All six available models in one article
With the first implementation of the Bulldozer microarchitecture, everything has been clear to everyone for a long time - including AMD :) Over the past year, we managed to improve the yield of suitable processors, which made it possible to completely cope with the shortage and even begin to carefully manipulate prices. But faster processors than the FX-8150 have not yet appeared. And it is obvious that it is impossible to achieve a significant increase in performance without architectural improvements. They are just planned for this year, but in stages. That is, the first processors with Piledriver cores have recently been released, but they have become Trinity notebook APUs, where progress is most important for AMD at the moment (especially given the competitor's improved integrated graphics in Ivy Bridge). Desktop Trinity will have to wait until the end of the summer (in any case, such forecasts now seem most likely), but they will not bring anything to the high-performance segment - a configuration with a couple of modules, despite all the improvements in the architecture, is only a competitor for dual-core Ivy Bridge in terms of processor parts. However, in the same third quarter, Zambezi is also expected to change to Vishera, which promises increased performance at the same clock speed and the same four modules in the maximum configuration. The names of the first processors are already approximately known: FX-8350, FX-6300 and FX-4320 - respectively, four, three and two new "improved" modules. How significant the Piledriver improvements will be - then we will evaluate.
What to do now? The price cut is an obvious step: they looked a little too high even in the days of Sandy Bridge, and the new technical process allowed Intel to “pull up” the performance a little more for the same money. But you also need to somehow release new processor models to increase interest in your platform, which, as we have already written above, is very difficult without architecture improvements. More precisely, it is very difficult in the older segment, where the frequencies are already so high that there is no reserve for the heat pack. But two- and three-module models fit into 95 watts. On the one hand, too many. On the other hand, it still doesn’t work out to compete with Intel in this parameter, so why not use the 30 W reserve to increase performance? So the FX-4170 and FX-6200 were born, which are accelerated versions of the FX-4100 and FX-6100. But there is nothing new in them - in fact, this is nothing more than guaranteed factory overclocking. Moreover, for an additional payment of about $ 20 compared to their predecessors, i.e. economical overclockers, in general, will not receive anything new - all multipliers are unlocked, so that both the core frequencies and UnCore can be tuned independently (and it can very well be more degree than with factory overclocking). On the other hand, these processors are quite worthy of attention. Even if only as objects of express testing - so that you can get a more complete picture of the current AMD FX line.
We originally wanted to do a full review of the entire FX range, which now consists of seven models. Moreover, in fact, it was divided into two separate parts - processors with a 95 W TDP (4100, 6100 and 8100) and faster, but more "gluttonous" models designed for 125 W (4170, 6200, 8120 and 8150). However... However, the stumbling block turned out to be the FX-8100 - at the moment the only "full" (ie, four-module) processor with a TDP of 95W. This unique combination of characteristics makes it highly desirable (because eight cores, eight gigs, gaming graphics card, and with a cooler for $ 10) from large computer manufacturers - to the detriment of retail chains. In general, we have not yet been able to “get” this processor. Even resorting to the help of the manufacturer, nothing happened. Perhaps, over time, the deficit will “dissolve” (especially after another improvement in production - it’s not for nothing that AMD even planned to reduce the heat pack of the FX-8120 to the same 95 W), but at the moment the idea of beneficial FX had to be abandoned. However, no, it's not scary - if we didn't manage to get the processor, it's unlikely that buyers will have an easier time. Unless it is acquiring, but it seems to us that there are relatively few of those among the regular readers of this line of articles (and even fewer who are eager to purchase the “eight-core” “bulldozer”).
Test stand configuration
| CPU | FX-4100 | FX-4170 | FX-6100 | FX-6200 | FX-8120 | FX-8150 |
| Kernel name | Zambezi | Zambezi | Zambezi | Zambezi | Zambezi | Zambezi |
| Production technology | 32 nm | 32 nm | 32 nm | 32 nm | 32 nm | 32 nm |
| Core frequency (std/max), GHz | 3,6/3,8 | 4,2/4,3 | 3,3/3,9 | 3,8/4,1 | 3,1/4,0 | 3,6/4,2 |
| Starting multiplication factor | 18 | 21 | 15,5 | 19 | 15,5 | 18 |
| Number of cores/threads of calculation | 4/4 | 4/4 | 6/6 | 6/6 | 8/8 | 8/8 |
| L1 cache, I/D, KB | 2×64/ 4×16 | 2×64/ 4×16 | 3×64/ 6×16 | 3×64/ 6×16 | 4×64/ 8×16 | 4×64/ 8×16 |
| L2 cache, KB | 2×2048 | 2×2048 | 3×2048 | 3×2048 | 4×2048 | 4×2048 |
| L3 cache, MiB | 8 | 8 | 8 | 8 | 8 | 8 |
| UnCore frequency, GHz | 2 | 2,2 | 2 | 2,2 | 2 | 2,2 |
| RAM | 2×DDR3- 1866 | 2×DDR3- 1866 | 2×DDR3- 1866 | 2×DDR3- 1866 | 2×DDR3- 1866 | 2×DDR3- 1866 |
| socket | AM3+ | AM3+ | AM3+ | AM3+ | AM3+ | AM3+ |
| TDP | 95 W | 125 W | 95 W | 125 W | 125 W | 125 W |
| Price | N/A() | N/A(0) | $111(as of 01/11/16) | N/A(0) | N/A(0) | N/A(0) |
The most interesting model is the FX-4170: in fact, this is the first desktop processor in which the base frequency exceeds 4 GHz, i.e. AMD kept its long-standing promise. Unfortunately, as has been known for a long time, the efficiency of the new architecture is not so high yet, which makes it difficult to enjoy high frequencies, but... In low-threaded software (of which there is still a huge amount), this processor should be no worse than the top FX-8150. Maybe even a little better. And the FX-6200 can also be a reasonable compromise, fitting exactly between the 6100 and 8120: in multi-threaded applications, the processor is able to compete with the latter due to a higher clock frequency, and in low-threaded applications, due to it, it can even turn out to be the fastest. The situation, of course, spoils the relatively high level of TDP - after all, 4100, 6100 and even 8100 "fit" into 95 W, however, in principle, it is not so scary in reality. And it may very well be that in practice it will be unattainable. Moreover, the buyer of low-module processors now also has a choice: faster and hotter or colder and cheaper - the main thing is to use this opportunity correctly. That is why we decided not to complicate the picture, but limit ourselves to only six processors: two main characters, two of their "energy efficient" brothers and two top models. In the end, there have already been many comparisons of the FX family with other architectures (and with very different ones), but it's interesting how the alignment inside the line looks like.
Testing
Traditionally, we divide all tests into a number of groups, and show the average result for a group of tests/applications on the diagrams (for details on the testing methodology, see a separate article). The results in the diagrams are given in points, for 100 points the performance of the reference test system, the site of the sample of 2011, is taken. It is based on the AMD Athlon II X4 620 processor, but the amount of memory (8 GB) and video card () are standard for all tests of the "main line" and can only be changed as part of special studies. Those who are interested in more detailed information are again traditionally invited to download a table in Microsoft Excel format, in which all the results are shown both in converted points and in "natural" form.
Interactive work in 3D packages
As expected, in this low-threaded group, all the subjects lined up in terms of clock speed, and it is the highest in the FX-4170. The flagship of the line follows a little behind it, and then all the rest, and the best of the "others" was the FX-6200. On the one hand, the victory of the "turbocharged" models, on the other hand, is not so serious. That is, for such loads, the slowest FX-4100 is the best choice: it is the cheapest and fits into 95 W, lagging behind the leader by less than 10%. Multi-module processors, obviously, are the worst suited for such tasks - even Turbo Core does not completely save the situation.
Final rendering of 3D scenes
But in this group they are out of competition for obvious reasons, although the frequency is important even when all the executive devices can be loaded with work. Taking it into account, even "almost" out of competition - the result of the FX-8120 is 144 points: only 1 (!) More than the FX-6200. But the latter is cheaper, and the heat pack of these devices is the same. So it's obvious that the 6200 would easily outperform the "economical" and hard-to-find 8100. All in all, not such a bad result. FX-4170, of course, is incapable of such feats, but for dual-module processors, rendering is not quite the target area of application from the very beginning.
Packing and unpacking
But here the FX-4170 managed to impress us, taking third place in the family - after the older FX-8150 and 8120 (and losing only 1 point to the latter). Which is not surprising - multi-core models here are “pulled out” only by data packing by 7-Zip, while the three remaining subtests are quite content with one or two computation threads.
Audio encoding
And even in this group, traditionally gravitating towards multi-threaded processors, our today's heroes look very good. Yes, of course, they do not set records, but their positioning (primarily in terms of price) is fully justified.
Compilation
A similar picture, only a little more "torn". However, we initially had no particular doubts about this state of affairs: these two groups are very susceptible to the number of integer computation threads, and this is the best thing for Bulldozer. In turn, the dependence of performance on the clock frequency is an axiomatic, regardless of the specific architecture (the main thing is to be the same), which allows “accelerated” low-core processors to occupy an intermediate position between normal small- and multi-core ones. However, the dependence of the generated thermal power on the clock frequency also does not disappear anywhere, so in a particular situation you need to choose what is more important: performance, price or heat dissipation. And here AMD is not original - the same trends are observed in the assortment of Intel processors.
Mathematical and engineering calculations
The two main characters are the same. And of all the FX, only the 8150 is behind, and the size of this gap can be put in a claim to the flagship, and not vice versa.
Raster graphics
Another victory of frequency over quantity, which, in fact, there were no particular doubts from the very beginning: there are multi-threaded tests in this group, but there are also single-threaded tests in general. Therefore, earlier in it, FX suffered precisely because of low single-threaded performance, and what could not be solved architecturally can be fixed by brute force.
Vector graphics
Especially here, where even two FX-41x0 modules are redundant. As a result, the FX-4170 won everyone, and the FX-6200 took an honorable third place.
Video encoding
Applications of this group traditionally gravitate towards multithreading (not all equally, of course, but this is a common occurrence), however, it is quite possible to compensate for the “lack” of cores with an increased clock frequency. So the performance of the FX-6200 can be considered the best - it lags only behind the top FX-8150, overtaking two other four-module models (we did not test the FX-8100, but there is no doubt that it is slower than the 8120, and there cannot be). And the FX-4170 was just a little short of repeating this feat - it is literally one and a half percent slower than the FX-6100. But slower. But by one and a half percent :) But TDP is higher. But cheaper. In general, each approach has its advantages and disadvantages, so the main thing is to properly dispose of the freedom of choice that has appeared.
Office software
The only thing that somehow saves four-module processors is the multi-threaded FineReader. And even then - this is only true for the FX-8150, and already the 8120 is behind both the 6200 and, moreover, the 4170.
Java
The JVM simply loves multi-core processors, however, it is also ready to trade cores for frequency. In any case, the increase in clock frequencies allows you to win back at least half of the gap from the "more nuclear" processors and make the "ladder" more uniform.
Games
If we compare only the 6000 and 8000 families, there is a feeling that games really need multithreading. However, in reality this is not the case - the failure of the FX-6100 and FX-6200 is due to only one reason: F1 does not tolerate any “six-threaded” processors - both Phenom II and FX (and the LGA2011 platform in this game seems to be saved only by Hyper -Threading, bringing the number of threads to 12). Moreover, this game is just the best accelerated at 8000, which allows these processors to generally break away from 4000 a little. faster than both of them. At the same time, it is also the cheapest :) In general, a good gaming computer based on AMD FX is possible - just don't run after older models.
Multitasking environment
This experimental test has recently demonstrated good stability and predictability, so we once again decided to use it to look at the subjects from this point of view. Moreover, in their case, the question of “many slow” or “little fast” has been very poorly disclosed until recently.
The picture turned out to be similar to the Java machine - with the same architecture, more cores are always better, but the performance of each matters. In general, the release of the FX-4170 and FX-6200 is quite justified from the point of view of such “refined multi-threaded” applications: they filled the gaps between 4100, 6100 and 8120 remarkably well.
Total
For many years, songs have been heard in various forums: A computer is not bought for one year, but over time, the degree of utilization of multithreading by software will only grow, so you need to buy a multi-core processor. And for a year now we have been observing that the mass of application software remains generally single-threaded. Yes, of course, over the past five years, the number of programs that can use multithreading well has grown, but they are still far from 100%, firstly, and "many" often means "two" or at best "four" in the first place. -second. And some applications that are interesting to the mass user, we repeat, generally remain single-threaded. Entirely and completely or in no small part. Well, since our testing methodology is based on real applications, and not at all on synthetics, its overall score reflects the situation on average quite well. Due to resource-intensive applications, of course, multi-core processors are ahead, but the increase can not always justify the difference in price. The cheaper FX-4170 came head to head with the more expensive FX-6100, while the FX-6200 was only slightly behind the FX-8120 (and the latter doesn't have a TDP advantage either).
And even such a general result is largely due to the presence of well-parallel applications, which, as a rule, do not differ in mass demand. What happens in "domestic" areas of application is clearly visible in some of the diagrams: there, multi-cores have no advantages at all. It is clear why - if only a part of potential computation threads are involved, everything "rests" on single-threaded performance. You can increase it either by architectural improvements that allow you to do more work per cycle, or ... The usual extensive method - increasing the clock frequency. And from this point of view, the release of the FX-4170 and FX-6200 is absolutely the right decision. Indeed, as has been said more than once, the Bulldozer architecture cannot yet boast of high efficiency, but it is designed for high clock frequencies from 4 GHz and higher. And at such frequencies, it unfolds in full force. Unfortunately, in multimodule processors, power consumption and heat dissipation increase too much, while younger models previously had a certain margin. Which AMD disposed of in the right way.
Should I be afraid of a high TDP level? As it seems to us, the most timid ones still haven’t looked in the direction of AM3 + before :) Moreover, in a modern house there are many places where savings are more justified. A discussion of our recent testing of household light bulbs showed that even in 7520 AD, many people continue to use incandescent infrared heaters to light their homes. And if a person is not interested in the opportunity to save a few hundred watts on a banal chandelier, then why does he need an extra 30 watts from the processor? With heat release, the situation is a little more complicated, but far from hopeless - after all, we are not talking about something unprecedented. Only 125 W, which is quite within the power of many coolers for AM3+ and previous AMD platforms (especially considering that some Phenom II "pulled" officially 140 W as well). Moreover, in terms of heat dissipation, a huge (by today's standards) Zambezi crystal is an advantage, not a disadvantage: it is much easier to “remove” at least 125, at least 150 W from 300 mm² than to cool an overclocked Ivy Bridge.
In short, the new processors are a step in the right direction. If it is not yet possible to increase the performance of older models, it makes sense to "pull up" the younger ones in order to strengthen its position in the $100-150 segment. Moreover, Intel will not offer anything new in it until the end of the summer. Of course, strictly speaking, there is nothing so “new” in these two new FX - all multipliers in the entire family are unlocked, so an economical overclocker can also get an analogue of the FX-4170 by buying a cheaper FX-4100. And earlier it could too. On the other hand, such a “factory overclocking” is at least interesting as a guideline. And for the vast majority of users who do not overclock anything on their own, an additional degree of freedom in choosing a processor will also not be superfluous.
AMD FX-8350 | Meet the AMD FX Processor Lineup Powered by Piledriver
We, as observers of computer hardware, are not very interested in the problems that hardware manufacturers face. Many would agree that AMD's CPU division had a terrible year last year, starting with the voracious Bulldozer CPUs that slowly dropped in price over twelve months in response to third-generation Intel Core processors. Recently, new samples have arrived at our laboratory in Southern California. Talk of mismanagement, layoffs, and AMD's bad past shouldn't be too much of a concern for end-users. So let's get down to business.
Sometimes it is possible to predict the ending of an article. If AMD sent us an FX-8170 running at 200 MHz faster than its predecessor, one would assume that the processor will show the same shortcomings in low-threaded applications, probably outperform the Core i5-2500K in more intensive tasks, but compared to 77W chips its power consumption is simply terrible.
But instead we got a processor AMD FX-8350, which, by analogy with APU Trinity, introduced less than a month ago, is based on the Piledriver architecture. Experience tells us that in terms of performance per core and per clock, the Piledriver can outperform the Bulldozer design by 15%. It is also worth considering that AMD FX-8350 will operate at least 400 MHz faster than FX-8150. By the way, the architecture provided a noticeably smaller increase in speed for the Intel Core processor line. There is a good chance that today's comparison will turn out to be much more interesting than a complete teardown. FX-8150 in last year's review.
Meet the FX Family on Piledriver Architecture
True to its tradition since the Bulldozer, AMD has sent out the fastest models in the new lineup to the press, which will include eight-, six- and quad-core models. While they are all based on the Piledriver architecture, the chips themselves are called Vishera and will continue to be sold under the FX brand.
AMD Vishera Crystal
The Vishera processor occupies 315 square millimeters and consists of 1.2 billion transistors. Exactly the same figures characterize the previous generation Zambezi based on the AMD Bulldozer architecture.
| 2012 AMD FX processor line | |||||||
| Number of cores/threads | Base frequency, GHz | Max, Turbo frequency, GHz | North bridge frequency, MHz | TDP, W | Price, $ | OPN | |
| FX-8350 | 8/8 | 4 | 4,2 | 2200 | 125 | 195 | FD8350FRW8KHK |
| FX-8320 | 8/8 | 3,5 | 4 | 2200 | 125 | 169 | FD8320FRW8KHK |
| FX-6300 | 6/6 | 3,5 | 4,1 | 2000 | 95 | 132 | FD6300WMW6KHK |
| FX-4300 | 4/4 | 3,8 | 4 | 2000 | 95 | 122 | FD4300WMW4MHK |
Two of the four processors use eight processing cores, or four Piledriver modules, as AMD calls them. The base frequency of the flagship model AMD FX-8350 is 4 GHz. In low-threaded applications, Turbo Core technology can increase this figure to 4.2 GHz, although to a greater extent the increase in chip speed will be associated with the base frequency. How much does Turbo Core speed up AMD FX-8350? Not much. In the single-threaded iTunes benchmark, the result improved by only three seconds.
In the FX-8320, the base frequency is lowered to 3.5 GHz, but within the allotted thermal package, Turbo Core increases it to 4 GHz (500 MHz acceleration is more important for the FX-8320). Both eight-core models are equipped with 8 MB of L2 cache (divided into 2 MB for each module) and 8 MB of L3 cache (common for all four modules of the chip). Suggested price for AMD FX-8350 is $195 and the FX-8320 is $169.
The FX-6300 already has three active modules (six cores), and the price has been reduced to $132. The 3.5 GHz frequency gives the architecture an advantage in multi-threaded tasks, while Turbo Core tries to compensate for the disadvantages in single-threaded applications by increasing the frequency to 4.1 GHz. Like four-module chips, the FX-6300 uses 2 MB of L2 cache per module (6 MB in total) and a shared 8 MB L3 cache. Fewer active resources, as well as a lower northbridge frequency (2 GHz), allow the FX-6300 to stay within the 95 W TDP, which is noticeably different from the 125 W of the older FX-83x0 processors.
The TDP of the dual-module processor FX-4300 also does not exceed 95 watts. The base frequency of 3.8 GHz in low-threaded applications is increased to 4 GHz, and the northbridge operates at a frequency of 2 GHz like the FX-6300. However, the L3 cache capacity is reduced to 4 MB, and the price is only $10 lower than the three-module chip, which will encourage many buyers to spend an extra $10.
The AMD architecture does not feel much shortage of memory bandwidth, however, the dual-channel DDR3 controller officially supports data transfer rates of 1866 MT/s. To reduce the overall cost of the platform, we will use low-latency DDR3-1600 modules, especially since the test results (except for Sandra 2013 Beta) show that spending more money on faster memory will not increase speed.
The entire FX line has an unlocked multiplier, which makes overclocking much easier. Does Vishera have enough overclocking potential? What about 5.125 GHz with closed loop liquid cooling?
AMD FX-8350 | Overclocking and platform compatibility
Overclocking
Although AMD processors have not been named the fastest for several years, the company is trying to keep the attention of enthusiasts by providing features that are important to them. Software that allows you to change configuration settings in Windows in real time, unlocked multipliers, and platforms with more PCI Express slots are just some of the key features that AMD offers to people who know how to use them to increase the performance of their systems.
Those overclockers who were disappointed in the scalability of FX Zambezi processors with a conventional cooling system will be very surprised this time, even though we are considering a very similar architecture on the same 32nm core.
With CPU and Northbridge voltages of 1.375 V and 1.175 V, respectively, we were able to achieve stable operation AMD FX-8350 at 4.8 GHz at full load. In the screenshot above, a single-threaded test is running to "spin up" the chip, but the highlighted maximum temperature corresponds to the peak load of the entire test package.
AMD FX-8350 could go even faster, but the key factor here is to set the voltage so that the temperature does not exceed 70 degrees Celsius. At this point, the temperature sensor causes the cores to lower their frequency (the image above proves this), preventing the chip from overheating and negatively affecting performance. Without giving multi-threaded loads, we even managed to run tests at 5.125 GHz (for this you need to set the voltage to a 1.4375 V for the CPU and 1.2 V for the north bridge).
Obviously, in most user systems, cooling will be the bottleneck. AMD's reference heatsink and fan are not enough, to put it mildly, and a powerful third-party solution will increase the cost of a platform with an FX processor. But exclusively for testing, we used the closed-loop liquid cooling system that AMD offered with the FX processors last year. It will cost you about $70. In such a case, you can consider $300 as an alternative. Luckily, we have test results.
Overclocking to 4.8 GHz is enough to AMD FX-8350 outperformed in a multi-threaded environment like 3ds Max 2012, but it didn't help the AMD Piledriver architecture outperform in a single-threaded iTunes application. Of course, if you're willing to spend an extra $30 on and even more on a third-party cooler, then its base frequency can be increased relatively easily from 3.5 GHz to 4.5 GHz and face AMD FX overclocking.
Compatibility
All four FX Piledriver chips are compatible with the existing Socket AM3+ processor interface. In order for older motherboards to recognize the new FX series processors, you need to update the BIOS. However, boards that previously experienced problems with FX processors are unlikely to get rid of this shortcoming.
Asus added support for the FX line in 2011 as an experiment. However, the company has not yet released the updates needed to fix the blue screen of death in some situations. Therefore, we do not believe that the old AM3 platforms will provide adequate support for FX processors, and with the arrival of Piledriver, the problems will not go anywhere. AMD assured that the problem is not global, and suppliers can solve it with updates. But some board makers don't seem to care much about upgrading older products.
AMD FX-8350 | Piledriver architecture: what has changed compared to Bulldozer
The current AMD x86 architecture was covered in great detail in the review FX-8150 (AMD FX-8150 Review: From Bulldozer to Zambezi and FX). All these principles can be transferred to the Piledriver architecture. However, AMD engineers gained experience when they put the Bulldozer concept into practice. We know that the manufacturing process has evolved over the year, even though the company is still using the 32nm core in its Vishera processors. Therefore, one should not be surprised if the new architecture turns out to be just a modified old one, where the changes made only turned out to be long-planned "work on the bugs".
Entry Improvements
After the premiere of AMD Bulldozer, it was argued that branch prediction is one of the weaknesses of the architecture. The Piledriver module concept includes some shared resources across two execution pipelines, and the architects tried to minimize bottlenecks in the preprocessor by implementing one branch prediction queue per thread. The company claims that Piledriver has improved prediction accuracy.
Piledriver adds support for several ISA extensions that we first reviewed in the review APU Trinity. The unified multiplication block was introduced a year ago in Bulldozer. That version was called FMA4 and allowed instructions to have up to four operands. But in the upcoming Haswell architecture, Intel only plans to use the simpler three-operand FMA3 instruction set, so AMD retains that advantage in Piledriver. Another extension is called F16C. It includes support for converting up to four half precision values to floating point values at a time. The Intel architecture already has such an extension, so Piledriver is simply catching up with the competitor. This is not to say that Bulldozer felt an urgent need for FMA3 / F16C, because compiler-level support was added only in Visual Studio 2012.
integer block
Each of the two integer cores uses a separate load/store unit capable of two 128-bit loads per clock or one 128-bit store per clock. AMD found that in some cases Bulldozer was unable to locate stored data in a register file that was already there. After the fix, instructions get into an integer block faster.
The integer core still has two execution units and two address generation units (simply called AGen). This time, AGen's abilities have been expanded, and they can process MOV instructions. When the activity of the AGen block is low, the architecture will redirect MOV instructions on these channels.
One important change is the increase in the fast lookup buffer (TLB) for the L1 cache from 32 entries to 64 entries. Because L2 TLB has a fairly high 20 clock latency, increasing the L1 cache hit rate can significantly improve performance in data-intensive applications. This is especially true for server environments, however, according to AMD engineers, some games are also sensitive to this, although they did not expect this.
L2 cache optimizations
Hardware sampling in L2 has also been improved. The minimum latency hasn't changed, which is why Sandra 2013's cache latency hasn't improved. However, the prefetcher and L2 cache are used more efficiently, and AMD says average latency (very hard to measure) should be reduced. The same Sandra 2013 module shows minimal changes in L3 latency, and the Vishera architects confirm that no changes were made to the common L3 cache for all FX processor modules.
Putting It All Together: Five Architectures at 4GHz
What effect do all these changes have on Piledriver performance? To compare relative results, we will test five different architectures at 4 GHz.
However, this is a very generalized result. The subtests show how each platform affects the overall gaming performance score.
It looks like the only processor out of the picture is the AMD Phenom II X6 1100T, and only by a couple of percent. All others show the same results because the Graphics subtest isolates GeForce GTX 680 .
In the Physics subtest, processor performance plays a major role, as Futuremark divides the world into several separate regions, creating many threads.
AMD FX-8350 faster than AMD Phenom II X6 1100T. But a six-core chip has lower power consumption, and if its speed is not much lower, the efficiency may be higher. This will be a disaster for AMD.
Unfortunately, the platform based on the Vishera chip is unlikely to be able to catch up with the Intel processors in terms of efficiency, since the average power consumption results differ dramatically.
Here AMD can be proud of itself. AMD FX-8350 finishes second among the processors we have chosen (CPUs were selected in such a way that there was not a big spread in the results). It was very interesting how AMD FX-8350 will resist Core i5-3570K, and as a result, the new processor outperformed the more expensive model by 12 seconds.
The fact that AMD is asking for a new flagship that ran nearly 10 minutes faster than FX-8150, less than $200, only adds value to it. But what about efficiency?
AMD FX-8350 almost 13% more efficient than its predecessor. But more importantly, the new processor is more efficient than the Phenom II X6 1100T.
A year ago, Phenom surpassed Bulldozer in this regard. And it was undeniable that AMD introduced a processor that consumed more power and ran slower. Today, we recognize higher performance, improved efficiency and more attractive pricing. Is this enough to get a recommendation? Core i5-3570K, even spitting on serious differences in efficiency and energy consumption. This is our opinion in 2012.
As already mentioned, AMD FX-8350 was much stronger FX-8150 and allows AMD to regain ground lost by the Bulldozer architecture. The Piledriver doesn't fix all of the Bulldozer's shortcomings, but tweaking the design and power consumption allows the company to increase the clock speed without affecting the 125W TDP. The changes are not so significant, but they are enough to create a good alternative to the best Intel Core i5 models.
Of course, if AMD decided to ask for a new chip for $245 as FX-8150, processor AMD FX-8350 I would expect the same "success" as last year's model. However, the recommended price does not exceed $200. Thereby AMD FX-8350 it is on par with the Intel Core i5-3470, with a locked multiplier, which was behind in many benchmarks. In the same tests, the new FX chip outperforms Core i5-3570K for $230. And only in single-threaded applications, Intel processors remain out of reach.
But there is still the issue of power consumption. In Russia, thank God, electricity is relatively cheap. And it's unlikely anyone will bother with an extra 50W unless you need a loud cooler to dissipate it. But, for example, people in Denmark pay $0.40/kWh, and the difference is only 10 W between Core i5 and AMD FX-8350 in idle time it will cost a few dollars a month. A system running 24/7 under load will already ask for an extra $15 per month. Advantage for Intel.
Anyway, let's try to make a recommendation. Considering that professional users prefer AMD more than Intel, we believe that many enthusiasts will be seriously interested AMD FX-8350, as opposed to Bulldozer chips, and rightly so. Everyone wanted to see more speed, improved efficiency, and a lower price, and AMD delivers it all. Are there any compromises? Yes, I have. Performance in single-threaded applications did not impress us at all, and power consumption is still a sore point. But AMD FX-8350 for less than $200, it's definitely usable in a mid-range workstation.
We would choose AMD FX-8350 for the new system? Most likely no. While the AMD architects have done a great job over the past year, performance is still highly dependent on the type of task. And given that electricity is constantly becoming more expensive, and productivity is at a close level, we would choose a more efficient option.
IntroductionOver the past few years, AMD's position in the desktop processor market has steadily deteriorated. At first, due to the lack of new progressive microarchitectures, the company had to drop the prices of its processors over and over again, and as a result, we came to the conclusion that AMD completely left the high-end performance CPU segment. Then there was an epic failure at all - the release of processors with the Bulldozer microarchitecture, on which a huge amount of hope was initially placed. The Bulldozer was expected to be a product capable of competing with the older LGA 1155 and LGA 2011 Intel processors, but in reality the new microarchitecture turned out to be slow and power hungry. As a result, the Bulldozer has become a niche product, capable of at least some resistance to mid-range Intel processors only thanks to twice the number of cores. And even then, one could talk about comparable performance only with a multi-threaded load in mind, and, moreover, turning a blind eye to the gigantic level of power consumption by modern standards. In other words, the release of Bulldozer failed to strengthen AMD's position in the desktop processor market.
Fortunately, a series of market failures, generously seasoned with management troubles, difficult financial situations, staff cuts and the search for a new strategy, did not become an obstacle to the work of engineers, and a year after the Bulldozer announcement, we saw a second, improved version of this microarchitecture - Piledriver. Testing of the senior member of Vishera's new line of desktop processors, the FX-8350, showed that the past year was well spent. The FX-8350 made it possible to significantly increase the performance of AMD's flagship platform. Benchmark scores rose by an average of 15 percent, more than the gain that Intel's processors made from the Sandy Bridge to Ivy Bridge microarchitecture. Wishing to enhance the effect of its successful microarchitecture update, AMD has chosen a very democratic pricing policy, and now the Socket AM3+ platform looks much more attractive than before. Vishera processors were talked about in a positive way, and we have no doubts that the new product in the face of AMD FX-8350 will be able to attract a noticeable number of supporters to its side.
At the same time, if you dig deeper, it turns out that processors with Piledriver microarchitecture are not so progressive. In fact, all the improvements made are purely cosmetic, and the superiority of the FX-8350 over the FX-8150 is explained by a whole combination of factors, among which a significant role is played not so much by improvements in the microarchitecture, but by a clock frequency increased by 400 MHz, as well as more aggressive operation of the technology. turbocharging. For end users, the sources of increased performance are not so important, but on the other hand, it turns out that the superiority of the older Vishera model over the older Zambezi model cannot be automatically transferred to other representatives of the respective families. With the FX-8350, AMD tried to crank up the clock speed to the maximum to impress us, but simpler and cheaper modifications of the FX processors with a new microarchitecture may not provide the same noticeable advantage. Especially if you take into account the fact that among the FX of the "new wave" there are also options with a reduced third-level cache.
That is why we decided to conduct a separate test of the younger Vishera models that were not included in our first review. In this article, we will analyze whether all processors of the FX series with the Piledriver microarchitecture can be considered a successful update of the model range, or only the FX-8350 deserves a positive review.
FX range with Piledriver microarchitecture
For reviews of new Socket AM3+ processors with the Piledriver microarchitecture, AMD sent out to the press the older version of Vishera, FX-8350. However, the revamped FX line actually includes four models: the said FX-8350 and slower, lower clocked variants with eight, six, and four cores. To some extent, they are all similar: according to a long tradition, AMD remains a supporter of unification and uses a semiconductor crystal of a single design in its Vishera line. It provides four dual-core modules and an 8-megabyte L3 cache. Such a Vishera semiconductor crystal consists of 1.2 billion transistors and has an area of 315 square meters. mm.However, in the lower models of processors, this crystal is not fully used. The manufacturer can disable one or two dual-core modules, or cut the amount of L3 cache. On the one hand, this forms a line of different-sized offers, and on the other hand, the manufacturer acquires the opportunity to sell partially defective semiconductor devices. In its younger generation Zambezi processors, AMD varied the number of active dual-core modules - this is how the six-core FX-6000 and the quad-core FX-4000 were obtained. In Vishera, differentiation has become deeper - the possibility of halving the working cache memory of the third level has also gone into action. As a result, the AMD FX line based on the new Piledriver microarchitecture, which so far consists of only four models, looks like this.
Let's get acquainted with its representatives in a little more detail.
AMD FX-8350
The flagship processor of the Vishera family has already been reviewed by us in separate review. Nevertheless, we remind you that it embodies the maximum modification of the processor in Socket AM3+ version at the moment. This means that the FX-8350 has eight dual cores and an 8 MB L3 cache, and its nominal frequency is set to 4.0 GHz, which allows the manufacturer to call this CPU the first desktop processor to conquer the 4 GHz mark.
AMD sees the FX-8350 as a competitor to Intel's Core i5-3570K, but this estimate is too optimistic in our opinion. However, the official cost of the CPU, set at $195, makes it possible to compare it with slower Core i5 variants with Ivy Bridge microarchitecture, against which the new product looks more confident.
Unfortunately, there is no information yet about how fast AMD is going to increase the frequencies of its Vishera line. It looks like the FX-8350 will remain the flagship for quite some time. Moreover, according to the latest data, the release of the next iteration of the microarchitecture with the code name Steamroller is postponed to 2014.
AMD FX-8320
Obviously, at a frequency of 4.0 GHz with a heat dissipation not exceeding 125 W, not all Vishera semiconductor crystals can work, the production of which uses far from the most modern 32-nm process technology. Therefore, the top eight-core FX-8350 in AMD's lineup is supplemented by a slower FX-8320 modification with similar characteristics, but at a lower clock speed. That is, like its older brother, the FX-8320 has eight twin cores and an 8-MB cache in the third level, but at the same time its nominal frequency is only 3.5 GHz - 500 MHz less than the flagship. This processor accelerates to the coveted 4 GHz only when Turbo Core technology is activated. However, the TDP of the slower eight-core version of the Vishera remains at 125W.
Overall, the frequency response of the FX-8320 is similar to that of the Zambezi-designed FX-8150. However, at the same time, its cost is set at $169, and the Core i5-3450 is indicated by marketers as a direct competitor in the Intel line.
AMD FX-6300
Vishera's six-core model, the FX-6300, is obtained from a full-fledged semiconductor die by disabling one of the four dual-core processor modules. In terms of its own frequencies, it is close to the FX-8320. Nominally, the six-core operates at 3.5 GHz, and with a decrease in the number of executable threads, it can accelerate to 4.1 GHz. At the same time, the reduced number of processing cores and moderate clock speeds allowed the manufacturer to install a tougher 95-watt thermal package for the FX-6300. However, in the end, the frequencies of the six-core Vishera are inferior to the frequencies of the FX-6200 processor, which is based on the microarchitecture of the previous generation Bulldozer.
Disabling two cores on the original chip did not affect the size of the L3 cache, it, like the full-fledged Vishera, has a volume of 8 MB. However, the L2 cache, which in the Piledriver microarchitecture (as well as in the Bulldozer) is individual for each pair of cores, is smaller in the FX-6300. It quite expectedly consists of three 2-megabyte parts, that is, its total volume is 6 MB versus 8 MB for eight-core.
Another difference between the six-core Vishera modification is the reduced frequency of the north bridge built into the processor. For the FX-8350 and FX-8320, this node operates at a frequency of 2.2 GHz, while for the FX-6300, the north bridge frequency is 200 MHz lower. This translates into some decrease in the speed of the processor memory controller and L3 cache. However, practice shows that this difference is not critical.
But the FX-6300 is much cheaper compared to the eight-core ones. The manufacturer estimates it at $132, and the outdated Core i5-2300 is indicated as a direct competitor for this processor.
AMD FX-4300
The FX-4300 is perhaps the strangest member of the new wave FX series. This processor costs $122, which is only $10 cheaper than the FX-6300, but at the same time it is significantly weaker in performance. Firstly, the number of processing cores in it has been reduced to four - two of the four dual-core modules in this CPU are blocked. Secondly, the L3 cache in the FX-4300 has also been reduced: its volume in this case is 4, not 8 MB. As a result, we get a “half” of the FX-8350, only the price of this half is less than that of the flagship, not half at all.
The frequencies are not impressive either: nominally, the FX-4300 operates at 3.8 GHz, and thanks to the turbo mode it accelerates to 4.0 GHz. The northbridge of this processor, like that of the FX-6300, operates at a frequency of 2.0 GHz. All this allows the quad-core Vishera to remain within the 95-watt thermal package, but at the same time it noticeably loses in terms of characteristics to the older quad-core of the previous generation, the FX-4170, which has a frequency of 200-300 MHz higher, and the L3 cache is full-size. As a result, the superiority of the quad-core Vishera over the quad-core Zambezi raises certain doubts.
However, AMD itself is aware of the weakness of the FX-4300 characteristics, pointing out the dual core of the Sandy Bridge generation, Core i3-2120, as a direct competitor for this CPU. Moreover, based on current prices, a quad-core A10-5800K processor for the Socket FM2 platform, based on the same Piledriver microarchitecture, seems to be a more profitable purchase than the FX-4300. Its price is exactly the same, but at the same time it has an integrated graphics core and has slightly higher clock speeds.
How We Tested
Since this test, our methodology has undergone significant changes. Namely, we switched to using the latest Microsoft Windows 8 operating system. Of course, such a transition does not fundamentally change anything, after all, the software environment does not affect the computing performance of platforms so much. But, nevertheless, it should be borne in mind that the Windows 8 kernel contains a number of innovations.First, the scheduler of the new OS, unlike previous systems, is initially optimized to work with all modern processor microarchitectures using SMT and CMT technologies. This means that neither for modern AMD processors with dual cores, nor for Intel processors supporting Hyper-Threading technology, no patches are required: everything works in the most optimal way right out of the box. Secondly, the core of the new operating system has undergone significant changes aimed at improving its functioning on various kinds of mobile computers. And although we are focusing on desktop tests, these changes indirectly affect our case as well. Windows 8 uses RAM more economically, parks idle cores more aggressively, and tries to reduce the cost of processor cycles. Thirdly, in Windows 8, the Aero interface was replaced by a fundamentally new Metro interface, which has increased 2D performance. And fourthly, the new system contains an update to DirectX version 11.1 (Direct3D 11.1, DXGI 1.2, WDDM 1.2, etc.).
As for the direct test participants, we compared the Vishera generation AMD FX-8350, FX-8320, FX-6300 and FX-4300 processors both with their predecessors with the Zambezi design and with modern Intel offerings based on the Ivy Bridge design. . The honor of the old Bulldozer microarchitecture was defended by the older processor models in each weight category: the eight-core FX-8150, the six-core FX-6200 and the quad-core FX-4170. Intel processors were represented by the older LGA 1155 quad-core with support for Hyper-Threading technology, Core i7-3770K; simpler quad-core Core i5-3570K and Core i5-3470; dual-core processor with Hyper-Threading, Core i3-3240; and the youngest carrier of the Ivy Bridge microarchitecture at the moment - the Pentium G2120 processor.
As a result, the composition of the test systems included the following software and hardware components:
Processors:
AMD FX-8350 (Vishera, 8 cores, 4.0-4.2 GHz, 4 x 2 MB L2, 8 MB L3);
AMD FX-8320 (Vishera, 8 cores, 3.5-4.0 GHz, 4 x 2 MB L2, 8 MB L3);
AMD FX-6300 (Vishera, 6 cores, 3.5-4.1 GHz, 3 x 2 MB L2, 8 MB L3);
AMD FX-4300 (Vishera, 4 cores, 3.8-4.0 GHz, 2 x 2 MB L2, 4 MB L3);
AMD FX-8150 (Zambezi, 8 cores, 3.6-4.2 GHz, 4 x 2 MB L2, 8 MB L3);
AMD FX-6200 (Zambezi, 6 cores, 3.8-4.1 GHz, 3 x 2 MB L2, 8 MB L3);
AMD FX-4170 (Zambezi, 4 cores, 4.2-4.3 GHz, 2 x 2 MB L2, 8 MB L3);
Intel Core i7-3770K (Ivy Bridge, 4 cores + HT, 3.5-3.9 GHz, 4 x 256 KB L2, 8 MB L3);
Intel Core i5-3570K (Ivy Bridge, 4 cores, 3.4-3.8 GHz, 4 x 256 KB L2, 6 MB L3).
Intel Core i5-3470 (Ivy Bridge, 4 cores, 3.2-3.6 GHz, 4 x 256 KB L2, 6 MB L3);
Intel Core i3-3240 (Ivy Bridge, 2 cores + HT, 3.4 GHz, 2 x 256 KB L2, 3 MB L3);
Intel Pentium G2120 (Ivy Bridge, 2 cores, 3.1 GHz, 2 x 256 KB L2, 3 MB L3).
CPU cooler: NZXT Havik 140;
Motherboards:
ASUS Crosshair V Formula (Socket AM3+, AMD 990FX + SB950);
ASUS P8Z77-V Deluxe (LGA1155, Intel Z77 Express).
Memory: 2 x 4 GB, DDR3-1866 SDRAM, 9-11-9-27 (Kingston KHX1866C9D3K2/8GX).
Graphics Card: NVIDIA GeForce GTX 680 (2 GB/256-bit GDDR5, 1006/6008 MHz).
Disk subsystem: Intel SSD 520 240 GB (SSDSC2CW240A3K5).
Power supply: Corsair AX1200i (80 Plus Platinum, 1200 W).
Operating system: Microsoft Windows 8 Enterprise x64.
Drivers:
AMD Chipset Driver 12.10;
Intel Chipset Driver 9.3.0.1025;
Intel Management Engine Driver 8.1.2.1318;
Intel Rapid Storage Technology 11.6.0.1030;
NVIDIA GeForce 306.97 Driver.
Performance
Overall PerformanceTo assess the performance of processors in common tasks, we traditionally use the Bapco SYSmark 2012 test, which simulates the user's work in common modern office programs and applications for creating and processing digital content. The idea of the test is very simple: it produces a single metric that characterizes the weighted average speed of the computer. With the release of Windows 8, the SYSmark 2012 benchmark has been updated to version 1.5, and we now use this adapted version.
Vishera processors show good progress in performance compared to their predecessors. The FX-8350 outperforms the FX-8150 by 19 percent, largely due to its increased clock speed. However, the role of the new microarchitecture should not be downplayed either. Even the FX-8320, which operates at a slightly lower frequency than the older Zambezi, is not inferior to it in terms of performance: the FX-8320 outperforms the FX-8150 by 10 percent. Approximately the same difference in results is observed in AMD six-core processors of different generations: FX-6300 and FX-6200. But the quad-core FX-4300 outperforms the FX-4170 by only 6 percent, since the old quad-core processor with the Bulldozer microarchitecture operates at a fairly high clock frequency and, unlike its successor, has a full-size 8 MB L3 cache.
However, the successes of AMD processors, which are clearly visible when comparing representatives of different generations, do not change the place of FX processors against the backdrop of competing Intel offers. Octa-core FX continue to be outperformed by quad-core Ivy Bridge, and AMD's hexa-core and quad-core processors are only comparable in performance to the Core i3 or Pentium.
A deeper understanding of the SYSmark 2012 results can provide insight into the performance scores obtained in various system usage scenarios. The Office Productivity scenario models typical office work: word preparation, spreadsheet processing, e-mail, and Internet browsing. The script uses the following set of applications: ABBYY FineReader Pro 10.0, Adobe Acrobat Pro 9, Adobe Flash Player 10.1, Microsoft Excel 2010, Microsoft Internet Explorer 9, Microsoft Outlook 2010, Microsoft PowerPoint 2010, Microsoft Word 2010 and WinZip Pro 14.5.
The Media Creation scenario simulates the creation of a commercial using pre-captured digital images and video. For this purpose, popular Adobe packages are used: Photoshop CS5 Extended, Premiere Pro CS5 and After Effects CS5.
Web Development is a scenario that simulates the creation of a web site. Applications used: Adobe Photoshop CS5 Extended, Adobe Premiere Pro CS5, Adobe Dreamweaver CS5, Mozilla Firefox 3.6.8 and Microsoft Internet Explorer 9.
The Data/Financial Analysis scenario is dedicated to the statistical analysis and forecasting of market trends that are performed in Microsoft Excel 2010.
The 3D Modeling scenario is all about creating 3D objects and rendering static and dynamic scenes using Adobe Photoshop CS5 Extended, Autodesk 3ds Max 2011, Autodesk AutoCAD 2011 and Google SketchUp Pro 8.
The last scenario, System Management, performs backups and installs software and updates. Several different versions of Mozilla Firefox Installer and WinZip Pro 14.5 are involved here.
Despite the fact that according to the final index, the performance of Vishera processors seems not high enough, there are common special cases when the situation for some of them is completely different. Modern AMD microarchitectures are weak under low-threaded loads, as they do not provide sufficient specific performance per core. But in those scenarios where the load is multi-threaded, they are ready to show very good performance, as they offer a larger number of cores than Intel competitors.
Vivid examples of such cases where the FX series looks worthy are 3D modeling tasks or financial and settlement tasks. In them, the eight-core FX-8350 and FX-8320 perform at the level or even better than the quad-core Core i5. However, options with a reduced number of cores can no longer boast of equally confident performance. Even in such favorable cases for AMD processors, the six-core and quad-core Vishera can only be compared with the dual-core Core i3.
Gaming Performance
As you know, the performance of platforms equipped with high-performance processors in the vast majority of modern games is determined by the power of the graphics subsystem. That is why, when testing processors, we choose the most processor-intensive games, and measure the number of frames twice. The first pass tests are carried out without turning on anti-aliasing and setting far from the highest resolutions. Such settings allow you to evaluate how well processors perform with a gaming load in general, which means they allow you to speculate about how the tested computing platforms will behave in the future, when faster variants of graphics accelerators appear on the market. The second pass is performed with realistic settings - when choosing FullHD-resolution and the maximum level of full-screen anti-aliasing. In our opinion, such results are not so interesting, but they answer the frequently asked question about what level of gaming performance processors can provide right now - in modern conditions.
The Bulldozer microarchitecture in games showed all its worst sides. Fortunately, its recent update, Piledriver, has charted a way out of this stalemate. The speed of Vishera in games compared to Zambezi has increased very noticeably. As a result, the FX-8150 is defeated not only by the next-generation octa-core FX-8350 and FX-8320, but also by the six-core FX-6300. But the gaming performance of the FX-4300 is somewhat frustrating. AMD completely unnecessarily cut the L3 cache in it, and as a result, with a game load that is sensitive to the speed of the memory subsystem, the new generation quad-core FX often loses to its predecessor, the FX-4170 processor.
However, noting the improvements in the gaming capabilities of Vishera processors with eight and six cores, we should not forget that Intel CPUs continue to demonstrate significantly higher speed in gaming applications. Platforms based on any Core i7 and Core i5 produce more frames per second than systems with older AMD FX processors, and Core i3 class processors can compete on equal terms with the FX-6300.
This means that adherents of AMD solutions, who while away their time playing 3D games, can only appeal to the fact that the speed in gaming applications is limited from above by the graphics subsystem, which does not allow current processors to "turn around" in full force. Therefore, in real conditions, the difference between fast and slow CPUs can be almost imperceptible. However, this is actually a rather weak argument. As can be seen from the tests, there are games in which processor power affects the number of fps even at maximum quality settings. In addition, in the near future we are expecting the release of a number of new 3D shooters, in which nothing is known yet about the influence of the CPU on graphics performance.
Application Tests
To measure the speed of processors during information compression, we use the WinRAR archiver, with the help of which we archive a folder with various files with a total volume of 1.1 GB with the maximum compression ratio.
WinRAR version 4.2 received high-quality optimization for multithreading, so the speed of AMD FX processors in it is quite good. It is thanks to this that the eight-core FX generations of Vishera overtake the Core i5, and the six-core FX-6300 “props up” their results from below. However, the speed improvement we see with the Piledriver microarchitecture media does not extend to the FX-4300. AMD has deprived this CPU not only of half of the cores, but also of half of the L3 cache, as a result of which it loses both the FX-4170 and the Core i3-3240.
The performance of processors under cryptographic load is measured by the built-in test of the popular TrueCrypt utility, which uses AES-Twofish-Serpent "triple" encryption. It should be noted that this program is not only capable of efficiently loading any number of cores, but also supports a specialized AES instruction set.
Cryptographic payload is a great option to bring out the strengths of AMD microarchitectures. Here, the FX-8350 is even faster than the older LGA 1155 processor, the Core i7-3770K, while the slower FX-8320 is only a tiny bit behind. The successes of older versions of Vishera extend to the six-core and quad-core processors with the Piledriver microarchitecture. The FX-6300 manages to outperform the Core i5-3570K, while the FX-4300 is well ahead of the competing Core i3 processor. At the same time, the contribution to such success of the new architecture is not so noticeable. The FX-8320 outperforms the FX-8150 by just 2 percent, the FX-6300 outperforms the FX-6200 by 5 percent, and the FX-4300 is completely behind the FX-4170. In other words, the 15% advantage of the flagship Vishera model over the corresponding Zambezi, which we admired when we first got acquainted with the new FX line, is primarily the result of clock speeds turned to the maximum.
When testing the speed of transcoding audio, the Apple iTunes utility is used, with the help of which the contents of a CD are converted to AAC format. Note that a characteristic feature of this program is the generation of an exclusively single-threaded load.
Unfortunately, under a single-threaded load, modern Socket AM3+ processors cannot offer decent performance. Due to the weakness of individual Bulldozer cores (and now Piledriver), even the flagship FX-8350 shamefully lags behind the Pentium G2120. Vishera-designed processors are a little faster than their predecessors, but this hasn't improved things at all.
With the release of the eighth version of the popular Wolfram Mathematica scientific computing package, we decided to return it to the number of tests used. To assess the performance of systems, it uses the MathematicaMark8 benchmark built into this system.
Wolfram Mathematica 8 is another example of an application where AMD processors are in trouble. In this case, the load is not single-threaded, but the failure of the FX series is again associated with the peculiarities of its microarchitecture. Modern AMD processors have only one FPU for each of their dual-core modules, and this ultimately leads to low floating point performance.
We measure performance in Adobe Photoshop CS6 using our own test, which is a creatively redesigned Retouch Artists Photoshop Speed Test that includes typical processing of four 24-megapixel digital camera images.
AMD FX processors do not shine with high performance in Adobe Photoshop either. If performance in this application is measured not by applying resource-intensive filters to an image, but by simulating typical versatile processing, then the performance of eight-core and six-core FX processors with the Piledriver microarchitecture turns out to be worse than that of the dual-core Core i3-3240. But this weak result can actually be interpreted as progress, since FX, built on the microarchitecture of the previous generation, are outperformed even by the Pentium G2120.
We also tested in the graphics program Adobe Photoshop Lightroom 4.2. The test script includes post-processing and export to JPEG of two hundred 12-megapixel images in RAW format.
Adobe Lightroom can process photos in several threads at once, which inevitably affects the results and improves the performance of AMD FX processors. However, even the fact that the advantage of Vishera variants with a different number of computing cores over the corresponding Zambezi versions reaches 16 percent does not give them the opportunity to overtake the Core i5 on the Ivy Bridge microarchitecture. The younger version of the new generation FX is completely behind the FX-4170, which once again hints at the failure of AMD's decision to reduce the L3 cache in the FX-4300.
Performance in Adobe Premiere Pro CS6 is tested by measuring the render time to H.264 Blu-Ray format of a project containing HDV 1080p25 footage with various effects applied.
Processing HD video content is one of the most favorable workloads for AMD multi-core processors. In addition, the introduction of the Piledriver microarchitecture played a very positive role in applications of this type. The advantage of Vishera models over Zambezi with the same number of cores reaches 21 percent, and on average it is about 12 percent. As a result, the FX-8350 provides better performance than the core i5-3570K, the FX-8320 almost falls short of the Core i5-3470 in speed, and the FX-6300 and FX-4300 confidently outperform Intel's Ivy Bridge dual-core processors, including those models. , which implement Hyper-Threading technology.
The x264 HD Benchmark 5.0 is used to measure the speed of transcoding video to H.264, based on measuring the processing time of original MPEG-2 video recorded at 1080p at 20 Mbps. It should be noted that the results of this test are of great practical importance, since the x264 codec used in it underlies numerous popular transcoding utilities, for example, HandBrake, MeGUI, VirtualDub, and so on.
High-definition video transcoding should also be included among the tasks favorable for AMD solutions. This is clearly seen in the performance indicators of the 8000-series FX processors. They are able to compete not only with the Core i5, but can even outperform the LGA 1155 flagship Core i7-3770K. However, the high performance under multi-threaded load, which we note in the maximum modifications of Vishera, is by no means typical of simpler modifications. Note that the FX-6300 is 70% behind the FX-8350, and the FX-4300 is more than twice as slow as the older eight-core. As a result, representatives of the Socket AM3 + line with six and four cores, even in the most favorable case for them, fall short of the younger Core i5. Their destiny is rivalry with Core i3 class processors.
We measure computational performance and rendering speed in Autodesk 3ds max 2011 using the specialized test SPECapc for 3ds Max 2011.
Rendering is another good example of a multi-threaded workload where AMD processors can play to their strengths. The new Piledriver microarchitecture also performs well here. As a result, the FX-8350 outperforms the FX-8150 by 20 percent and is even faster than the Core i5-3570K. The lower-clocked FX-8320 outperforms the FX-8150 by only 6 percent, but that's enough for it to fit in with the current Ivy Bridge-designed Core i5s. The advantage of the six-core FX-6300 over the FX-6200 of the previous generation is 12 percent, but it lags behind the Core i5 series. The quad-core FX-4300 performs about the same as the FX-4170 in terms of performance, and this puts it on a par with the Core i3.
power usage
Getting acquainted earlier with the FX-8350 processor, we came to the conclusion that the new Piledriver microarchitecture did not lead to noticeable progress in the efficiency of Socket AM3+ processors. AMD's 32nm process offerings continue to be wildly voracious compared to their 22nm competitors. However, the Vishera modifications we are considering in this article with lower frequencies and a reduced number of cores may turn out to be a little more economical than the flagship model. Moreover, the six-core and quad-core modifications have a maximum calculated heat dissipation of 95 W, and not 125 W, like their “full-fledged” counterparts.To get a complete picture of the power consumption of all processors in the updated AMD FX line, we conducted special testing. The new Corsair AX1200i digital power supply we use in our test system allows us to monitor the consumed and output electrical power, which we use for our measurements. The following graphs, unless otherwise noted, show the total consumption of systems (without monitor) measured at the output of the power supply and is the sum of the power consumption of all components involved in the system. The efficiency of the power supply itself is not taken into account in this case. During the measurements, the load on the processors was created by the 64-bit version of the LinX 0.6.4 utility. In addition, in order to correctly assess idle power consumption, we activated the turbo mode and all available energy-saving technologies: C1E, C6, Enhanced Intel SpeedStep and AMD Cool "n" Quiet.
At idle, all processors and platforms show approximately the same consumption. Being idle, any modern processors go into specialized energy-saving states, in which their consumption is extremely small and amounts to a few watts. Under such conditions, the energy appetites of other system components and the efficiency of the motherboard's power converter come to the fore, which mask the purely processor power consumption.
Already with a single-threaded load, you can trace the trends that are unpleasant for AMD products. The fastest and most power-hungry LGA 1155 processor, the Core i7-3770K, consumes significantly less power than the most economical member of the AMD FX series. At the same time, the performance of AMD processors under this kind of load is significantly lower than that of Intel processors. Yes, the Vishera generation began to consume electricity not as wastefully as the Zambezi series, but it is still impossible to talk about some kind of qualitative change.
The picture with the power consumption of the new FX processors and with a multi-threaded load is no better. The FX-8350 turns out to be the most gluttonous in today's CPU testing. Under load, it consumes 4 percent more than the FX-8150 and two-thirds the consumption of Intel's Core i7-3770K. Other representatives of the Vishera-designed FX series slightly improve power consumption compared to their predecessors in the Zambezi series, but like the flagship model, they are completely incomparable in this characteristic with those CPUs offered by a competing manufacturer today. Meanwhile, it should be noted that the FX-6300 and FX-4300 endowed with a 95-watt thermal package look much better compared to other Socket AM3+ processors. If old Core i5 processors manufactured using 32nm technology took part in our testing, then, perhaps, AMD's new dual-module quad-core processor could be compared with it in terms of energy costs. But, of course, in terms of the speed achieved in this case, such processors are completely incomparable.
In other words, in terms of performance per watt, modern AMD processors are hopelessly behind their Intel competitors. If, when testing performance, we often manage to find cases where older eight-core Visheras can compete with quad-core Ivy Bridges, then when measuring power consumption, the illusions are dispelled. AMD products are worse.
Overclocking
The AMD FX processor series is classified by the manufacturer as an overclocking solution. All of them have unlocked multipliers, that is, they allow for simple overclocking. At the same time, the frequency potential of both Zambezi and Vishera is quite sufficient to get a serious increase in performance, however, at the cost of a significant increase in power consumption. However, overclocking is an important competitive advantage for the FX series. Intel processors are also free to overclock only if their price exceeds $220. AMD, on the other hand, does not fix any obstacles for economical enthusiasts, which partly determines the popularity of the Socket AM3 + platform.Talking about the Piledriver microarchitecture, AMD emphasized the increase in its frequency potential. Therefore, we hoped that the Vishera processors would be more interesting in overclocking than their predecessors, whose average result "under the air" was a frequency of 4.6 GHz. However, our first tests showed that the new products do not have much improvement - the first FX-8350 that got into the laboratory overclocked only up to 4.7 GHz.
However, it would be wrong to judge the overclocking prospects of a whole family of CPUs by one instance, so we did not fail to explore the overclocking potential of all four Vishera models. As part of the tests, we did not set ourselves the goal of getting the highest possible overclocking, the task was different - to determine the frequency at which the new FX processors can operate in advanced user systems for a long time in 24/7 mode. Therefore, we limited the voltage on the processors to 1.55 V, which was recommended to us for the Vishera by AMD specialists, and the cooling was carried out by a serial air cooler NZXT Havik 140. The stability of the system in the overclocked state was confirmed by the OCCT 4.3. set).
First of all, we repeated the tests of the FX-8350 processor. There was a hope that the 4.7 GHz we got in our previous experiments was an unsuccessful overclocking, which is typical only for our first copy of the CPU. But since then we've had a chance to try a couple more FX-8350 samples, and they could not improve this result. So 4.7 GHz is quite a typical frequency of an overclocked flagship processor of the FX series when using serial air cooling.
Many people think that not the most successful semiconductor crystals are selected by the manufacturer for junior processor models, so their overclocking potential is significantly lower. In fact, this trend is observed quite rarely, and our sample FX-8320 did not confirm it. It confidently overclocked to 4.6 GHz, which is only 100 MHz less than the result of a more expensive brother.
The six-core processor FX-6300, in which one of the four dual-core modules is disabled on the semiconductor chip, showed approximately the same results during overclocking as the "full-fledged" Vishera processors. The maximum frequency at which it remained stable with air cooling was 4.7 GHz.
Overclocking of the AMD FX-4300 was especially hopeful, as a number of sources published information about the possibility of increasing their clock frequency to 5 GHz with air cooling. However, this information was not confirmed. Our copy of the processor with a halved number of active computing cores and with a halved L3 cache was able to work stably only at a typical for all Vishera frequencies of 4.6 GHz. At least not without involving potentially dangerous levels of processor core voltage.
Thus, any FX processors based on the Piledriver microarchitecture, regardless of the number of cores, overclock approximately the same when air-cooled - up to 4.6-4.7 GHz. This is a little better than the previous generation of AMD FX could provide, but we are not talking about any qualitative change in their frequency potential. Nevertheless, overclockers should be quite pleased with the results, which are typical for processors manufactured using 32nm technology.
At the same time, it should be noted that the FX-6300 and FX-4300 processors heat up rather weakly during overclocking with an increase in voltage to 1.5-1.55 V. For a six-core processor, the maximum temperature in our testing was 65 degrees, and for a quad-core one, a completely ridiculous 53 degrees. This means that no one bothers to raise the supply voltage and achieve stability at higher frequencies. However, we don't approve of this approach: an excessive increase in the supply voltage can lead to degradation of the processor chip, so it is unacceptable for continuous operation.
conclusions
In this study, we have actually repeated our first test Socket AM3+ processors with the Piledriver microarchitecture, with the only difference being that now not only the older Vishera model took part in it, but the entire lineup. And this allowed us to somewhat reconsider our attitude towards the novelties proposed by AMD. And that's why.The flagship FX-8350 really looks quite interesting. It provides a noticeable increase in performance compared to previous generation AMD offerings and, under multi-threaded load, is capable of competing with the top models of LGA 1155 processors of the Ivy Bridge generation. Given the democratic cost, the FX-8350 can be recommended for installation in inexpensive desktop systems aimed at solving resource-intensive tasks, such as processing and creating high-resolution content or final rendering. However, before you decide on this AMD proposal, you should also accept its drawbacks. Among them is not only a monstrous level of power consumption by modern standards, but also a lack of universality, expressed in modest performance in everyday and general applications, the vast majority of which cannot split the load into eight cores. Separately, we note that 3D games are among such unfavorable tasks for AMD processors.
If, nevertheless, you liked the FX-8350, then it makes sense to turn your attention to the FX-8320 model as well. It is noticeably cheaper, but, in fact, it offers everything the same - in professional applications, its speed is at its best. Moreover, given that all modern Socket AM3+ processors belong to the Black Edition series, that is, they have unfixed multipliers, it is not difficult to overclock the FX-8320 to the flagship level or even higher. This allows us to call the FX-8320 one of the most interesting options for enthusiasts in terms of the ratio of multi-threaded performance and price. The only pity is that all this does not negate the shortcomings of Vishera - high power consumption and low speed with a low-threaded load. So, frankly, the FX-8320 is a good narrow-minded, but not comprehensive option.
Vishera's six-core modification, FX-6300, leaves a dull impression at first glance. This processor has one of the four dual-core modules deactivated, so its peak performance is low compared to Intel's quad-core processors, even under multi-threaded workloads. This is quite natural, because two modern AMD cores are similar in speed to one Intel core: this rule of thumb was also observed in today's testing. However, the first impression is deceptive, it is easily dispelled if you look at the price list. AMD has far-sightedly set the FX-6300 at such a price that it competes not with the Core i5, but with the Core i3. And this move opens up certain prospects for the six-core Vishera: as an alternative to Intel's dual cores, it looks very good. Moreover, with such a comparison, the FX-6300 even has its own trump cards. For example, unlike representatives of the Core i3 series, it can be overclocked.
But the youngest representative of the updated Vishera series, the FX-4300 processor, completely disappointed. On it, AMD was too carried away with the cuts, disabling not only half the cores, but also half the L3 cache. As a result, it turned out that the FX-4300 is not at all faster than the FX-4170 with the Bulldozer microarchitecture and slightly outperforms it, only in terms of economy. As a result, the performance of the quad-core Vishera is noticeably worse than that of Intel's Core i3, but the price is close to that of the FX-6300. Therefore, this model is of no real interest even to loyal fans of AMD products, who are likely to prefer the similarly priced A10-5800K with integrated graphics and higher clock speeds.
Thus, the temptation of Vishera-designed processors is primarily provided not by the advantages of the Piledriver microarchitecture, but by their low prices. From this position, the middle models look the most interesting: FX-8320 and FX-6300. It is on them that we recommend paying attention first of all, unless, of course, you are afraid of additional expenses for paying bills from the energy company. And do not forget - the performance of AMD processors is good under multi-threaded loads, but they are not peculiar to Intel's omnivorousness. Therefore, for everyday use, the Socket AM3 + platform is not very suitable, it is better to use it if your goal is to build an inexpensive workstation.
Bulldozer 2.0 or 1.5 yet?
A year ago, we studied the first-born of the new AMD architecture, and until recently, the FX-8150 remained the company's top solution. Despite some ambiguity, despite the expansion of the range of cheaper devices, the FX-8170, which was once actively discussed by many in the first quarter of this year (as well as in the second and third), did not come out, and talk about it subsided. The reasons are clear: last year it was not possible to release it, and this no longer made sense, as the company worked on architectural improvements in the new generation of processors. The Piledriver microarchitecture, which replaced the Bulldozer, debuted on the market in the first half of this year. At first, however, only as part of mobile APUs, but from the very beginning it was obvious that its expansion into the desktop segment (both as APUs and as the basis of high-performance "classic" processors) was only a matter of time. And not so big. Accordingly, the release of the new FX on the old architecture did not make sense, because it could not improve anything radically, but it was quite possible to spoil mass for a more modern competitor: it is obvious that the superiority of the already planned FX-8350 over the FX-8150 is greater than it would be over FX-8170.
And at the beginning of October we waited for desktop Trinity . In principle, Vishera could have been announced even then - in fact, we (like, apparently, most of the test labs) received samples of the A10-5800K and FX-8350 at the same time. However, to enhance the intrigue, the company decided not to share, but to present to the public APU and multi-module solutions separately - in order to finally spend October “under its own sign”. However, today all the veils of secrecy have fallen, so that we can not only evaluate the pros and cons of the new flagship of the line, but also introduce them to all readers. What are we going to do now.
Test stand configuration
| CPU | FX-8350 | FX-8150 | A10-5800K | Phenom II X6 1100T |
| Kernel name | Vishera | Zambezi | Trinity | Thuban |
| Production technology | 32 nm | 32 nm | 32 nm | 45 nm |
| Core frequency std/max, GHz | 4,0/4,2 | 3,6/4,2 | 3,8/4,2 | 3,3/3,7 |
| 8/8 | 8/8 | 4/4 | 6/6 | |
| L1 cache (total), I/D, KB | 256/128 | 256/128 | 128/64 | 384/384 |
| L2 cache, KB | 4×2048 | 4×2048 | 2×2048 | 6×512 |
| L3 cache, MiB | 8 | 8 | - | 6 |
| UnCore frequency, GHz | 2,2 | 2,2 | - | 2,0 |
| RAM | 2×DDR3-1866 | 2×DDR3-1866 | 2×DDR3-1866 | 2×DDR3-1333 |
| video core | - | - | Radeon HD 7660D | - |
| socket | AM3+ | AM3+ | FM2 | AM3 |
| TDP | 125 W | 125 W | 100 W | 125 W |
| Price | $218() | N/A(0) | $111() | N/A(0) |
The two main characters of the article are obvious - FX-8150 and FX-8350. As you can see, the old and new processors are very similar in terms of performance characteristics: four modules equipped with a large amount of third-level cache memory, manufactured using the same 32 nm process technology. They also have similar TDP and frequencies. More precisely, the frequency of the cache memory and the maximum frequency of the cores in Turbo mode are generally the same, but the starting one for a beginner has increased by more than 10%. Accordingly, with multi-threaded loading, the new processor has a significant advantage over the old one, even without taking into account intensive innovations. But single-threaded performance will increase only due to the architecture.
Therefore, after some deliberation, we decided to add the A10-5800K to the list of subjects. Yes, of course, these are processors of completely different classes - here there are half as many modules, and there is no cache. But the architecture is the same, the frequencies are close - in general, it will be interesting to compare processors in a low-threaded software environment. And for someone - not only in it: it may turn out that many consider the A10-5800K performance level sufficient for themselves, which, coupled with good video, lower power consumption and a much lower price, will easily outweigh the difference in performance :)
Another competitor taken directly from last year's article is the Phenom II X6 1100T. It is interesting to us because the FX-8150 sometimes overtook the old flagship slightly, and there were also tests in which it completely lagged behind it, which made a number of fans of AMD products discouraged. Now there is no direct competition between FX and the older Phenom II, because, formally continuing their supplies (and even lowering prices), the company has reduced the shipment of top Phenom II to a minimum, but there are a lot of them on hand, so compare the performance with the new top solution interesting and helpful.
| CPU | Core i5-2500 | Core i5-3570K | Core i7-2600 | Core i7-3770K |
| Kernel name | Sandy Bridge QC | Ivy Bridge QC | Sandy Bridge QC | Ivy Bridge QC |
| Production technology | 32 nm | 22 nm | 32 nm | 22 nm |
| Core frequency std/max, GHz | 3,3/3,7 | 3,4/3,8 | 3,4/3,8 | 3,5/3,9 |
| Number of cores/threads of calculation | 4/4 | 4/4 | 4/8 | 4/8 |
| L1 cache (total), I/D, KB | 128/128 | 128/128 | 128/128 | 128/128 |
| L2 cache, KB | 4×256 | 4×256 | 4×256 | 4×256 |
| L3 cache, MiB | 6 | 6 | 8 | 8 |
| UnCore frequency, GHz | 3,3 | 3,4 | 3,4 | 3,5 |
| RAM | 2×DDR3-1333 | 2×DDR3-1600 | 2×DDR3-1333 | 2×DDR3-1600 |
| video core | HDG2000 | HDG4000 | HDG2000 | HDG4000 |
| socket | LGA1155 | LGA1155 | LGA1155 | LGA1155 |
| TDP | 95 W | 77 W | 95 W | 77 W |
| Price | $229() | $284() | $340() | $431() |
Well, since we extracted this “old man” from the archive, it is also logical to take two pairs of Intel processors. We compared the FX-8150 with the Core i5-2500 and Core i7-2600, as it fell in the price range between them. True, it fit in only at the beginning, and then it fell considerably in price: first, to the level of older Core i5, recently it “rolled down” to the younger ones, making room for the FX-8350 and FX-8320. Judging by preliminary information about prices, this pair should not compete with Core i7 at all, initially “dancing” somewhere in the region of older Core i5. But for the sake of clarity, we will take not only the Core i5-3570K, but also the Core i7-3770K (it is possible without the “K”, since it is noticeably cheaper and literally half a percent slower, but limited overclockable, unlike any FX) . And how the final retail prices will behave - this is not predictable at first. In the end, they are initially overpriced for new items, but processors for LGA1155 have not been such for a long time, so we won’t be surprised if somewhere we sometimes some retailers will sell the FX-8350 at very close prices to the Core i7.
| Motherboard | RAM | |
| AM3+ | ASUS Crosshair V Formula (990FX) | |
| AM3 | ASUS M4A78T-E (790GX) | Corsair Vengeance CMZ8GX3M2A1600C9B (2×1333; 9-9-9-24-2T, Unganged Mode) |
| FM2 | MSI FM2-A85XA-G65 (A85) | G.Skill F3-14900CL9D-8GBXL (2×1866; 9-10-9-28) |
| LGA1155 | Biostar TH67XE (H67) | Corsair Vengeance CMZ8GX3M2A1600C9B (2×1333/1066; 9-9-9-24 / 8-8-8-20) |
Testing
Traditionally, we divide all tests into a number of groups and show the average result for a group of tests/applications on the diagrams (for details on the testing methodology, see a separate article). The results in the diagrams are given in points, for 100 points the performance of the reference test system, the site of the sample of 2011, is taken. It is based on the AMD Athlon II X4 620 processor, but the amount of memory (8 GB) and video card () are standard for all tests of the "main line" and can only be changed as part of special studies. Those who are interested in more detailed information are again traditionally invited to download a table in Microsoft Excel format, in which all the results are shown both in converted points and in "natural" form.
Interactive work in 3D packages
A traditionally low-threaded group of applications, also gravitating towards Intel products, so there was no talk of intercompany competition initially, and even now it is not particularly observed. Something else is more interesting: three out of four AMD processors demonstrate the same level of performance, but the FX-8350 noticeably breaks away from them. We recall that in this mode, the clock frequencies of the other trio (when the new flagship also enters) are approximately the same, and we conclude that some Bulldozer bottlenecks in the Piledriver really managed to be “expanded”: the FX-8150 slightly outperforms the A10-5800K only due to the presence cache memory of the third level, but it fails to fight on equal terms with the FX-8350, where there is a cache and a new architecture.
Final rendering of 3D scenes
Also a very interesting picture. Such a load was previously devastating for Bulldozer: only four vector units did not allow competing not only with the Core i5 (there are also four of them, but faster), but also with the Phenom II X6. And Piledriver has been improved internally. Plus, we managed to increase the frequencies at full load, so that the FX-8350 is no longer inferior to the above-mentioned processors, significantly outperforming its predecessor. It is clear that this is still not enough to compete with the eight-thread Core i7 of the second and third generations (the first one also works at the level of the new Core i5), but the performance of the Core i5-3450 in such an inconvenient group of tests is very OK.
Packing and unpacking
It is noteworthy that the once top Phenom II X6 1100T, despite the large number of cores (one test out of four can use them) and the presence of L3 (which is important for all four), is only 5% ahead of the A10-5800K, where both computational threads and less cache. A good answer to those who are still sure of the wrong choice of path - they say, it was not necessary to develop a new architecture, but it was worth just transferring Phenom II to a new technical process. Well, they would do it - so what? As you can see, there are tasks that are “convenient” for construction equipment. Actually, the FX-8150 also looked good in archiving tests, having performance at the level of the second generation Core i5 and some Core i7 of the first. And the FX-8350 added 10%, which already allows it to overtake all the third generation Core i5 and get closer to the second generation Core i7.
Audio encoding
And an even more convenient load for the FX line, under which the FX-8150 easily outperformed any Core i5. But it did not claim to be equal to the Core i7, and the FX-8350 is capable of this. The reason is clear: a 10% increase in performance per thread with the full utilization of all eight “half-cores” adds up to a 10% increase in clock frequency, which in the end gives already 20%.
Compilation
The same is true here. But the performance gain is somewhat less - cache memory is important for these applications, and the L3 characteristics in the new microarchitecture have not changed much (if only due to internal optimizations, both the volume and the frequency have remained the same). But even about 15% is also very good, since it allowed us to break away far from the Phenom II X6 1100T (the progress compared to which the FX-8150 was, to put it mildly, not impressive), and even with modern Core i7 "wrestle" by almost equal.
Mathematical and engineering calculations
Let's return to the low-threaded (for the most part) group of tests, in which so far there are no hints of direct competition with Intel products. But, which is also important, the FX-8350 in it is the fastest of the AMD processors, and earlier all FX were inferior to the older Phenom II. In general, let the victory over yourself, but not unimportant.
Raster graphics
This group occupies a neutral position: on the one hand, there is an increase over its predecessors, and a considerable one; on the other hand, the lagging behind the older Intel processors was only reduced, and not completely overcome. However, the FX-8150 overtook the only Core i3-2100, and all the Core i3s (including the latest ones), and even the Core i5-2400s (and slower ones, respectively), are already behind the FX-8350, so not everything is like that. simply.
Vector graphics
This group of programs does not like new modules - and this is putting it mildly. In general, as before, the Phenom II is still faster in them. But at least not by 20%, as it was last year, so the effect of the architecture update should not be underestimated.
Video encoding
But here the effect is extremely small, since the video encoding programs treated the FX-8150 very well. On the other hand, higher frequencies still have an effect, so if earlier the Core i5-2400 was faster than all AMD processors, then the FX-8350 is on average faster than even the i5-2500/2500K.
Office software
Not for the first time, we are not reaching the competitors of “other colors”, but we managed to break away from the “compatriots”. There is still a lot in common between such diametrically opposed classes of software as "office" programs and 3D modeling packages;) In the latter, perhaps the gap from Core is smaller, but the principles are often similar. Although, it would seem, where is Maya - and where is Word!
Java
Another example of a "favorable" load for "building modules", which the FX-8150 still could not manage in full, only slightly outperforming the Phenom II X6 1100T. As for the victory over the Core i5, firstly, it was predictable (eight threads against four), and secondly, it did not last long: the appearance of the third-generation Core i5 “overthrew” the Bulldozer from not the first, but the prize. But the FX-8350 managed not only to restore the status quo, but also to get as close as possible to the Core i7 for LGA1155. Albeit only to the "old man" Core i7-2600 - but this is also the result.
Games
As we (and not only us) have said more than once, for any video card there is a certain level of processor performance, after which the latter ceases to be significant. However, as we can see, AMD processors have not yet reached the level corresponding to the GTX 570, so they have room to grow. And the new architecture is better than the old one in this regard. Although Phenom II X6 1100T is not the best representative of the old Phenom II X4 980 due to the peculiarities of the F1 2010 operation on six computational threads, the latter is also in terms of performance between the A10-5800K and FX-8150, i.e. between the Piledriver without L3 and Bulldozer with L3. And Piledriver with L3 (i.e. FX-8350) is over 5% faster! But eight threads of computing are not very relevant for the time being, so gamers may very well like the new dual-module FX-4300 the most: with comparable performance, but significantly cheaper.
Multitasking environment
During the operation, this experimental test showed good predictability and repeatability, so it did not present any revelations this time either: everything is similar to multi-threaded "individual" applications. Well, since the bottleneck in the form of cache memory performance of the third level has not disappeared, the situation is closest to compilers - the FX-8150 was already able to bypass both the Phenom II X6 1100T and any Core i5, and the FX-8350 by another 10% faster, allowing it to get as close as possible to the Core i7. And just such types of loads show why four “dual-core” modules are needed. How, however, and why do Intel processors need Hyper-Threading technology.
Total
Starting from the very first test results, we had a vague regret about the unfulfilled: how simple everything would be if this processor was called FX-8150 and came out a year ago! Indeed: the superiority over the best Phenom II X6 1100T is noticeable, not nominal; despite the large number of one-two-threaded applications in the testing methodology (and in real life), the final result is at the level of the older second-generation Core i5 with a noticeable lead over the latter - such a representative of the new architecture would look extremely convincing. But history (including computer history) does not tolerate the subjunctive mood. Therefore, we have what we have - it is already October 2012, and not 2011. It only helped that Intel also did not speed up the third generation of Core too much relative to the second, and 10% "on average" is one and a half times less than the 15% achieved by the new generation of AMD.
If you look not only on average, then there are 17-20%. However, Intel also has more than “its own” 10%. Moreover, if you look closely, the most significant increase in performance in multi-threaded groups of tests. Which is connected not only and not so much with the architecture, but with the fact that both companies increased the frequencies in this mode. Intel (formally) to a lesser extent, but in the third-generation Core, Turbo Boost was also “turned up”. But AMD had to sharply increase the starting frequency, which can serve as an indirect confirmation that the Turbo Core (albeit already 3.0) is still less efficient than the competing earlier development. However, indirect confirmation is not needed here - a direct one is enough: the TS only works when some of the modules (or at least their halves) are turned off, and the TV increases the frequencies even at full load.
In addition, it becomes clear why the FX-8170 did not come out: the superiority over it would be less. Here, Intel somewhat blurred the triumph of Ivy Bridge, first increasing the starting frequencies of the second generation (in the i5-2550K and i7-2700K models) and only then releasing their successors, while AMD could not do this, if only for reasons of prestige. Therefore (in fact) the increase in performance between 2011 and 2012 for companies differs by about a factor of two. Fans of conspiracy theories may see this as collusion and / or Intel's desire to support a competitor a little. Although, in fact, the company went a little differently: yes, top models accelerated slightly, but energy-efficient ones grew significantly. In particular, they managed to release the Core i7-3770T (the first desktop Core i7 with a 45W TDP) and the i7-3612QM (the first eight-threaded notebook processor with a TDP of 35W - previously only dual-core Core i7s were capable of this). And in the region of 65 W, the frequencies have grown a lot: the Core i7-3770S - 3.1-3.9 GHz (i.e. + 10% frequency at full load) and HDG 4000. And now remember that at a constant frequency of 3.2 GHz Ivy Bridge scored 195 points - the real i7-3770S should be slightly better (if only because of the active "boost" in low-threaded applications ). Maybe a little worse, but it doesn't matter: as we saw today, the FX-8350 is only capable of 186 points - without any video core (except with the Radeon HD 4290 chipset, which is comparable to the second generation HDG at best) and at 125 W against a faster processor part with video and 65 watts.
On the other hand, it's not that scary. The older multi-module FX still do not pretend to be used in laptops or monoblocks - there is Trinity for this. Which are quite out of place there - even from a comparison of the A10-5800K and FX-8350, we can conclude that, on average, the second one is “only” one and a half times faster, and in programs for domestic use (and close to them in terms of the logic of work) due to the traditional low-flow of the latter, the difference is reduced to 10-15%. Plus the most powerful integrated video core. The A10-5800K, of course, is also not suitable for compact systems, but even in the desktop line, AMD has the A10-5700 with a TDP of 65 W with exactly the same video part and slightly lower processor performance. And the destiny of FX in such conditions is systems with a deliberately “heavy” load and a considerable number of other energy-consuming peripherals. First of all, powerful discrete video cards come to mind that can “gobble up” 200-300 W, against which the difference in TDP between different processors quickly fades and evaporates. It is clear that competition in this segment is not easy, however ... However, under a multi-threaded load, the new FX are already unconditionally inferior to at least Core i5, and often reach the level of Core i7. Plus, the complete freedom of overclocking, coveted by a considerable number of potential buyers. Which, by the way, generally makes (at least for them) a conversation about power consumption not very relevant, but it is aggravated by the fact that (as we already wrote) some FX disadvantages become advantages - it is easier to “remove” the necessary amount of heat from a large crystal than from small.
Ultimately, the following overall picture emerges. The second generation of FX did not become a radical breakthrough (just as the first did not) - the revolution is once again postponed, already before the Steamroller. However, the company managed to improve the performance of end devices and prove the viability of the new architecture in practice. Of course, the fact that the proof turned out to be two-stage is somewhat disappointing, but the main thing is that it did turn out. Moreover, such situations happened in the past - what was the first Phenom worth (and the first FX were definitely not such a failure). This is not enough to compete in the top segment, but in the mass (and so) it is quite possible - in any case, the FX-8000 are and will continue to be the cheapest processors that support eight computation threads (along with some younger Xeon models) for the foreseeable future. , of course, but these are very specific products, and still somewhat more expensive). Thus, for those who demand it, they can be a very good choice. Moreover, the new generation does this with much fewer reservations than the previous one.
At the moment, when building a computer based on an AMD processor, if you want to get a high-performance system with a multi-core processor and eight megabytes of L3 cache, then there is simply no alternative to the Vishera series. The rest of the models of the “A” line on the Kaveri and Richland cores do not have a third-level cache and there are only 4 cores in them. Their CPU performance was cut to ensure the normal operation of the integrated graphics core. The last bastion of performance is the Vishera, which is being phased out. Therefore, I suggest you read this review and have time to buy a good stone.
The youngest of the line, the lowest productive relative to its counterparts. However, even it has a cache of the third level, although it could not do without minuses: the cores got under the knife and there are only 4 of them left. Naturally, this was done to reduce the cost of the model. The speed of the cache has been cut, its volume is only 4 MB. But, it is worth considering that this processor was released simultaneously with the release of the core itself in October 2012, respectively, it was replaced by more productive versions, which will be described below.
Link to Yandex Market.
FX-4350
Favorably differs from the FX 4300 by the L3 cache size of 8 MB and its speed. The clock frequency has been increased by more than 10%, and the overclocking potential has been pushed back by the same amount - 5.1 GHz. At the same time, it is not much more expensive, although it heats up noticeably more - heat dissipation is 125 watts. You have to put in good cooling. Released in April 2013, the 4350 was supposed to replace the 4300, which it did well.
Link to Yandex Market.
FX-6300
Released in October 12, the 6300 has 8 cores, but they have not yet been optimized for speed. L3 cache 8 MB, L2 cache 6 MB. Heat dissipation is very low for AMD processors, which are famous for their "hot temper", is only 95 watts. The clock frequency of operation is 3.5 GHz.
A fairly good option in terms of price / performance ratio, but it is better to look at other models before making a final choice.
Link to Yandex Market.
FX-6350
Released almost a year after the announcement of the Vishera core, improved in terms of performance and stability. Has an increased frequency compared to its predecessor 6300. Standard clock frequency is 3.9 GHz. It lends itself well to overclocking to 4.6 GHz in air and 4.8 GHz in water.
Link to Yandex Market.
FX-8320
So we got to full-fledged eight-core processors. The basis of the entire Vishera line is the FX 83XX index, which has 8 cores, L2 and L3 caches of 8 MB, and is capable of running in 8 full threads. very popular with overclockers who managed to overclock this model from stock 3500 MHz to an incredible 4640 MHz with air cooling (cooler) and up to 4860 MHz with water cooling. 
At the same time, at a price it compares favorably with the Intel / / Core i7 family. The only thing that casts a shadow on this superiority is TDP 125 Wt, but this problem is small.
Link to Yandex Market.
FX-8350
The updated version of the processor has become more productive due to factory overclocking and an increase in typical power consumption by one and a half times from 101 W to 160 W! Accordingly, you will have to take great care of cooling. But in return, you get stock 4GHz and overclocking potential up to 5GHz. It is with this potential that the FX-8350 has earned the love of overclockers from all over the world. Here she has no equal, because no one else produces mass-produced processors with eight cores that can operate at a frequency of 5000 MHz. 
Everyone is looking for the justification for such overclocking for himself. If you need a platform to play the latest games at 1920×1080 resolution and ultra settings, then only top-end Intel Core i7 processors can become an alternative to the FX-8350.
Link to Yandex Market.
FX-9370
The novelty appeared in June 2013. The stock frequency is now 4.4 GHz, overclocking to 5.1 GHz is not difficult: 
If the budget allows you to buy this model, then it will be a good competitor to the top Core i7. In conjunction with a good graphics card of the GTX770 level and above, you will be able to play all modern games at ultra settings with complete freedom. It's hard to say whether such a performance margin will last for a long time or the gaming industry will give us another voracious monster, but today there are very few alternatives to the FX-9370 processor.