Ed.: Originally, this story contained a page exploring resonant clock mesh technology, from Cyclos Semiconductor, which was expected to surface in AMD's Piledriver-based SoCs. Upon discussing this in greater depth with the company, however, "the timing of the products and the implementation of resonant clock mesh caused [the technology] to not be productized with "Piledriver" based processors." As such, we've removed that page to avoid any confusion.
As someone who reviews computer hardware, the challenges facing a business matter a lot less to me than the products it sells. We can all agree that the last year was pretty awful for AMD’s processor team, beginning with the power-hungry Bulldozer-based FX CPUs that slowly slid down in price over 12 months in response to a more compelling family of third-gen Core processors from Intel. But every delivery to my SoCal lab represents an opportunity. Chatter about mismanagement, layoffs, and a troubled past matter little in the evaluation of a new CPU. So, let's get down to business.
Sometimes it’s possible to predict the outcome of a story. Had AMD shipped an FX-8170 running 200 MHz faster than its former flagship, I would have guessed that it’d exhibit the same issues with lightly-threaded workloads, it would probably outperform a Core i5-2500K in more intensive tasks, but its power consumption would probably look pretty nasty compared to modern 77 W Ivy Bridge-based chips.
Instead, we have an FX-8350, which centers on the same Piledriver architecture as the Trinity-based APUs introduced less than one month ago. Experience tells us that, per core and per cycle, Piledriver can be as much as 15% faster than a Bulldozer-based design. Add to that the fact that FX-8350 operates at least 400 MHz faster than FX-8150. Oh, yeah. And Ivy Bridge only gave Intel’s line-up single-digit gains. There’s every chance that today’s performance comparison is going to be a lot more interesting than the overwhelmingly negative judgment passed on FX-8150 one year ago.
Meet The Piledriver-Based FX Family
In a faithful return to its Bulldozer approach, AMD sent out the fastest model in its new line-up for review, but is actually introducing eight-, six-, and four-core configurations. Of course, we know they all center on the Piledriver architecture, but the SoC itself is referred to as Vishera, and it’s still branded as FX.
A complete Vishera processor measures 315 square millimeters and is composed of 1.2 billion transistors. Those are the exact same figures given for Zambezi, the previous-generation SoC based on AMD’s Bulldozer architecture. Whatever alterations the architects made—from ISA extensions to a larger L1 DTLB—surface area and transistor count don’t change enough to move the dial on those contextually-irrelevant but interesting-to-know-anyway specifications.
| AMD's 2012 FX Family | |||||||
|---|---|---|---|---|---|---|---|
| Core Count | Base Clock | Max. Turbo | NB Clock | TDP | Price | OPN | |
| FX-8350 | 8C / 8T | 4.0 GHz | 4.2 GHz | 2200 MHz | 125 W | $195 | FD8350FRW8KHK |
| FX-8320 | 8C / 8T | 3.5 GHz | 4.0 GHz | 2200 MHz | 125 W | $169 | FD8320FRW8KHK |
| FX-6300 | 6C / 6T | 3.5 GHz | 4.1 GHz | 2000 MHz | 95 W | $132 | FD6300WMW6KHK |
| FX-4300 | 4C / 4T | 3.8 GHz | 4.0 GHz | 2000 MHz | 95 W | $122 | FD4300WMW4MHK |
Two of the four SKUs boast eight integer cores, or four Piledriver modules, however you choose to label AMD’s compute units. The flagship, FX-8350, features a base clock rate of 4 GHz. Turbo Core technology is able to push that to 4.2 GHz in lightly-threaded workloads, though most of the chip’s speed-up undoubtedly comes from its default state. How much does Turbo Core really do for FX-8350? Not much. In iTunes, our single-threaded benchmark finishes three seconds faster with the feature on.

An FX-8320 drops the base clock rate to 3.5 GHz, though Turbo Core pushes that to 4 GHz under defined thermal limits (a 500 MHz speed-up is likely more meaningful to FX-8320). Both eight-core models include 8 MB of L2 cache (split into one shared 2 MB slice per module) and 8 MB of L3 cache (shared between all four of the SoC’s modules). AMD is suggesting a $195 price tag on FX-8350 and a $169 price on FX-8320.
FX-6300 comes equipped with three active modules (six cores) and drops pricing all the way to $132. A 3.5 GHz base clock rate helps take advantage of the architecture’s strengths in threaded apps, while a 4.1 GHz peak Turbo Core setting tries to compensate for lackluster single-threaded speed. Like the four-module parts, FX-6300 exposes 2 MB of shared L2 per module (totaling 6 MB in this case) and a shared 8 MB L3 cache. Fewer active resources (along with a slightly slower 2 GHz northbridge clock) allow FX-6300 to fit within a 95 W thermal ceiling, down from the 125 W limit imposed by both FX-83x0 processors.
A lone dual-module CPU, FX-4300, is also rated for 95 W. Its base 3.8 GHz clock rate is sped up to 4 GHz in lightly-threaded apps, and a 2 GHz northbridge frequency matches the FX-6300. A drop to 4 MB of shared L3, however, and a price tag just $10 under the triple-module chip will likely encourage most folks to spend the extra $10 bucks.
Although AMD’s architecture doesn’t seem particularly bandwidth-starved, Vishera’s dual-channel DDR3 memory controller officially supports 1866 MT/s data rates. Frankly, we’d rather use lower-latency 1600 MT/s modules to maximize value, particularly since our results show that you don't gain any performance (outside of Sandra 2013 Beta) after spending more on faster memory.
The entire FX line-up also features an unlocked ratio multiplier, which serves to simplify overclocking. Is Vishera scalable enough to make tuning worthwhile? What do you think about 5.125 GHz using a closed-loop liquid cooler?
Overclocking
Although AMD’s CPUs haven’t been the fastest for years now, the company continues to nurture enthusiast support through products that give power users certain features important to them. Software able to manipulate configuration settings in real-time on the Windows desktop, unlocked ratio multipliers, and platforms with far more PCI Express connectivity than the competition are just a few of the nods AMD gives to the folks who know and appreciate how each of those capabilities can improve platform performance.
The same overclockers who might have been disappointed in the scalability of AMD’s Zambezi-based FX CPUs using mainstream cooling might have something to get excited about this time around, even though we’re looking at a very similar architecture manufactured on the same 32 nm node.

Using a 1.375 V CPU voltage and a 1.175 V northbridge voltage, I was able to get FX-8350 running stably at 4.8 GHz under full load. In the screen capture above, I'm running a single-threaded test to spin the chip up, but the highlighted maximum temperature is where our benchmark suite peaked.

The FX-8350 wanted to go even faster, but the key here is a voltage setting low enough that you avoid hitting 70 degrees Celsius. At that point, the thermal monitor starts cycling cores to throttle down (evidenced in the image above), keeping the chip from getting any hotter and negatively impacting performance. So long as I didn’t trigger any threaded workloads, I was even able to run benchmarks as high as 5.125 GHz (requiring a 1.4375 V CPU voltage and 1.2 V northbridge setting).

Clearly, cooling capacity is going to be a bottleneck for most folks. I find AMD’s reference heat sink and fan to be dismally insufficient, and a beefy third-party upgrade is going to add to the FX’s cost. But, for the sake of testing, I used the closed-loop liquid cooler AMD offered with its FX processors last year to reach those overclocks. Something comparable will run you $70 or so. At that point, you’d probably want to consider a Core i7-3770 for $300 as an alternative. Fortunately, I have a -3770K in our benchmark results.
Overclocking to 4.8 GHz is enough to put the FX-8350 ahead of Intel's Core i7-3770 in a threaded workload like 3ds Max 2012, but it's not going to help AMD's Piledriver architecture overcome Ivy Bridge in a single-threaded test like iTunes. Of course, if you're willing to spend an extra $30 on a Core i7-3770K and even more on an aftermarket cooler, you can take the 3.5 GHz CPU to 4.5 GHz relatively easily and flip the script on AMD's overclocked FX, too.
Compatibility
All four Piledriver-based FX chips are compatible with the existing Socket AM3+ interface. Current FX-enabled motherboards need a BIOS update to recognize the new processors. However, boards that previously exhibited trouble with FX processors probably won’t get fixed. Asus’ Crosshair IV Formula, for example, should work with Bulldozer-based CPUs by AMD’s own claims.
Asus did add experimental support for the FX family back in 2011. However, the company never rolled in the update needed to stop a severe BSOD issue. As a result, many older AM3 platforms expected to properly support FX don’t, and we expect those problems to persist with Piledriver. We’ve heard assurances from AMD that the issue isn’t widespread, and that motherboard vendors are able to resolve it with an update. But certain manufacturers can’t seem to be bothered to bring older products up to date. Complaints about the Crosshair IV Formula and other motherboards are chronicled in this thread.
The very foundation of AMD’s current x86 architecture was covered in great depth back when I reviewed the FX-8150 (AMD Bulldozer Review: FX-8150 Gets Tested). All of those tenets carry over to the company’s Piledriver update. However, we know that AMD’s engineers learned a number of lessons as they took the original Bulldozer concept from theories and diagrams to actual silicon. We also know that process technology evolved over the last year, even if the company continues to use a 32 nm node for manufacturing its Vishera-based CPUs. It should come as no surprise, then, that today’s reformulation is the result of several tweaks flagged for improvement a long time ago.
Front-End Improvements
In the days that followed AMD’s Bulldozer introduction, branch prediction was suggested as one of the architecture’s possible weaknesses. The module concept involves certain shared resources feeding two execution threads, and the architects attempted to minimize bottlenecks in the front-end by implementing one prediction queue per thread behind a 512-entry L1 and 5000-entry L2 branch target buffer. For Piledriver, the company claims the accuracy of its predictor is better.
Piledriver adds support for a couple of ISA extensions that we first covered in our mobile Trinity-based APU coverage. The fused multiply-add was introduced a year ago in Bulldozer. That specific version was called FMA4, though, and allowed an instruction to have four operands. But Intel only plans to support a simpler three-operand FMA3 instruction set in its upcoming Haswell architecture, so AMD preempts that addition with Piledriver. The other extension, F16C, enables support for converting up to four half-precision to floating-point values at a time. Intel’s Ivy Bridge architecture already includes this, so its implementation on Piledriver simply catches AMD up. Not that Bulldozer was suffering without FMA3/F16C; compiler support was only added in Visual Studio 2012.
Inside The Integer Cluster
The two integer clusters in each compute module feature an out-of-order load/store unit capable of two 128-bit loads/cycle or one 128-bit store/cycle. AMD discovered that there were certain cases where Bulldozer wouldn’t catch store data already in a register file. By rectifying this, instructions are fed into the integer clusters more quickly.
Within each integer core, we’re still dealing with two execution units and two address generation units (referred to simply as AGens). Those AGens are more capable this time around in that they’re able to perform MOV instructions. When AGen activity is light, the architecture will shift MOVs over to those pipes.
One of the most notable changes is a larger translation lookaside buffer for the L1 data cache, which grows from 32 entries to 64. Because the L2 TLB has fairly high 20-cycle latency, improving the hit rate in L1 can yield significant performance gains in workloads that touch large data structures. This is particularly important in the server space, but AMD’s architects say they noticed certain games demonstrating sensitivity to this too, which isn’t something they had expected.
L2 Cache Optimizations
Hardware prefetching into the L2 is improved as well. Minimum latency doesn’t change, which is why cache latency doesn’t look any better in our Sandra 2013 benchmark. However, as the prefetcher and L2 are used more effectively, average latency (much more difficult to measure with a diagnostic) should be expected to drop, AMD claims. The same Sandra 2013 module also reflects very little change in L3 latency, and Vishera’s architects confirm that no changes were made to the L3 cache shared by all modules on an FX package.
Putting It All Together: Five Architectures At 4 GHz
What effect do all of those adjustments have on Piledriver's per-cycle performance? We ran five different architectures at 4 GHz to compare their relative results.

In iTunes, which we know to be single-threaded, the FX-8350 demonstrates significant gains over the Bulldozer-based FX-8150. But a Phenom II X6 1100T operating at the same frequency is still faster. And that's before we look at the Sandy and Ivy Bridge architectures, which jump way out in front of anything from AMD.

Notice that the Core i7 is listed as a quad-core CPU capable of addressing four threads. I disabled Hyper-Threading in this test to isolate core performance. Had it been turned on, Intel's client flagship would have likely finished in first place.
Nevertheless, we're most interested in the gain realized by shifting from FX-8150 to FX-8350, and it is significant. Again, though, Thuban's six cores manage to outmaneuver Vishera's quad-module configuration. AMD is using a clock rate advantage to keep this latest architecture in front of its older design. Thuban really doesn't want to run at such high frequencies, even as it's able to get more done per cycle.
| Test Hardware | |
|---|---|
| Processors | AMD FX-8350 (Vishera) 4.0 GHz (20 * 200 MHz), Socket AM3+, 8 MB Shared L3, Turbo Core enabled, Power-savings enabled |
| AMD FX-8150 (Zambezi) 3.6 GHz (18 * 200 MHz), Socket AM3+, 8 MB Shared L3, Turbo Core enabled, Power-savings enabled | |
| AMD Phenom II X4 980 BE (Deneb) 3.7 GHz (18.5 * 200 MHz), Socket AM3, 6 MB Shared L3, Power-savings enabled | |
| AMD Phenom II X6 1100T (Thuban) 3.3 GHz (16.5 * 200 MHz), Socket AM3, 6 MB Shared L3, Turbo Core enabled, Power-savings enabled | |
| Intel Core i7-3770K (Ivy Bridge) 3.5 GHz (35 * 100 MHz), LGA 1155, 8 MB Shared L3, Hyper-Threading enabled, Turbo Boost enabled, Power-savings enabled | |
| Intel Core i5-3570K (Ivy Bridge) 3.4 GHz (34 * 100 MHz), LGA 1155, 6 MB Shared L3, Turbo Boost enabled, Power-savings enabled | |
| Intel Core i5-3470 (Ivy Bridge) 3.2 GHz (32 * 100 MHz), LGA 1155, 6 MB Shared L3, Turbo Boost enabled, Power-savings enabled | |
| Intel Core i5-2550K (Sandy Bridge) 3.4 GHz (34 * 100 MHz), LGA 1155, 6 MB Shared L3, Turbo Boost enabled, Power-savings enabled | |
| Motherboard | ASRock Fatal1ty 990FX Professional (Socket AM3+) AMD 990FX/SB950 Chipset, BIOS 1.9 |
| Gigabyte Z77X-UD3H (LGA 1155) Intel Z77 Express, BIOS F17 | |
| Memory | G.Skill 16 GB (4 x 4 GB) DDR3-1600, F3-12800CL9Q2-32GBZL @ DDR3-1600 at 1.5 V |
| Hard Drive | Crucial m4 256 GB, SATA 6 Gb/s |
| Graphics | Nvidia GeForce GTX 680 2 GB |
| Power Supply | Cooler Master UCP-1000 W |
| System Software And Drivers | |
| Operating System | Windows 7 Ultimate 64-bit |
| DirectX | DirectX 11 |
| Graphics Driver | Nvidia GeForce Release 306.97 |
Notice that we're using DDR3-1600 memory in all of our test platforms (a fact that we'll point out again on the Sandra 2013 Beta results page). The FX-8350 officially supports modules running at up to 1866 MT/s, so this could be a potential performance issue if the platform is bottlenecked by bandwidth. Fortunately, it's not. We ran spot-checks on several benchmarks in order to make sure our decision to use 16 GB of DDR3-1600 instead of 8 GB of DDR3-1866 wouldn't reflect poorly on Vishera.

Aside from Sandra's memory throughput result, real-world benchmark's aren't affected. WinRAR is one of the most bandwidth-sensitive apps in our suite, and you can see that it stops scaling at 1600 MT/s.

| 3D Game Benchmarks And Settings | |
|---|---|
| Benchmark | Details |
| The Elder Scrolls V: Skyrim | Game Settings: High and Ultra Quality Settings, Anti-Aliasing: FXAA, V-sync: Disabled, 1680x1050, 1920x1200, 2560x1600, Custom Demo, 25-second Fraps run |
| Battlefield 3 | Game Settings: Ultra Quality Settings, Anti Aliasing: Disabled and 4xMSAA (Deferred)/High (Post), Anisotropic Filtering: 16x, Vertical Sync: Off, 1680x1050, 1920x1080, 2560x1600, Demo: Going Hunting, 90-second Fraps run |
| World of Warcraft: Mists of Pandaria | Game Settings: Ultra Quality Settings, Anti Aliasing: 1x AA and 8x AA, Anisotropic Filtering: 16x, Vertical Sync: Disabled, 1680x1050, 1920x1080, 2560x1600, Demo: Crushblow to The Krazzworks, DirectX 11 |
| Audio Benchmarks And Settings | |
| Benchmark | Details |
| iTunes | Version: 10.4.1.10, 64-bit Audio CD ("Terminator II" SE), 53 min., Convert to AAC audio format |
| Lame MP3 | Version 3.99 Audio CD "Terminator II SE", 53 min, convert WAV to MP3 audio format, Command: -b 160 --nores (160 Kb/s) |
| Video Benchmarks And Settings | |
| Benchmark | Details |
| HandBrake CLI | Version: 0.9.8 Video: Big Buck Bunny (720x480, 23.972 frames) 5 Minutes, Audio: Dolby Digital, 48 000 Hz, Six-Channel, English, to Video: AVC Audio: AC3 Audio2: AAC (High Profile) |
| MainConcept Reference v2.2 | Version: 2.2.0.5440 MPEG-2 to H.264, MainConcept H.264/AVC Codec, 28 sec HDTV 1920x1080 (MPEG-2), Audio: MPEG-2 (44.1 kHz, 2 Channel, 16-Bit, 224 Kb/s), Codec: H.264 Pro, Mode: PAL 50i (25 FPS), Profile: H.264 BD HDMV |
| Application Benchmarks And Settings | |
| Benchmark | Details |
| WinRAR | Version 4.20 RAR, Syntax "winrar a -r -m3", Benchmark: 2012-THG-Workload, 1.35 GB |
| WinZip | Version 17 WinZip Commandline Version 3, ZIPX, Syntax "-a -ez -p -r", Benchmark: 2012-THG-Workload, 1.35 GB |
| 7-Zip | Version 9.20 (x64) LZMA2, Syntax "a -t7z -r -m0=LZMA2 -mx=5", Benchmark: 2012-THG-Workload, 1.35 GB |
| Adobe Premiere Pro CS 6 | Hollywood Sequence to H.264 Blu-ray, Output 1920x1080, Maximum Quality |
| Adobe After Effects CS 6 | Create Video which includes three Streams Frames: 210, Render Multiple Frames Simultaneously: on |
| Cinebench | Version 11.5 Build CB25720DEMO CPU Test single- and multi-threaded |
| Blender | Version: 2.63, Cycles Engine Syntax blender -b thg.blend -f 1, Resolution: 1920x1080, Anti-Aliasing: 8x, Render: THG.blend frame 1 |
| Adobe Photoshop CS 6 (64-Bit) | Version: 11 Filtering a 16 MB TIF (15 000x7266), Filters:, Radial Blur (Amount: 10, Method: zoom, Quality: good) Shape Blur (Radius: 46 px; custom shape: Trademark sysmbol) Median (Radius: 1px) Polar Coordinates (Rectangular to Polar) |
| ABBYY FineReader | Version: 10 Professional Build (10.0.102.82) Read PDF save to Doc, Source: Political Economy (J. Broadhurst 1842) 111 Pages |
| 3ds Max 2012 | Render Space Flyby, 1440x1080, from Y: RAM Drive |
| Adobe Acrobat X Professional | PDF Document Creation (Print) from Microsoft PowerPoint 2010 |
| SolidWorks 2010 | PhotoView 360, 01-Lighter Explode.SLDASM Benchmark File, 1920x1080 Render, 1.44 Million Polygons, 256 AA Samples |
| Visual Studio 2010 | Google Chrome Compile, Scripted |
| Synthetic Benchmarks And Settings | |
| Benchmark | Details |
| PCMark 7 | Version: 1.0.4 |
| 3DMark 11 | Version 1.0.3 |
| SiSoftware Sandra | Version: 2013 Beta Processor Arithmetic, Multimedia, Cryptography, Memory Bandwidth, .NET Arithmetic, .NET Multimedia |

Futuremark’s PCMark 7 is on notice, in a way. In testing integrated graphics for Core i5-3570K, -3550, -3550S, And -3570T: Ivy Bridge Efficiency, I discovered that the application was weighing performance from the Ivy Bridge architecture’s Quick Sync feature incredibly aggressively. That won’t affect today’s story, given our use of Nvidia’s GeForce GTX 680, but the fact that AMD’s FX-8350 gets ranked below Intel’s Core i5-2550K indicates to us that the test isn’t fully taxing the Piledriver-based processor.
The FX’s performance in the Productivity and Entertainment suites mirrors the Overall benchmark’s results, though the Creativity and Computation sub-tests show the new FX-8350 doing much better. We’ll need to fire up the real-world metrics in our armada of benchmarks to draw more definitive conclusions.





Futuremark’s attempt to capture overall gaming performance with its 3DMark benchmark sees AMD’s FX-8350 improve on the FX-8150 by several hundred points, placing it behind the pricier Core i5-3570K and Core i7-3770K CPUs.
That’s a really big generalization, though. The sub-tests are where we see how each platform affects the overall gaming experience.

The only CPU that seems to impact graphics performance is AMD’s Phenom II X6 1100T, but only by a couple of percentage points. Everything else is on almost-identical footing as the Graphics suite purposely isolates our GeForce GTX 680.

The Physics sub-test is where processor performance plays a predominant role, as Futuremark partitions the simulated world it creates into several isolated regions, creating multiple threads.
When it’s fully taxed, the FX-8350 is able to best Intel’s Core i5-3570K, a $230 quad-core chip. Turning on Hyper-Threading on the Core i7, however, kicks performance up another notch.

The Combined sub-test also exhibits plenty of difference between processors. It’s threaded insofar as the simulated world is partitioned into multiple regions. The outcome isn’t the same as the Physics test, though, because there’s a considerable graphics workload applied as well. As we’ll see in the real-world game testing, AMD’s chips have a tendency to bottleneck our GPU, and so the FX-8350 slides down in the field behind three Ivy Bridge-based processors.




Remember that AMD’s Bulldozer architecture added hardware acceleration for AES encryption and decryption, and this naturally carries over to Piledriver. Basically, this capability is bottlenecked by memory bandwidth—the faster you can feed the CPU, the better it performs in the AES-256 benchmark.
After running a handful of spot-checks on performance, we decided to use 16 GB of DDR3-1600 memory instead of 8 GB of DDR3-1866, particularly since the RAM drive we use for alleviating storage bottlenecks would carve some of that memory away anyway, and the extra bandwidth did nothing to speed up our results. Nevertheless, FX-8350’s loss to FX-8150 is a result of that lower data rate. Had we used DDR3-1866 modules, the FX-8350 would have matched the Ivy Bridge-based chips at DDR3-1600 closer to 20 GB/s.




It doesn’t appear that Vishera’s cache latencies—often thought to be one of Zambezi’s performance issues—are any different. AMD's architects confirm the L3 isn't changed. The L2's minimum latency isn't any different, either. But average L2 latency should drop as a result of optimizations.

A nearly last-place finish for FX-8150 turns into a second-place prize for FX-8350 in our 3ds Max 2012 workload. This time last year we were wondering how AMD could justify charging more for its new flagship than a Core i5-2500K. Now, by bringing the price of FX-8350 down to $200, it’s beating a pricier Core i5-3570K and nipping at the heels of a $320 CPU.

Both FX-based processors actually do pretty well in Blender, similar to what we saw last year (even if this time around we’re using the newer Cycles engine). In fact, even the old Thuban-based Phenom II X6 manages to best Intel’s Core i5-3570K.

The Cinebench single-threaded test illustrates what we’ve known for a year: AMD’s per-core performance is pretty dismal compared to Intel’s. Piledriver at least allows AMD to match the older Stars architecture in the Phenom II X6 and X4 parts. However, Intel’s Ivy Bridge-based CPUs—particularly the Hyper-Threading-equipped Core i7—are much stronger.
Brute-forcing performance with higher clock rates and as many as eight integer clusters allows FX-8350 to snag a second-place finish in the threaded benchmark.

The same speed-up that changed AMD’s position in 3ds Max also affects the results in SolidWorks. The FX-8150 loses out to the $200 Core i5-3470. Meanwhile, the new model manages to top the Core i5-3570K, landing in second place yet again.

Our older Photoshop workload is composed of threaded filters able to tax each multi-core processor. Given the FX-8350’s first-place finish, it clearly scales to at least eight threads. The improvement over FX-8150 is quantifiable. Meanwhile, AMD’s older four- and six-core chips get left in the dust.

We have a second Photoshop benchmark that exploits the application’s support for OpenCL. The workload is completely different, so you can’t compare it to the chart above. That'd make sense though, since this test takes longer to run.
AMD’s platform just doesn’t do as well. It’s OpenCL-based, so Nvidia’s GeForce GTX 680 and PCI Express 2.0 do play a role. However, this is also a scenario AMD is telling us to expect more of moving forward, so I think it’s still very relevant. All four AMD chips slide back to the rear of the field behind Ivy and Sandy Bridge-based processors from Intel.

We move back over to workloads that isolate processor performance and AMD’s FX-8350 shows us what it’s working with once more. The FX-8150 didn’t do poorly in this test previously, so its position right behind the Core i5-3570K isn’t altogether surprising. But the -8350 soundly bests the Core i5 and pulls right up behind the Core i7. We’ve told this story before, and we know how it ends: when you’re dealing with a threaded application, AMD’s modular architecture does well.

At the same time, when you swap over to a usage model not as well adapted for parallelism, the module concept chokes up. After Effects CS 6 is no friendlier to AMD’s design than CS 5.5 was. A higher base clock rate does help the Piledriver-based FX-8350 land in front of the almost two-year-old Phenom II X6 1100T, but it trails the full complement of chips from Intel.

Presented with another threaded title like ABBYY’s FineReader, the FX-8350 barely trails Intel’s Core i7-3770K. Then again, FX-8150 was already outperforming the Core i5-3570K, so the tweaked architecture and higher clock rate only reinforce that result. Give this design a threaded application and it does well. Got it.

I swear I’m not purposely trying to mix things up by dropping in single-threaded tests right after the heavily parallelized ones. This is the same order I try to use in every review.
As if to remind us why FX-8150 got so thoroughly shredded upon its introduction, our Lame encoding workload shows a 4 GHz FX-8350 narrowly edging out the 3.7 GHz Phenom II X4 980 (which was introduced a year and a half ago, mind you). Although Piledriver appears to be an improvement over Bulldozer, AMD still has not matched the per-clock performance of its prior architecture, unfortunately.

We see the same outcome when we convert a PowerPoint presentation to PDF format—a single-threaded task. Vishera fares better than Zambezi, but all four Intel processors get this job done before the first AMD processor wraps it up.

Our Visual Studio benchmark sees FX-8350 cutting more than four minutes from our compile job—a better-than-10% speed-up compared to FX-8150. AMD’s latest cannot come close to Core i7-3770K, but the Piledriver-based processor does manage to best Intel’s Core i5-3570K.


Corel just introduced WinZip 17, and we’re finally satisfied with the application’s ability to fully tax multiple cores simultaneously. The latest version of WinZip supports OpenCL-based acceleration on AMD, Intel, and Nvidia hardware, and we enabled the feature throughout testing. This is only applied to files larger than 8 MB, though. Our test folder contains thousands of files, and there aren’t many larger than 8 MB. So, it makes very little difference whether we run our benchmark with or without OpenCL support turned on.
Regardless, in a massive shift from last year when we tested the FX-8150 under WinZip 14 and it placed dead-last, the FX-8350 finishes in second place. The FX-8150 takes fourth—and against a more capable field, no less.
AMD would likely point at our test and rightly note that an increasing number of performance-sensitive applications are moving this direction—they’re either able to exploit multiple x86 cores simultaneously or leverage graphics resources. This gives the company’s hardware a growing advantage. But we remain concerned about the power it draws while it’s delivering marginally better benchmark results. More on that soon.

In light of WinZip’s rise to prominence once again, WinRAR looks like it’s now the least-able to exploit on-die resources. Sure, it sees plenty of speed-up from Intel’s efficient Ivy Bridge architecture. And the fact that an FX-8350 edges out an FX-8150 suggests those extra 400 MHz are going to use. But the new chip’s fifth-place finish is a tell-tale sign that the hardware is underutilized.

The outcome in 7-Zip isn’t too far off from WinZip 17. The Core i7-3770K does very well, but is followed closely by the FX-8350. AMD’s FX-8150 isn’t far behind that.
It’s almost surprising to see a Phenom II X6 take fourth place, but when you consider the four quad-core parts bringing up the back of the pack, it becomes clearer that 7-Zip has an affinity for cores, even if it takes of six of AMD’s Stars-based cores to beat four of Intel’s.

The threaded nature of MainConcept makes it easy for FX-8350 to assert its strengths. FX-8150 doesn’t do all that bad in this test either, landing amongst Intel’s Ivy Bridge-based Core i5s.

The same holds true in HandBrake, where FX-8350 behaves a lot more like a Core i7 than a Core i5. AMD’s FX-8150 ends up 20 seconds behind, dropping it in between Core i5 models.
But when you consider that Ivy Bridge didn’t yield as much of a performance gain for Intel as we might have expected, the shift from Bulldozer to Piledriver actually helps AMD make up ground it had previously given up.

This probably could have gone on the Productivity page, but iTunes reminds us that single-threaded workloads are not FX-friendly.



Historically, we’ve seen AMD’s processors bottleneck the performance of certain games at low resolutions and mainstream quality settings. Using a GeForce GTX 680 at Battlefield 3’s Ultra quality preset, however, reveals no such limitation (even with anti-aliasing disabled completely).
Of course, this only applies to the single-player campaign, which tends to be GPU-heavy. The multi-player element of Battlefield 3 is more taxing on processor performance. But because it’s difficult to create a repeatable benchmark involving 63 other players, we’ll move on to another game notorious for its emphasis on CPU speed.



When AMD’s Bulldozer emerged, we were particularly disappointed that it wasn’t able to resolve the performance deficits suffered by AMD's older architectures in processor-bound games. In fact, our Skyrim tests demonstrate that the FX-8150 underperforms a Phenom II X4 980 across the board, even at 2560x1600.
The Piledriver-based FX-8350 does not completely ameliorate AMD’s gaming issues, but it does improve on the Bulldozer architecture, at least pushing past the Phenom II flagship.
I presented AMD with my findings and we came up with a couple of sources for the speed-up, which, combined, likely explain what’s happening. Most obvious is the 400 MHz-higher base clock rate, enabled through power optimizations. Second is Piledriver’s larger L1 DTLB (from 32 entries to 64), which can benefit performance when a game searches through large data structures spanning multiple pages.
At the end of the day, AMD still has work to do in improving game performance. But Piledriver certainly does help rectify the slide backward we saw Bulldozer taking relative to some of AMD’s previous quad-core parts in processor-bound games.



World of Warcraft is another title where AMD simply hasn’t done well in the past. At 1680x1050, the difference between each of our contenders is most obvious. FX-8150 manages to outmaneuver the Phenom II CPUs. However, FX-8350 improves performance even more than the Bulldozer architecture. That doesn’t stop Intel’s Ivy Bridge-based chips from outperforming, though.
Naturally, as resolution increases, the significance of graphics starts outweighing the influence of CPU speed. But even at 2560x1600 using the Ultra quality preset, it’s still clear that Vishera serves up a nice gain over Zambezi, and that both AMD packages continue to get beaten by Intel’s Sandy and Ivy Bridge-based processors.
It’s always interesting to do a full analysis of performance before taking a look at the power consumption log files I generate while the benchmarks run.

It's a little unexpected to see the Piledriver-based FX-8350 (the blue line) clearly using less power over the course of our benchmark suite than the Bulldozer-based FX-8150 (the green line), even though the previous-gen part operates at a base frequency 400 MHz lower.
Before I get into the specifics, I want to point out a small change I’m making in this review. Normally, I cut the power log off as soon as the last test finishes. This fails to account for idle power use, though, as the script is constantly starting and stopping benchmarks. Today, I’m giving each system exactly 600 seconds (10 minutes) to idle at the end of the run. As a result, average power use is brought down and energy consumption increases (due to the longer measurement period), but we do get a more realistic look at how these systems do when they’re allowed to rest.
I don’t have a chart for each machine’s idle power consumption. Looking over each log, however, tells us that the Core i5 and Core i7 idle the lowest (roughly 79 and 80 W, respectively). The Phenom II X6-based system pulls about 102 W from the wall when it’s not doing anything. And the machine with an FX-8150 draws 92 W. FX-8350 fares no better, idling at 92 W as well. But it finishes the whole suite so much faster that efficiency almost certainly improves.
Average power consumption and the time taken to finish all of our benchmarks will shed some light on Vishera’s efficiency.

Across our benchmark suite, the FX-8350-based system used 10 W less than the same machine with FX-8150, despite its higher clock rate and better performance.
We have a pretty good idea that the FX-8350 is quicker than AMD’s old Phenom II X6 1100T. But the six-core chip’s lower power consumption could translate to better efficiency if it isn’t significantly slower. That’d be a disaster for AMD.
The fact that both chips from Intel average dramatically lower power use across the run makes it almost impossible for the Piledriver-based Vishera platform to catch up.

This is something for AMD to be proud of. Its FX-8350 finishes in second place (of the CPUs I charted—I tried to pick and choose carefully to keep the graph from getting too hectic). For the record, though, I also had to know how FX-8350 did against Core i5-3570K, and it finished 12 seconds before the pricier Intel chip.
Even though it wraps up the total workload less than 10 minutes sooner than the FX-8150, the fact that AMD is charging less than $200 for its desktop flagship completely changes the processor’s value proposition. How efficient is it, though?

FX-8350 is almost 13% more efficient than its predecessor. Perhaps more important, it proves to be more efficient than Phenom II X6 1100T.
One year ago, the Phenom was a more power-friendly choice than the Bulldozer-based FX. There was no way to escape the fact that AMD had put out a CPU that used more power and performed worse in a number of key applications. Now we’re at least able to acknowledge better performance, better efficiency, and a more attractive price. Is all of that enough to garner a recommendation?
We ran her through the twisties. We did the drag racing. We measured her mileage. We even popped her hood to figure out what makes this little speedster tick. In every way, today’s FX-8350 is better than the FX-8150 that preceded it. Faster. More efficient. Cheaper. And, when you look at the fairly minor improvements Intel made to its Ivy Bridge architecture, FX-8350 competes more readily against higher-end competition.
I’m going to have to resign myself to forgetting what the FX brand once meant, though. Almost a decade ago, an FX was something that made Intel scramble to respond. It represented cutting-edge. And it set you back more than $700 bucks. Today, we have to ooh and aah over performance victories against the middle of Intel’s desktop line-up—its Core i5-3470 and -3570K—all the while shrugging off fairly severe discrepancies in energy efficiency. There. I’m done. Back to 2012.
As I was saying, with my barometer of success recalibrated, FX-8350 is a much stronger contender than FX-8150 was. It reclaims ground that AMD’s Bulldozer architecture gave up. The Piledriver architecture doesn’t cure all that afflicted Bulldozer, but subtle design and process tweaks adjust power use down, allowing the company to nudge its flagship’s clock rate up without violating a 125 W TDP. The changes aren’t dramatic, but they’re substantial enough to create a good comparison against Intel’s highest-end Core i5. So there’s that.
Of course, if AMD had excitedly recognized good progress and tried to charge the same $245 it thought FX-8150 was worth a year ago, I’d be setting FX-8350 aside as quickly as I did with last year’s model. Instead, the company is asking for less than $200. That puts the FX-8350 on par with Intel’s Core i5-3470—a multiplier-locked part that it outperforms in a great many demanding desktop apps. In those same applications, the FX is usually able to beat the $230 Core i5-3570K, too. It’s only when you look back at the single-threaded stuff that AMD continues to get creamed.
But then there’s power to consider. In the United States, we’re blessed to have relatively inexpensive energy. We tend not to flip out over 50 W unless dissipating that heat requires a noisy fan. But if you’re in Denmark paying $.40/kWh, just the 10 W difference between Core i5 and FX-8350 at idle costs you several bucks per month. Under load, you’re looking at up to a $15-a-month difference for a system running 24/7. Advantage: Intel.
So, let’s try to distill all of this down into a recommendation. Recognizing that the power user community gives AMD more latitude than Intel, I anticipate a greater number of enthusiasts getting excited about FX-8350 than any of the Bulldozer-based CPUs, and rightly so. More speed, significantly improved efficiency, and a sensible price tag are exactly what I was hoping to see, and AMD delivers them all. Are you asked to make compromises? Yeah. Single-threaded performance still isn’t impressive, and power consumption remains a sore subject. But for less than $200, I can certainly see FX-8350 at the heart of a budget-oriented workstation.
Would FX-8350 be my first choice in a new build, though? Probably not. Although I’m impressed by the work AMD’s architects have done in the last year, performance remains too workload-dependent. And, inexpensive energy aside, I’m going to go with the more efficient implementation when all else is close to equal.







