Our story back in May, Is This Even Fair? Budget Ivy Bridge Takes On Core 2 Duo and Quad, put a spotlight on the per-clock performance and efficiency improvements made to Intel's architectures over the last five years. We were smacked in the face by the fact that the company's old Core 2 processors, tuned up through overclocking, were often able to fly past the newer budget-friendly Ivy Bridge-based parts.
Once you dip down under the Core i5 family, you're dealing with dual-core CPUs limited by locked multiplier ratios. Factor in a stubborn base clock that requires an enthusiast-oriented chipset to adjust, and we're left with an inability to coax extra performance from the Celeron, Pentium, and Core i3 families.
Thankfully, AMD isn't following suit. The company continues catering to frugal enthusiasts with numerous quad-core Black Edition and K-series processors sporting unlocked multipliers and a more receptive HyperTransport reference clock. Naturally, we felt compelled to explore AMD's offerings in the same way, hand-picking several options that were priced attractively.
Our oldest selection, the Phenom II, first appeared in 2009. Rather than use a four-year-old review sample, we went out and bought a Phenom II X4 965 Black Edition to get a more accurate snapshot of the chip's current overclocking headroom. This once-mighty flagship packs four cores running at 3.4 GHz and a hefty 6 MB of shared L3 cache.
Later in 2009, AMD released the affordable Propus-based Athlon II X4 processors, which were basically Phenom IIs without the on-die L3. We dug out an Athlon II X4 640 test sample running at 3 GHz to represent this configuration. Although the chip's multiplier is locked, we can still overclock it by manipulating the reference clock.
We knew we'd need a modern Vishera-based processor to pit against our former favorites, and AMD sent over FX-4350 and FX-6350 samples. The former sports a pair and the latter comes equipped with a trio of Piledriver-based modules, which in turn pack two integer cores each. Being a budget-minded experiment, the six-core FX is the fastest CPU we're testing.
Lastly, we noticed that the Athlon X4 750K finally reached North America. Although we couldn't get a sample from AMD, we went ahead and bought one to benchmark. This two-module Trinity-based processor lacks integrated graphics and represents the most affordable multiplier-unlocked quad-core chip available.
By comparing AMD’s K10 and Piledriver architectures, stock and overclocked, we hope to see where each excels, hopefully determining the CPUs most deserving of your hard-earned buck.
Before diving head-first into the performance data, let’s cover the platforms we used to test and the overclocks those components achieved.
Socket AM3+: Gigabyte GA-990FXA-UD3
As a budget-oriented piece emphasizing overclocking, we wanted affordable hardware with enthusiast-friendly features. Gigabyte's GA-990FXA-UD3 Socket AM3+ motherboard fit the bill for testing our FX, Phenom II, and Athlon II processors.
Based on the AMD 990FX/SB950 chipset, the board packs a ton of connectivity, including six SATA 6Gb/s ports, USB 2.0, USB 3.0, eSATA, and FireWire. We're able to appreciate the company's Ultra Durable 4 technology suite, 8+2 phase power design, solid capacitors, and dual BIOS chips in case something goes wrong while we're overclocking.
Socket FM2: Gigabyte GA-F2A85X-UP4
Next, we chose the feature-rich GA-F2A85X-UP4 for testing our Athlon X4 750K. Based on AMD’s A85X Fusion Controller Hub, this enthusiast-oriented Socket FM2 board is built for APU tweaking. It sports Gigabyte’s Ultra Durable 5 technology, again, solid capacitors, and a beefy power stage rated at 60 A.
A trio of PCI Express x16 slots accommodate up to two graphics cards at x8 and x8 transfer rates. The platform also supports Dual Graphics, though we wouldn't recommend going that route. Four DDR3 memory slots support dual-channel configurations at up to 2400 MT/s. Here, again, you get ample connectivity via six USB 3.0 ports, seven SATA 6Gb/s connectors, and one eSATA 6 Gb/s interface. Other notable features include dual BIOSes, Lucid Virtu Universal MVP GPU virtualization, and 3x USB Power with On/Off Charge technology.
Overclocking
Given the flexibility of these CPUs, I took a slightly different approach to overclocking than I would in an Intel-focused story. In my prior piece, I knew the chips in question offered quite a bit of headroom, so I picked frequencies that'd be easily attainable for most folks. Capping the clock rates low helped represent commonly-used settings.
In contrast, AMD's processors ship at clock rates that leave little headroom with their boxed heat sinks and fans. So, I used the same Xigmatek setup that took Intel's Wolfdale architecture to 4.5 GHz. It made sense in this value-oriented experiment to keep aftermarket cooling in the $20 to $30 dollar range. Then I doubled up on component airflow by adding a second 120 mm fan to avoid VRM-related throttling. Although I chose not to push crazy voltages, I did tweak the CPU-NB frequencies, and squeezed everything I could (within reason) from each setup, while respecting the thermals.
Die-hard enthusiasts from both camps might be crying foul, wondering why I pushed AMD's chips harder or shied away from larger air coolers. The bottom line is that we're evaluating the performance of inexpensive CPUs and stacking them up relative to each other. I suspect that the AMD crowd is going to push its chips just as hard, and with good air circulation and a respectable heat sink, you shouldn't have any trouble achieving results like mine.





| System Test Configurations | |
|---|---|
| CPU | AMD FX-6350 (Vishera) 3.9 GHz (19.5 * 200 MHz), Socket AM3+, 8 MB Shared L3, Turbo Core enabled, Power-savings enabled, Overclocked to 4.52 GHz (22.5 * 200.92), 1.404 V load, 2612 MHz HT Link, 2210 MHz CPU-NB Frequency |
| AMD FX-4350 (Vishera) 4.2 GHz (21 * 200 MHz), Socket AM3+, 8 MB Shared L3, Turbo Core enabled, Power-savings enabled, Overclocked to 4.72 GHz (23.5 * 200.92), 1.440 V Load, 2612 MHz HT Link, 2411 MHz CPU-NB Frequency | |
| AMD Phenom II X4 965 Black Edition (Deneb) 3.4 GHz (17 * 200 MHz), Socket AM3, 6 MB Shared L3, Power-savings enabled, Overclocked to 4.02 GHz (20 * 200.91), 1.392 V Load, 2008 MHz HT Link, 2411 MHz CPU-NB Frequency | |
| AMD Athlon X4 750K (Trinity) 3.4 GHz (17 * 200 MHz), Socket FM2, No L3 Cache, Turbo Core enabled, Power-savings enabled, Overclocked to 4.30 GHz (43 * 100), 1.464 V Load, 2000 MHz CPU-NB Frequency | |
| AMD Athlon X4 640 (Propus) 3.0 GHz (15 * 200 MHz), Socket AM3, No L3 cache, Power-savings enabled, Overclocked to 3.6 GHz (15 * 240), 1.428 V Load, 1920 MHz HT Link, 2400 MHz CPU-NB Frequency | |
| CPU Cooler | Xigmatek HDT-S1283 120 mm air cooler |
| Motherboards | Gigabyte GA-990FXA-UD3, Socket AM3+, AMD 990FX, BIOS FC (02-05-13) |
| Gigabyte GA-F2A85X-UP4, AMD A85X, BIOS F4 (03-13-13) | |
| RAM | 8 GB (2 x 4 GB) Crucial PC3-12800 kit Stock: DDR3-1600, CL 8-8-8-24 at 1.5 V Overclocked: Athlon II/Phenom II: DDR3-1600/1607, CL 8-8-8-24, Athlon X4 750K/FX-6350/FX-4350: DDR3-1866/1875 9-9-9-24 @ 1.6 V |
| Common | |
| Graphics | Sapphire Radeon HD 7970, 950 MHz GPU, GDDR5-5700 |
| System Drive | Samsung 840 Pro 256 GB, SATA 6Gb/s SSD |
| Power | Corsair Professional Series HX1050, 1050 W, 80 PLUS Silver |
| Software and Drivers | |
| Operating System | Windows 8 Professional x64 |
| Graphics Driver | AMD Catalyst 13.2 Beta 7 |
| Benchmark Configuration | |
|---|---|
| 3D Games | |
| Borderlands 2 | Version 1.0.28.69606, DirectX 9, Custom Run, Fraps Test Set 1: Medium Quality Settings, Low PhysX, 8x AF Test Set 2: Highest Quality Settings, Low PhysX, FXAA,16x AF |
| Crysis 3 | Version 1.0.0.2000, Direct X 11, Custom Run, 60-Sec. Fraps Test Set 1: Lowest Quality Settings, No AA, 1X AF Test Set 2: Medium Quality Preset, FXAA, 8x AF Test Set 3: Very High Quality Preset, 2x SMAA, 16x AF |
| The Elder Scrolls V: Skyrim | Version 1.8.151.0.7, Custom Run, 25-Sec. Fraps Test Set 1: High Preset, No AA, 8x AF, FXAA Enabled Test Set 2: Ultra Preset, 8x AA, 16x AF, FXAA Enabled |
| F1 2012 | Version 1.2, Direct X 11, Built-in Benchmark Test Set 1: High Quality, No AA Test Set 2: Ultra Quality, 8x AA |
| Far Cry 3 | V. 1.04, DirectX 11, 50-sec. FRAPS "Amanaki Outpost" Test Set 1: High Quality, No AA, Standard ATC., SSAO Test Set 2: Ultra Quality, 2x MSAA, Enhanced ATC, HDAO |
| Hitman: Absolution | V. 1.0.446.0, DirectX 11, Built-in Benchmark Test Set 1: Medium Quality Preset, No MSAA, 2x AF Test Set 2: Ultra Quality Preset, 2x MSAA, 16x AF |
| StarCraft II: Heart of the Swarm | Version 2.0.9.26147, Custom Run "Harvest of Screams" Campaign Mission, 60-Sec. Fraps Test Set 1: High Preset, No AA, 8x AF, FXAA Enabled Test Set 2: Ultra Preset, 8x AA, 16x AF, FXAA Enabled |
| Tomb Raider | Version 1.00.722.3, Direct X 11, Custom Runs, "Chasm Monastery", "Mountain Village", 45-Sec. Fraps Test Set 1: High Quality Preset Test Set 2: Ultimate Quality Preset |
| Audio/Video Encoding | |
| HandBrake CLI | Version: 0.98, Video: Video from Canon Eos 7D (1920x1080, 25 frames) 1 Minutes 22 Seconds, Audio: PCM-S16, 48,000 Hz, Two-Channel, to Video: AVC1 Audio: AAC (High Profile) |
| iTunes | Version 10.4.1.10 x64: Audio CD (Terminator II SE), 53 minutes, default AAC format |
| LAME MP3 | Version 3.98.3: Audio CD "Terminator II SE", 53 min, convert WAV to MP3 audio format, Command: -b 160 --nores (160 Kb/s) |
| TotalCode Studio 2.5 | Version: 2.5.0.10677, MPEG-2 to H.264, MainConcept H.264/AVC Codec, 28 sec HDTV 1920x1080 (MPEG2), Audio:MPEG2 (44.1 kHz, Two-Channel, 16-Bit, 224 Kb/s) Codec: H.264 Pro, Mode: PAL 50i (25 FPS), Profile: H.264 BD HDMV |
| Abobe Creative Suite | |
| Adobe After Effects CS6 | Version 11.0.0.378 x64:Create Video, Three Streams, 210 Frames, Render Multiple Frames Simultaneously |
| Adobe Photoshop CS6 | Version 13 x64: Filter 15.7 MB TIF Image: Radial Blur, Shape Blur, Median, Polar Coordinates |
| Adobe Premiere Pro CS6 | Version 6.0.0.0, 6.61 GB MXF Project to H.264 to H.264 Blu-ray, Output 1920x1080, Maximum Quality |
| Adobe Acrobat X Pro | Version 10.0.0.396: Print PDF from 115 Page PowerPoint, 128-bit RC4 Encyption |
| Productivity | |
| ABBYY FineReader | Version 10.0.102.95: Read PDF save to Doc, Source: Political Economy (J. Broadhurst 1842) 111 Pages |
| Autodesk 3ds Max 2012 | Version 14.0 x64: Space Flyby Mentalray, 248 Frames, 1440x1080 |
| Blender | Version 2.64a, Cycles Engine, Syntax blender -b thg.blend -f 1, 1920x1080, 8x Anti-Aliasing, Render THG.blend frame 1 |
| Compression | |
| 7-Zip | Version 9.28, LZMA2, Syntax "a -t7z -r -m0=LZMA2 -mx=5" Benchmark: THG-Workload-2012 (1.3 GB) |
| WinRAR | Version 4.2, RAR, Syntax "winrar a -r -m3" Benchmark: THG-Workload-2012 (1.3 GB) |
| WinZip | Version 17.0 Pro, Syntax "-a -ez -p -r" Benchmark: THG-Workload-2012 (1.3 GB) |
| Synthetic Benchmarks and Settings | |
| 3DMark 11 | Version: 1.0.1, Performance Suite |
| PCMark 7 | Version: 1.0.4, System, Productivity, Hard Disk Drive benchmarks |
| SiSoftware Sandra 2013 | Version: 2013.01.19.11, Processor Arithmetic, Cryptography, Memory Bandwidth Benchmarks |
We’re going to shoot through the performance tests in the same order as last time, starting with 3DMark 11, and once again focusing on the Physics and Combined modules.
For the most, part these processors place in our chart according to their cost. And while overclocking helps somewhat, only the Athlon X4-750K is able to ascend the hierarchy. The Athlon II X4 640 trails behind, seemingly suffering from its lack of L3 cache. Even overclocked to 3.6 GHz with a 2400 MHz CPU-NB frequency, it gets smoked by the stock 3.4 GHz Phenom II X4 965 Black Edition.

Our overclocking efforts yield impressive gains in SiSoftware Sandra's Arithmetic component. Sporting three active modules, the FX-6350 sits alone and unchallenged at the top.
Check out the difference in the architectures, though. Piledriver shines in the integer-heavy test, since each module features two cores, but trails in Whetstone due to a shared floating-point unit.

This chart may appear unsorted, but we're ranking each configuration by the average of its three scores.
A look at hashing performance puts the Athlon X4 750K slightly above the Phenom II X4 965 Black Edition. The older K10 architecture lacks the hardware-acceleration of AES supported by Vishera and Trinity.

We used DDR3-1600 at CAS 8 timings for all five processors, making apparent the improvements AMD made to its memory controller that improve bandwidth.
Overclocking the RAM and CPU-NB frequencies further augment memory throughput, though we don't expect serious gains in our real-world benchmark workloads.
You can't tell from the chart, but we measured these gains incrementally, so we know that the Athlon X4 750K gains as much through CPU-NB overclocking as it does from the jump to DDR3-1866. In contrast, the Athlon II X4 640 doesn't gain anything from a CPU-NB boost of 400 MHz.

Our audio and video encoding suite immediately highlights the strengths and weaknesses of AMD’s past and present architectures. For starters, the single-threaded iTunes and LAME workloads expose lackluster per-clock performance across the board. No surprise there, though.


That all changes as we switch to more optimized titles like HandBrake and TotalCode Studio. The sub-$100 quad-core CPUs offer big value, even before we consider overclocking.


Benefiting from a third Piledriver module, and able to execute 50% more threads simultaneously, the FX-6350 lays down impressive performance relative to the rest of the field.
Perhaps more surprising is the strength of Deneb, despite its age. Overclocked to 4 GHz, the Phenom II X4 breezes past the FX-4350 and overclocked Athlon X4 750K in all four A/V-oriented metrics.
AMD’s FX-6350 dominates these tests, so long as the software effectively utilizes its six integer cores.

The Athlon X4 750K is also impressive, right up until it falls flat in Premiere Pro. Overclocking and tweaking helps tremendously, allowing the inexpensive CPU to match the stock FX-4350. We thought it might be starved for memory bandwidth in stock form, but later debunked this theory by dropping the memory data rate to 1600 MT/s with a 4.3 GHz overclock. This only added three seconds back to the timer.


The quad- and hexa-core chips fare well in After Effects. Unlike what we saw in iTunes and LAME, where AMD's newer CPUs give up superiority in per-clock measurements, the Piledriver architecture does better in our single-threaded Acrobat workload.

AMD’s aging K10 architecture is strong throughout our threaded productivity-oriented applications. The 3.6 GHz Athlon II matches the 4.7 GHz FX-4350 in 3ds Max 2012, and even beats it in Blender. The six-core FX-6350 zips past both four-core Piledriver-based chips, but barely holds off an overclocked Phenom II X4. We imagine it'd likely fall to a Phenom II X6.



The FX-6350 shines in 7-Zip, evidence that this app fully utilizes all six integer cores. The Phenom II X4 holds its own in WinZip, but trails in WinRAR and 7-Zip.
Finally, the lack of L3 cache appears to be a major weakness for the Propus-based Athlon II in WinRAR. Overclocking by 20% yields a 10% improvement. Even then, though, the lower-clocked Phenom II still finishes 21% quicker.



Moving onto 3D gaming, we start with our Tom’s Hardware Borderlands 2 manual run, which takes approximately 105 seconds to complete.

Again, we have to caution against putting too much emphasis on Fraps' minimum frame rate data, since the same consistency issue we experienced in the Intel story happens here as well. I’ll stay consistent and chart unadjusted minimum FPS here also, though this may be the last time. While this variance is usually insignificant, it can also be game-changing. For example, Fraps reports that the overclocked FX-4350 drops to 35 FPS, yet the single lowest second logged is 57 FPS.

Each processor delivers a nice playable experience at Medium settings, though a couple of those dramatic drops at the beginning of the run (gaming on the Athlons) did translate to noticeable stutter.

While there is evidence of a CPU bottleneck preventing the complete utilization of our beefy Radeon HD 7970 graphics card, every one of these processors is at least capable of delivering an acceptable Borderlands 2 experience. Only the stock Athlon X4 750K and overclocked Athlon II X4 640 register a brief drop below 40 FPS.

We use a brutal Crysis 3 sequence to punish each processor. Consequently, we'll step down and also include the lowest-quality settings.


The FX-6350’s complete and total dominance proves that Crysis 3 scales well beyond four threads. While there are a couple dips below 30 FPS, each quad-core CPU remains playable at these entry-level settings.


Upped to medium-quality detail settings, my subjective impression of smoothness doesn't completely match the benchmark data. Each quad-core processor spends a few seconds under 30 FPS. But the main problem goes beyond that information. Through the test sequence, not one platform feels completely fluid. Average frame rates are reported higher than they actually feel. I’d call the Athlon II X4 640 unacceptable until you overclock it, though the others are at least playable. Without question, stock and overclocked, the FX-6350 is a pleasure to game on by comparison.


At Very High quality with 2x SMAA enabled, our platform starts suffering from a graphics bottleneck. Yet, scaling amongst these processors is still evident. While the FX-6350 takes a huge performance hit, it's still noticeably smoother than the quad-core models. Basically, an AMD FX-6300 or Intel Core i5 is my minimum CPU recommendation for enjoying Crysis 3. This hard-hitting title isn’t one for low budgets, and it prefers an FX-8350 or Core i7 if you have one.
F1 2012 is one of Codemasters’ popular DirectX 11 racing titles based on its own Ego Engine 2.


Each processor remains playable through the Ultra detail preset, so our F1 2012 analysis becomes all about architectural scaling. In general, Piledriver does well compared to K10, and even the Athlon X4 750K manages to outpace the quad-core Phenom II at 3.4 GHz. I don’t doubt that some of this is a result of increased memory bandwidth, as F1 2012 is one of those rare titles where tweaking the RAM and CPU-NB frequencies can yield almost as much extra performance as processor overclocking.
As an example, running at 4.7 GHz, the FX-4350 averages 101.3 FPS (a gain of 4.5 FPS). But our tweaked overclock manages 105.5 FPS at the same 4.7 GHz.


Next up is another well-threaded DirectX 11 title, Ubisoft’s Far Cry 3, based on the Dunia Engine 2.


The Athlon II is the only CPU to dip under 40 FPS. Everything else delivers acceptable performance at the High quality detail preset.
While AMD's Phenom II outpaces the company's Athlon X4 750K at 3.4 GHz, overclocking puts them on even ground. Also notable is that this is the fourth title in a row where an overclocked FX-6350 tops the charts.


Far Cry 3 has become one of my favorite titles for evaluating gaming platforms. In general, I find that configurations that stay above 30-32 FPS in our test sequence meet my demands through the rest of the campaign. The extra 15 or so minutes I spent playing on the Athlon II X4 640-based config were enough to know that, at Ultra quality, this processor needs to be overclocked. Like many of the dual-core processors I’ve evaluated, those dips under 30 FPS are significant enough that they detract from the game, necessitating a drop to the Very High preset.
I wouldn’t call anything less than a quad-core Phenom II smooth, though the overclocked Athlon II X4 640 and stock Athlon X4 750K also receive a passing grade. As I mentioned in the Intel equivalent of this piece, Ultra details at 1920x1080 require a strong graphics card, which you'll want to complement with a similarly-capable platform. If money is tight, though, an overclocked Athlon X4 750K could be the ticket on a strict budget.
IO Interactive’s Hitman: Absolution is a Direct X 11 title based on the Glacier 2 graphics engine. We’re using the built-in benchmark routine for a worst-case look at how the massive crowd technology punishes our crop of processors. Keep in mind that frame rates are generally much higher, since most of the game isn't as wide-open as the crowed Chinatown map area.


We know that dual-core processors struggle in this benchmark, and the FX-6350’s finish up top suggests that scaling continues beyond four threads. Despite a frequency advantage, the overclocked Propus-based chip can’t match Deneb at its stock settings, so we also have to assume the Phenom II’s L3 cache helps performance in this title. The stock Phenom II X4 965 BE performs well, going toe-to-toe with the overclocked Athlon X4 750K and almost matching AMD's FX-4350.


While the lasting appeal of Blizzard’s popular StarCraft franchise is no doubt found within the multiplayer experience, the single-player campaigns are also a lot of fun, making them a worthy starting point for testing. Rather than using our existing Wings of Liberty map, I found that the “Harvest of Screams” mission from the Heart of the Swarm expansion really taxed my Core i5-based gaming rig. Our 60-second benchmark takes place as Kerrigan leads approximately 150 Zerg units in to destroy the mission’s final Protoss base.
I purposely delayed my attack a couple of minutes to build up more Zergling than the mission required. Plus, I kept the game camera zoomed out and centered over the action. As a result, frame rates drop substantially as more units come into view and join the battle. This may be more brutal than your own play style. After all, a Core 2 Duo E6600 is Blizzard's recommended processor. But without question, too little compute power will force you to rethink your strategy, zoom the camera in to involve fewer units, or avoid large-scale maps altogether.


StarCraft II is CPU-intensive, but unfortunately isn’t optimized for quad-core processors. Operating at the lowest clock frequencies, AMD's Athlon II X4 640 struggles the most. Its frame rates are pinned in the mid-teens once all of the units join the battle. A lack of L3 cache must be hurting it as well. Despite a 200 MHz overclocked advantage over the stock Phenom II, it still trails by about 12%.


Cranking up graphics and texture quality for this second set of numbers gets us closer to how we'd play the game with a fast GPU under the hood. Even so, our results remain CPU-limited. Overclocking is pretty much imperative with the Athlon II, since its stock-frequency frame rates hovered between 12 and 16 throughout the battle. Even then, it's barely adequate at 3.6 GHz, stuck below 20 FPS during intense action.
The Athlon X4 750K is a nice step up, competing with the Phenom II and matching AMD's FX processors once it gets overclocked. For some reason, the FX-4350 suffers a strange performance drop as the fog of war lifts. This didn't seem to be a result of textures loading, and it was repeatable on subsequent runs. More important is that, during intense action, the FX-4350 tops our chart with the highest frame rates.
The Elder Scrolls V: Skyrim remains a popular title thanks to the modding community and numerous downloadable expansion packs. While it’s not necessarily a demanding title, it remains a very relevant benchmark. I'll often pick a fight within the City of Markarth to take a worst-case look at playability, but the Tom’s Hardware’s standard 25-second run through Riverwood is almost as demanding and far easier to repeat amongst editors.


Despite an official quad-core CPU recommendation, Skyrim doesn’t effectively utilize more than two cores. Both Athlon processors are out-classed by the Phenom II and FX, although the Athlon X4 750K recovers once it's overclocked.
Clear scaling is apparent in our testing, but all of these processors deliver playable performance.


Although it's still playable using the Ultra detail setting, AMD's Athlon II X4 640 looks weak, and I don’t doubt that there are user-created mods that'd bring it to its knees. In fact, the 640 is the only CPU in our two-part series that drops under 40 FPS. Worse, the chip even falls under that mark when we overclock it. This has to be attributable to L3 cache, since the Propus architecture is completely outclassed by Deneb. When we chart out frame rate over time, we see that the 3.6 GHz Athlon II only peaks above the Phenom II's minimum at the very end of our test sequence.
Tomb Raider is one of 2013's biggest hits, in my opinion. It takes powerful graphics hardware to handle the Ultimate quality settings, which enable realistic TressFX hair.
We already know the Radeon HD 7970 has what it takes to deliver playable frame rates at that Ultimate preset, and we'll include the graphics-heavy “Chasm Monastery” level you typically see on Tom’s Hardware. But we're putting far more weight on the CPU-intensive outdoor “Mountain Village” level. Used together, these two benchmarks provide a worst-case look at the game’s CPU and GPU requirements.


Normal hair effects are used with the High quality preset. That flat area in the middle of our line graphs, where all the processors appear at the same level, is the cinematic sequence. The Athlon II X4 640 trails early on, but breezes through this part of the game, delivering 100+ frames per second.


As expected, frame rates plummet once we step outdoors and overlook “Mountain Village”. A lack of L3 cache appears to be the Athlon II X4 640’s weakness yet again, though it remains playable, briefly dipping below 30 FPS.


Overclocking yields small, insignificant gains, mainly because this test is almost exclusively GPU-bound. TressFX hair enabled by the Ultimate quality preset completely changes the flat cinematic portion of our run, and the mighty Radeon HD 7970 drops to 30 FPS, no matter which host processor backs it. Once the camera zooms off of Lara, we see a huge frame rate spike before control is shifted back to the user. Similar cinematic sequences are unavoidable, and a big part of the game.


Without a doubt, it takes powerful graphics hardware to drive the Ultimate detail preset smoothly. But parts of this game smack the processor around, also. In this test, the Athlon II X4 640 fails, making it difficult to control Lara’s maneuvers. In fact, the lack of consistency caused me to scrap a couple of benchmark runs after misjudging my zip-line approach and blindly leaping straight off the cliff. Of our test samples, only the two FX-series chips remain above 30 frames per second through our 45-second run.
Power-saving features are enabled for each processor at its stock settings, but get disabled to help with higher and more stable overclocks. We chart peak input watts during active idle, full Prime95 load, and throughout our gaming tests.
Our bench rig is way overpowered with Corsair's 80 PLUS Silver-rated HX1050. A quality 450 W power supply could have been sufficient, though overclocking the Radeon HD 7970 would have put more stress on the whole platform.

The Trinity-based Athlon X4 750K tops our chart with the three lowest power consumption measurements. On average, the older quad-core Phenom II consumes the most in stock form. Of course, our overclocking efforts required voltage bumps, resulting in higher power use under load. The jump at idle is more a result of disabling power-saving features.
We’ll summarize our performance testing relative to the least expensive processor in today’s round-up. Of course, what matters more than the averages is how each CPU performs in the workloads most important to you.
Sporting a quartet of physical cores, the Athlon II X4 640 often dominates its price bracket in productivity-oriented metrics and overall measures of performance. However, that advantage disappears in today’s comparison, as all five of these CPUs juggle at least four simultaneous threads.

There is no other way to put it: the Athlon II X4 640 is simply outclassed. Worse, perhaps, is that, even cranked up to 3.6 GHz with a 2.4 GHz CPU-NB frequency, the Propus-based processor is not able to drive our games sufficiently well. We know that a lack of L3 cache stifled performance. After all, the stock 3.4 GHz Phenom II X4 fared significantly better.
The good news is that the enthusiast-friendly Athlon X4 750K and Phenom II X4 965 Black Edition both get the job done for just a bit more money AMD's FX-4350 performs even better. And most impressive of all is the FX-6350 with its six integer cores, topping the charts in seven out of eight games and completely dominating in Crysis 3.

While the Athlon II X4 640 came up short in our battery of compression utilities, where memory bandwidth likely affected performance, its quad-core architecture proved more capable in productivity applications and Adobe's Premiere Pro than the twin-module composition of AMD's Athlon X4 750K. Plenty of shared L3 cache helps the Phenom II X4 catch the newer Piledriver design in WinRAR, WinZip, and 7-Zip, while further improving our productivity benchmark scores.
Once again though, the FX-6350 shines brightest. Single-threaded tests like iTunes, LAME, and Adobe Acrobat cut into its overall dominance, but those two extra cores are put to good use in most of the other applications.
Today's story focused on AMD's processors, but we used the same tests, graphics hardware, and drivers that previously went into our exploration of Intel's architectures. The only difference was a forced update to StarCraft II: Heart of the Swarm that couldn't be avoided. Thus, it seems pertinent to create a summary of the data we generated.
In an effort to be as fair as possible, I’m switching up the overall game performance calculations a bit on this page and the preceding one. Rather than adding up the average FPS result for each game, I'm weighing the relative performance in each title equally, as I did with the applications. I’m also tossing out the two GPU-limited Tomb Raider tests and only using that title's outdoor sequence. Consequently, you should notice a larger spread between processors.

AMD’s chips shuffle into the hierarchy, just as we’d expect. Each one outpaces Intel's Core 2 Duo E8400, but none of them can challenge the Ivy Bridge-based Core i5 for its top spot. And keep in mind that many retail -3570Ks hit 4.4 GHz or more with air cooling, which would extend its lead.
The FX-6350's competitiveness at stock clock rates is impressive. It scales well with overclocking, too. And considering that we could probably squeeze close to the same performance from a less expensive FX-6300, we have to call out the competence of Vishera in its six-core configuration. Based on averages, the FX-6350 is a step above Intel’s Core i3-3225 and an overclocked FX-4350. In newer, well-threaded games, it's unlikely that a tweaked Phenom II X4 or Core 2 Quad could touch the FX-6350 at its stock settings.
If you're working with limited funds, AMD's Phenom II X4 965 Black Edition and Athlon X4 750K are both capable processors that sell for $100 or less. Unlocked multipliers mean that both CPUs are overclockable. The money you save buying one of these cheaper chips can be put toward graphics, aftermarket cooling, or even a new game.
Unfortunately, not all of AMD's processors deliver the performance we are looking for. In a threaded gaming suite favoring quad-core CPUs, a lack of L3 cache prevents the Propus architecture from matching the cheaper Pentium G2020's value. The Athlon II led in three out of eight games, and even then was limited to low-quality settings in a couple of them. The one benefit this quad-core chip offered was playable frame rates in Crysis 3 at the game's lowest detail settings. We had to overclock our Core 2 Duo E8400 all the way to 4.5 GHz to get similar performance, and even then its average frame rate was 10% lower.
But we can't praise the Athlon II for its behavior in Crysis without criticizing the slide show we saw in StarCraft II: Heart of the Swarm. In that tile, along with Skyrim, AMD's Propus design overclocked to 3.6 GHz couldn't catch the cheap 2.6 GHz Celeron G1610, much less a faster Pentium. Based on these eight games, I'd recommend skipping the Athlon II X4 640 for gaming builds. If cost keeps you from snagging an Athlon X4 750K or better, then a cheaper and generally quicker Pentium G2120 gets you more value.
If your computing needs have less to do with games and involve productivity-oriented apps, AMD has the upper hand in budget-friendly performance. Pentium and Celeron processors are fine for lightly-threaded workloads. But the Athlon X4 750K and Athlon II X4 640 are far more capable when it comes to multitasking.
The aging Phenom II design doesn't offer the per-clock performance of Intel's architectures, but its four physical cores blow past the dual-core competition in many of our benchmarks.

Finally, we come back to the Vishera-based FX with three Piledriver modules. The overclocked -6350 finishes several tasks faster than our Core i5-3570K control processor. Overall, it ends up a few percentage points behind. AMD's FX-6300/6350 simply crushes the similarly-priced Core i3s in our benchmarks, and AMD’s unlocked multipliers make the beat-down even more severe.

And then things get a little ugly. Although it leads AMD's other processors, the Athlon X4 750K consumes more power than even Intel's quad-core Core i5. Insofar as efficiency is concerned, AMD is still a couple of generations behind.
For the enthusiasts who enjoy playing with their PCs as much as on them, it's a lot more affordable to overclock an AMD platform. This is in sharp contrast to Intel's Sandy and Ivy Bridge architectures, which the company artificially locks up in order to upsell you on its K-series parts. At least in the mainstream space, AMD's only real weakness is gratuitous power consumption, which hurts efficiency, results in excess heat, and may lead to a noisier machine. I had the luxury of a quiet aftermarket cooler for my experiments, but in general, AMD's boxed coolers are both noisy and inadequate for accommodating higher voltages.
Unlocked multipliers are always appreciated. But we recognize that they mean very little if a CPU is already running close to its maximum clock rate. Thankfully, the processors we chose to test today had some room to scale up, giving us reasonably fast quad-core configurations under $100 and a hexa-core model at $120. An overclocked Athlon X4 750K is generally quicker than a Core i3-3220, and in workloads able to run across six threads, the FX 6300-series stomps Intel's similarly-priced dual-core chip.
What if you're a gamer, first and foremost? Tom's Hardware's editorial team believes that, moving forward, you're best off with a processor able to address four threads simultaneously. But as we saw today, not just any quad-core chip will do. Running at 3.0 GHz, and devoid of L3 cache, the Athlon II X640 is overwhelmed by some of our tests. Overall, it lags behind Intel's less expensive Pentium in measures of average frames per second. Overclocking helps, but the Propus architecture is still unable to match any of the other AMD chips we tested. The last reason to avoid the Athlon II, in my opinion, is AMD's own Trinity-based Athlon X4 750K. It’s the most affordable processor we benchmarked to survive all eight games (although it needed to be overclocked to really make Starcraft II enjoyable). You'll probably need aftermarket cooling and a voltage bump to make it past 4 GHz, but I think a slightly overclocked Athlon X4 750K has great potential in an affordable gaming rig.
AMD's Socket AM3/AM3+-based platforms offer plenty of gaming performance, too. As enthusiasts, we're always looking forward to the next upgrade. But at least for now, if you already own a Phenom II X4, your CPU is still good enough to plow through most modern titles. As with the Core 2 Quad based on Intel's Yorkfield design, you shouldn't need to upgrade any time soon if your overclocked Phenom II is still running well. It's only unfortunate that boxed Phenom II X4 Black Edition processors are so hard to find for $100.
Moving up the list, AMD's dual-module FX-4350 is typically able to outperform the company's older Phenom II X4. Overclocked, it destroys Ivy Bridge-based Core i3s in threaded productivity and content creation applications. It's also able to compete aggressively in games. But positioned between the FX-6300 and -6350, the $130 FX-4350 is a hard sell.
The shining star in today’s comparison is AMD’s FX-6350, which delivers solid performance in games, while besting Intel's Core i5 in a number of our other benchmark workloads. The cheaper FX-6300 is an even more attractive bargain, so long as you're willing to overclock it.


