Monday, June 10, 2013

Unlocked and Overclocked - Overclocking my AMD Phenom II 550 - Part 2

Previously, I posted about my experiences overclocking my Phenom II CPU that sits at the heart of my gaming rig. In order to see what difference the tweaks made to system performance, I had to benchmark the machine before and after and I'll attempt to draw some conclusions from the results.

Performance Testing

Given the amount of time I had spent producing a stable system, I wasn't keen on spending an age producing performance figures, but I still wanted to be able to highlight any gains in performance. With this in mind, I decided to use the following benchmarks:

  • Synthetic Tests
    • Cinebench - both the single and multi-threaded CPU rendering tests.
    • POV-Ray - again, both single and multi-threaded CPU rendering tests.
    • Unigine Heaven Benchmark - Full screen, 1920x1080, 8x Anti-Aliasing, 16x Anisotropy, with textures and shading set to high/maximum, occlusion, refraction and volumetric enabled, tessellation set to normal, with a trilinear filter. Vsync was disabled to see how fast the system could push out frames, despite the screen-tearing that would occur.
  • Real-World Tests
    • ARMA2:CO - 1920x1080, with the quality preference set to "high" and vsync enabled. I profiled the E08:Benchmark scenario (provided by ARMA2:OA) performance by running FRAPS for the duration.
    • Battlefield 3 - 1920x1080 with with graphics quality set to "ultra" and vsync enabled. I played through the car park segment of the Operation Swordbreaker mission, recording performance for 60 seconds using FRAPS.
    • Crysis - 1920x1080, details set to "very high", anti-aliasing set to x16 and vsync enabled. I ran the 64bit version of the "Assualt_Harbour" benchmark provided by the Crysis Benchmarking Tool and recorded frame times with FRAPS.
    • TES V: Skyrim - 1920x1080, with graphics options set to "ultra" and vsync on. I found an outdoor location that was near a giant's encampment with a dragon circling overhead and started benchmarking before attacking the giants, recording frame times with FRAPS for 60 seconds.

The first two synthetic tests (Cinebench and POV-Ray) are primarily to see how much additional raw processing power is unlocked by tweaking the CPU. I expected to see quite linear performance increases here, as the test are primarily CPU-bound. From Unigine Heaven through all the real-world tests, I expected to see varying performance gains; each engine will rely on CPU performance to a different degree, with multiple cores making more of a difference in some and clock-speed providing more of a boost in others.

The Results

Analysis

As expected, there were some variations in results, which I'll try to provide some analysis for below:

  • The synthetic CPU benchmarks (Cinebench and POV-Ray) produced pretty unsurprising results; performance appears to increase linearly with additional CPU horsepower. For example, when comparing the dual-core and tri-core results you can see there is a around 50% improvement with the additional core.
  • Unigine Heaven was a little disappointing: the only tangible improvement was a slight increase in minimum frame rate. However, I suspect this is because it's a GPU-focused test, without any other factors to impact performance, such as AI-related calculations, etc.
  • The most impressive improvement was produced by the ARMA2:CO benchmark. In fact the game is very CPU-dependent given it's primarily a military simulation title, as opposed to your usual run-and-gun FPS affair. The unlocked and overclocked CPU provided much less deviation in the frame rate, bringing up the minimum frame rate to almost 10 FPS over the stock configuration. Surprisingly, the overclocked dual-core configuration seemed to reduce performance, which leads me to believe there's some additional optimisation required at the higher clock speed (e.g. increasing CPU-NB bandwidth).
  • Looking at Battlefield 3, there's a similar story: the 3.4GHz tri-core brought up the minimum frame rate and you can clearly see on the frame-time graph a more consistent performance was produced throughout the benchmarking run.
  • Crysis seemed to gain similar improvements with both the 3.6GHz dual-core and 3.4GHz tri-core configurations. I suspect that the older title might not benefit from the addition core and the improvements seen with the 3.4GHz tri-core are tied to the marginal increase in clock speed. This theory is backed up by the frame time graph as all three CPU configurations seem to suffer from the same dips in performance at the same time during the benchmarking run.
  • The performance improvements in Skyrim were in line with ARMA2 and Battlefield 3, with the tri-core producing the most significant benefit.
Conclusion
In line with current understanding of game engines, newer titles seem to favour additional cores over CPU frequency. Fortunately for me, I have a Phenom II were I can easily unlock the third core, which instantly gives me better frame rates in the games I regularly play. I would have liked to try and push the clock speed of both the dual and tri-core configurations a bit higher, but I was seriously limited by the thermals of the system. Improving the cooling of the CPU could give me a bit more head room for me to increase voltages, both in the CPU core and the CPU-NB. Overall, I'm pleased with the additional performance I've unlocked in my PC and I'm very happy with the amount I've learnt in the process.