CPU Bottleneck / SLI Scaling revisited
Benchmarkextreme dives deep into the gray areas between hardware capabilities and gaming performance
Page 1 / 18 10/03/09
INTRODUCTION
If you have visited Benchmarkextreme in the past and have read any of the articles on this site, then you can probably tell that BenchEx is unique in quite some ways. Unlike other hardware websites, the major focus of benchmarkextreme is to look at gray areas that surround modern hardware performance. Previously published articles such as CPU Bottleneck Analysis and GTX 285 TRI SLI Analysis are a testament to this fact. I try to come up with novel ideas to try to analyze and quantify typically subjective entities such as the amount a piece of hardware might bottleneck your gaming performance, or what is the sweet spot for your overclocked CPU.
Even though most of the articles on this site are quite thorough, and have been received well by the readers, there were some instances when various aspects of the analysis were left untouched, mostly due to time constraints. For example the CPU Bottleneck Analysis article does a good job of testing dual GTX 280's at various CPU settings, but missed out a lot of the resolutions that most gamers play at. The same is true for the GTX 285 TRI SLI Analysis article. It had been on my agenda to come up with a more detailed article that will address some of the issues that were left out earlier. And after several weeks of benchmarking and data collecting, I can proudly present my latest article to my readers :).
The purpose of this article is to see how gaming performance is affected at various GPU configurations (single, dual and Tri) at varying CPU speeds AND at varying resolutions. To give you an idea of how many benchmark runs I performed : (10 benchmarks) X (3 GPU Configurations) X (5 CPU Speeds) X (3 resolutions) = 450 ! Yea...that's a lot of runs. But was it worth the time and effort? I will let the readers decide that.....read on...
TEST SETUP
Motherboard EVGA X58 TRI SLI
CPU I7 920 @ 203 QPI
RAM 4 GB DDR3 Patriot Viper @ 1620 8-8-8-8-20
GPU (3) X EVGA GTX 285 SLI (Forceware 193.8)
Monitor Gateway 30" XHD
OS Vista X64
BENCHMARKS USED
The following benchmarking software were utilized for this article:
1. Crysis Single player V 1.1.1
2. Cryostasis Tech Demo
3. Far Cry 2 (Ranch Small)
4. Tom Clancy's H.A.W.X
5. Lost Planet : Extreme Conditions Demo
6. PT Boats : Knights of the Sea
7. Stalker : Clear Sky Benchmark
8. The Last Remnant Benchmark
9. Warhammer : Dawn of War Demo (Fraps)
10. World in Conflict Demo
Besides Warhammer, all the benchmarks featured some sort of built in time demo that I used for benchmarking purposes.
A WORD ABOUT THE ANALYSIS
The analysis was run at three different GPU settings :
1) Single GTX 285
2) GTX 285 SLI
3) GTX 285 TRI SLI
The GPU clocks were kept at their stock 648 Core / 1476 Shader / 1242 Memory
The CPU QPI was fixed at 203 and the various multipliers used were 21 (4.26 GHZ), 20 (4.06 GHZ), 18 (3.65 GHZ), 16 (3.25 GHZ), and 13 (2.64 GHZ). The stock frequency of this CPU is 2.66 GHZ. So I basically went from stock, all the way up to 4.26 GHZ. By changing the multipliers, and leaving everything else the same, we ensured that the various hardware variables, such as ram speeds and Uncore frequency were kept constant and the only thing that varied was the CPU frequency.
Where ever possible, I provided both the average and minimum frame rates. However, some of the built in benchmarks did not return the minimum benchmarks and obviously I did not report the minimum benchmarks for such benchmarks. Another thing to keep in mind is that the results that I am going to display here are purely the raw FPS , and as such not an exact measure of your gaming experience. For example 100 FPS is no better than 70 FPS in terms of gameplay, since most LCDs have a refresh rate of 60 HZ yet it will be reported as a "gain" in FPS. So when I use the word "gain" in this article it will usually refer to FPS gains. In some instances, however, true gameplay gains will be experienced and I will point those out. Another thing I want to mention is that the scales on the graphs were adjusted, depending on the frame rates, to give a better idea of changes in FPS. Lastly, I have introduced a "minimum frequency recommendation system" for this article. This basically is a recommendation from Benchex to run your CPU at a certain minimum frequency for a desired combination of Game, Resolution and GPU. I am sure the readers will find this useful. However, I must stress that this recommendation is based on my personal experience with benchmarking and analysis of the games and should only be taken as a rough guideline at best.
Now, before we move on to the results, I would like to mention that what you are going to see is arguably the most extensive CPU bottleneck / SLI scaling analysis on the internet, and as such the information provided might be overwhelming for new readers. The way in which I presented some of the results, especially SLI scaling, is unique to my knowledge as I haven't seen that being used on any other hardware site.I am sure this will be a treat for those of you who are the number crunching, analytical type, yet others might find it tedious and overdone.....OK, enough talk...let's move on to the results now :)
| NEXT |
|---|