Cpu Bottleneck and Multi GPU Scaling with I7 920 Page 1 / 10 06/26/10
Introduction and Test Setup
Multi GPU scaling and the effect of CPU frequency on 3D performance is an interesting area for analysis. I have done various articles on this site that have discussed these topics in detail. One of my favorite articles was I7 920 bottleneck analysis where I did a thorough investigation with GTX 285 single, dual and tri SLI and how various CPU speeds affected gaming performance at different resolutions. Today's article is somewhat similar to that article but will feature a lot powerful hardware. Basically, this article will present :
- How GTX 480 scales in Dual and TRI SLI when gaming at 2560 X 1600 resolution, and how this scaling is affected by different CPU clock speeds.
- How GTX 480 scaling compares with ATI 5970 crossfire scaling at the same resolution, image settings and CPU clock speeds.
It might be worthwhile to note that I recently performed an analysis on 5970 corssfire scaling. In essence, this article can be considered a follow up to the original 5970 scaling analysis article, with the addition of tests run at various CPU clock speeds and also comparative numbers from GTX 480.
The following setup was utilized for this article:
I7 920 @ 2.67 - 4.3 GHZ (only multiplier was changed)
3 X 2 GB Patriot Viper 1620 MHZ 7-8-8-20
EVGA X58 Motherboard
Gigabyte ATI Radeon 5970 + Sapphire ATI Radeon 5970 (crossfire - Catalyst 10.4)
3 X PNY GTX 480 (Forceware 257.21)
150 GB 10,000 RPM Velociraptor
Dell 3008 wfp 30" LCD
Antec 1200 Watt PSU
Windows 7 64 bit
Software Used
I used a mix of DX9, DX10 and DX11 titles for this article :
World in Conflict Demo
Crysis Warhead 1.5
Far Cry 2
Just Cause 2
Battleforge Demo
Stalker Call of Pripyat Benchmark
METRO 2033
Dirt 2
FEAR 2
Lost Planet : Extreme Conditions Benchmark
Resident Evil 5 Benchmark
Batman Arkham Asylum
Warhammer : Dawn of War II Demo
Please note, that some of the tested games are demo versions and as such may not be representative of the performance and scaling of the final version of the game. Another thing worth noting is that to keep this analysis simple and short, I am only presenting the average frame rates. Typically, it is a good idea to report the minimum FPS when doing a CPU bottleneck analysis, but because of time constraints I chose to ignore that in this case. I also want to point out, that my intention here is to investigate raw scaling (both GPU and CPU). The scaling may or may not affect the gameplay, but that is not what this article will discuss. We are just investigating the combined theoretical performance gains by changing CPU clock speeds and varying the GPU horsepower.
Some Clarification
I have heard reports from around the internet that people are questioning the legitimacy of this article and this website in general. To be honest, I really don't care what people say because I have nothing to lose and no one to please. But I do feel that it is my duty to defend my integrity. One of these days I will add an "About us" page like all the other fancy hardware sites so readers don't mistake me for a money making, corporate funded entity.
This website is not a corporation and is not funded by a corporation and is free from all sorts of bias. Benchmarkextreme is purely owned by me and is the sole work of a PC performance analyst (I do not like using the word "enthusiast") who does this work in his free time. I built this web site about a year and a half ago, and during this time I have only written 14 articles (less than an article a month) -- A testament to the fact that this is something I do outside of my work and academic life. Every piece of hardware ever reviewed or analysed on this site has been in my possession, and was bought from an etailer / retailer, only to be sold off after being tested. The data for the 5970s appearing in this article was gathered at a time when 10.4 was the latest catalyst version. The video cards were then sold off, and I acquired the GTX 480s for the comparative test. This is the reason why this article uses CCC 10.4 and not 10.6.
If any one has an issue with this then be it. My job is to let the readers know that my work is 100% unbiased and I have no sympathy, affiliation or preference for any one brand over the other. In most cases, I do not even recommend the hardware that I test, and do not even mention the prices. Because in my opninion, most of what I test is overpriced to start off with, and not worth recommending at all. This is not about changing people's minds about hardware and brands. This is about analysing raw, unadeltrated performance that can be extracted from today's advanced computer hardware; and in my book the most extreme performer is the perfromance king, regardless of the brand. I just report my numbers, and how people react based on my numbers is their decision, and I have no interest in that whatsoever. OK, enough talk...now off to some extreme results. Next page please...
| NEXT |
|---|