Low performance with 8800 gtx SLI

[Alex843]

I just recently went SLI and I’m getting kind of poor performance it seems.

My specs are
E4300 @ 3.21
4 GB G.Skill DDR 800 Memory
Evga 680i Motherboard
2x 8800 gtx at stock
Windows XP Pro 32

Latest Nvidia drivers
And Crysis is updated to 1.2.

Doing the basic CPU benchmark I’m getting these numbers in 1920 by 1200
Avg 23.21 Min 13.78 Max 33

http://www.tomshardware.com/reviews/system-builder-marathon,1797-11.html

They’re getting Avg 38.2. Their system has a quad core at 3.0 ghz with a lot more cache than mine, but at a lower frequency. They also have vista. Those two things are pretty much the only things that separate my system from theirs. That’s a 40% difference. I’m running everything High, no vysnc, no aa, no af. As far as I can tell everything is identical settings wise. According to my NVIDIA control panel SLI is on. Can you guys help figure out this discrepancy?

 

[royalcrown]

they are also running a quad ciore with 12 megs of cache and not an e4300 !
The cache makes a huge difference ib gaming, and you just cannot replace 2 cores by running yours 200 mhz faster.

 

[Alex843]

they are also running a quad ciore with 12 megs of cache and not an e4300 !
The cache makes a huge difference ib gaming, and you just cannot replace 2 cores by running yours 200 mhz faster.

http://www.legitreviews.com/article/583/9/

Its obvious this game is GPU limited, the difference at 1024 by 768 on low settings between this quad core and a AMD 6400+ is 25%, not even close to the 40% I'm seeing here.
The only game I've ever seen that gets a significant boost in fps from adding 2 cores is Supreme Commander, this game apparently doesn't benefit.

 

[Alex843]

I just ran the benchmark with SLI turned off. I actully got a slight boots in frame rate. Now I'm averaging 24.28 Fps instead of 23.2.

 

[ovaltineplease]

what driver version?

Anyways, you should be getting higher performance than that, cause I was benching higher than that on "High" with 16xAF on 8800 GT SLI - but my processor is clocked quite a bit higher than yours (4.05ghz), there is a much higher cpu overhead when you're using SLI cards and Crysis has a fairly high cpu overhead to begin with; and when I clocked my CPU from 3.0ghz to 3.6ghz to 4.05 ghz using 8800 GT in SLI, there was a noticeable performance increase with every frequency jump. That said, a quad core of similar frequency to your dual core is undoubtedly going to have better performance for 2 reasons: 1, helper threads for physics/audio; 2, background applications/services can run on the extra cores - which reduces the overall cpu overhead.

I can't tell you that this is without a doubt the issue you are having, but its just within my suspicion that it could be. I don't think that a quad core is a requirement for an SLI setup, but if you are using a C2D then you will need a pretty high clocked C2D to make up the difference.

 

[royalcrown]

they are also running a quad ciore with 12 megs of cache and not an e4300 !
The cache makes a huge difference ib gaming, and you just cannot replace 2 cores by running yours 200 mhz faster.

http://www.legitreviews.com/article/583/9/

Its obvious this game is GPU limited, the difference at 1024 by 768 on low settings between this quad core and a AMD 6400+ is 25%, not even close to the 60% I'm seeing here.
The only game I've ever seen that gets a significant boost in fps from adding 2 cores is Supreme Commander, this game apparently doesn't benefit.

sorry, but I can get the same frame rate on the system in my sig and have on Very high at 1024 x768 and just under that at 1024 x 1028, so it's not you slI I don't think, you need more cache to get those framerates from that review, and 200 mhz isn't gonna change your cache or replace the 2 cores. Even crysis benefits a little from quad, I am not saying it is a beast on a quad and slow on a dual though, just that quads help.

 

[Alex843]

what driver version?

175.19. The latest I'm pretty sure.

 

[Alex843]

I'm not trying to argue with you guys when you say the qx9650 is a better cpu for gaming. I'm just saying there shouldn't be a 40% difference at this high of settings, you don't see sites reviewing cpu's with games maxed out. To test cpu performance they usually run at the lowest or close to lowest setting to eliminate a gpu bottleneck. At 1920 by 1200 there is going to be a gpu bottle neck in crysis, not saying the cpu doesn't matter at that point I’m just saying it shouldn’t make that drastic of a difference especially when this article here says the game doesn't fully utilize 4 cores. If you look up benchmarks it should support this theory.

http://blogs.pcworld.com/gameon/archives/005860.html

Then there is the cache difference. 1MB vs 3MB per core. While I didn't find that exact comparison I did find this

http://www.tomshardware.com/reviews/cache-size-matter,1709-5.html

512kb vs 2MB is 10% at the most in the games listed there. 1 vs 3mb should be even less of a difference.

For some reason SLI is working like it should in this game for me. I'm going to try a see if I can bench some other games to see if its actually doing anything at all right now.

Thanks for your reply's.

 

[SirCrono]

Try killing every unneccesary process before running Crysis (Antivirus, etc, remember to disconnect the internets before killing the antivirus) and see if there is a performance gain thanks to the extra cycles.

Also, a good idea would be to underclock your CPU to something like 2.8 GHz and seeing what kind of performance penalty you get from that, if it's about the same (percentage wise) as the underclock then you'll know you're cpu limited.

 

[vnderachiever]

Try reverting to nvidia driver 175.16

 

[Alex843]

I got it working much better now.

Well I tried under clocking like SirCrono said and it didn't have any real affect on performance. So then I tried looking for 175.16 to download, and I came across some new beta drivers "177.66" off of guru3d. Installed those and that did the trick I'm now getting 39.93 for an average.

Thanks your help everyone!