MultiGPU Setups and PCI-E Overclocking

[elitesix]

I recently read an old article (3xxx radeon series old - http://www.tomshardware.com/reviews/pci-express-2.0,1915-9.html ) concerning pci-e 4x-16x lanes and gpu performance. I wanted to run a test to see how much pci-e overclocking affected modern gpu since it stands to reason modern gpus saturate the pci-e bus more than older cards.

Specification:
Graphics: EVGA 275 SLI @ 693/1585/2510 core/shader/memory
CPU: i7 920 @ 4.1 Ghz (195x21), L3 Cache @ 3.9 Ghz, HT off
Memory: 3gb Corsair DDR3-1333@ 1950 CAS 9
MB: Asus P6T (both cards on 16x pci-e 2.0 slots)
PSU: Corsair TX850w

Test Setup:
PCI-E 100 mhz versus 109 mhz (115 mhz froze during testing)
Windows 7 Build 7100
Company of Heroes 2.601
Nvidia Driver Version 191.07 (V-sync off, Prerendered frames 0, AFR2, 16xAF, Prefer Maximum Performance, High Quality)
Game Settings: 1600x1200, 16xAF, 16xAA Quality, All In game Settings Maxed

I ran three runs of the ingame, built in benchmarking tool. It was not until the third benchmark run on both runs did harddrive thrashing cease (though the second run had very little). It is also of note that the average fps increased between the 1st to 2nd runs, and slightly increased again between the 2nd to 3rd runs, so we're not dealing with natural variation / benchmarking error.

Results (third benchmark run, displayed in average/max/minimum fps):
PCI-E 100mhz: 148.7/279/44
PCI-E 109mhz: 151.5/288/47

Percent Difference 100 versus 109:
Average FPS Increase 1.8%
Maximum FPS Increase 3.2%
Minimum FPS Increase 6.8%

Interesting results. I ask anyone with even more powerful cards (Especially ATI 58xx or 5970s) post their experiences with PCI-E overclocking and specifically, any test results.

-elitesix

EDIT: Although 115 mhz wouldn't work, I reran the test at 114mhz and got the following results average/max/min:
PCI-E 114mhz: 150.2/292/50

[jokers_greg]

not a bad improvement, i wonder how impactful pcie OCing is on FSB OCing. Most just lock it at 100, but if you can get an average FPS increase of 2% for a 9mhz OC, might as well. Have you tried 115?

[Bobnova]

In the 775 world you often have to oc the pcie in order to get the most out of your fsb.

I'd love to see this test redone with 5870s, we know they only lose 1-2% going from 16x to 8x, but not if they gain from PCIe overclocking.

[EarthDog]

Im wondering if this performance increase is the same or even exists with single cards...

[elitesix]

I think it's important to note that often a small percent difference in average fps, especially 1-2% range, is often benchmark error in a lot of games/benchmarks (though it does not seem to be in CoH in my runs). However, nearly a 7% difference in minimum fps suggests that something is actually going on.

I recommend that if anyone tests pci-e overclocking, they need to test minimum fps, because it seemed in my one test that's where the games affected by pci-e overclocking the most. And imho, minimum fps are just as important as average fps, if not moreso.

[ratbuddy]

I coulda sworn I'd read somewhere that overclocking the PCIe bus could potentially result in hard drive corruption. Was that only with PATA drives or something?

[jmdixon85]

It would still affect any drive if the disks controller was connected to the PCI-E bus. Interesting results tho. Would like to see this done with a single mid-range card? I'm not risking loosing my Win7 installation right now :)