Does OCing interfere with GPU performance

[Anymeans]

I have been OCing an E6600 to 3-3.2Ghz with no major dramas. I have 2G Corsair C5 RAM, GA965P-DS3 (rev3.3) mobo, and ASUS GTS 8800 640MB video card. C1E, EIST disabled, RGB at fast, memory ratio at 2, setting 9x333 or 8x400, voltage from 1.25-1.3375 depending.

OC is successful both on CPUz but also via 3DMark 06 where CPY scores significantly improve. However, often (but not always) I am seeing significant fall offs in video card performance when OCing. For example in the first 3DMark06 test in the first few seconds when the ship is landing on stock I would be getting 60-65 fps. When the problem occurs the fps can fall to 20-25!

This then continues throughout the test so that 3D scores for SM2 and SM3 are in the 2000's rather than >3600 normally.

Does anyone have any ideas what could be the cause and solution? Do I just give up CPU ocing as fps will generally be more impt for games?

Thanks

 

[Oerekum]

I'm not sure if it happens with that motherboard, but check if your PCI-e setting is at 16× and not 1×. I have the P5B deluxe (same chipset) and that happens to me if I do not bump the NB voltage enough.

 

[lavaheadache]

Your power supply could be struggling to provide adequate juice. What make/model are you using?

 

[swtethan]

did you lock your pci @ 100mhz? pci sync @ 33mhz also

 

[Anymeans]

Hi thanks for your replies.

PSU = Antec Trio 550 W

Not sure about PCIE thing... only setting is auto or running from about 90 to 105 or something from memory. All the guides Ive seen suggest just running at 100 which it is locked to.

Cant find any setting in BIOS around PCI sync (for setting at 33).

Another board suggested bumping MCh up by 0.1V... (is this the northbridge??) .. have done this but will need to test over a few days to see whether consistent performance. Any other ideas of what I could play with?

Thanks for your help.

 

[50cent1228]

Originally posted by: lavaheadache
Your power supply could be struggling to provide adequate juice. What make/model are you using?

thiss is the first thing i thought of ...it could be this!

 

[Rinaun]

Originally posted by: swtethan
did you lock your pci @ 100mhz? pci sync @ 33mhz also

This is most likely the problem. Set it to 33.33Mhz and set the frequency from auto to 106.

 

[Anymeans]

Sorry I wasn't clear. It PCIe is currently set to 100. Cannot find anything that adjusts PCI sync in my BIOS.

Getting varying opinions on increasing PCIE above 100...can anyone explain when should have on 100 and when above?

Thanks

 

[Thor86]

3DMock is not a stability testing suite. Maybe you should check and see if your system is stable then run the 3D benching testing.

 

[tersome]

Originally posted by: lavaheadache
Your power supply could be struggling to provide adequate juice. What make/model are you using?

The nvidia control panel pops up a warning if your GPU is underpowered.

It sounds like you're facing the 1x PCIE problem. There doesn't seem to be a single solution, but try setting your NB volts almost to the max, setting PCIE bus to 101mhz, making sure the link width is 16x and not 1x, etc.

 

[Agentbolt]

The link width is fudged, it has to be. Bumping your NB voltage might work, but what you really need to do is lock in your PCI-E frequency. It could need to be anywhere from 100-120 MHZ (mine needs to be locked to 120 MHZ for the link width to go back to 16). You will have to guess and keep checking until you find one that fixes it.

Use CPU-Z to check the link width under "mainboard" It will be 1. Setting the PCI-E frequency in your BIOS will fix this, I promise.

 

[Anymeans]

Thanks for comments guys.

Thor86... appreciate 3D is not a stability testing platform... been using Orthos for this and at 3G OC have not had CPU problems. My concern is the GPU results which I understand the std OC tests don't look to.

Tersome/Agentbolt... as indicated the PCIE setting is locked at 100. Have 'discovered' there are alot more sections to CPU-z than just the CPU section lol once you mentioned it Agent. The PCIE width is indicated as 16x currently... but this is with having increased MCH 0.1V... will reset computer to old settings and see whether I can reporoduce problem and look at this setting.

Thanks everyone.

 

[Anymeans]

OK guys... having now seen that I can check the PCIE link on CPU-Z I can confirm that this is the problem. After resetting Vcore to 1.25 and MCHto normal the probelm reappeared and I noted the PCIE link went to 1x (great when you can see where the problem is.. thanks guys). Tested this on 3D and problem manifested in framerate.

Then played with MCH volts, CPU volts and PCIE freq.

Results:

+0.1V MCH... link went to 16x
102 PCIE freq... link went to 16x
1.3V CPU (up 0.05) link stayed at 1x

So having two apparent solutions to the problem any advice on which is best for longevity of mobo or other elements (does the 0.1V turnup in a temp reading somewhere I should be looking at?)/stability etc.

Also I guess any advice on which is best to adjust if I increase CPU OC?

Really appreciate the help you have all given me.