Hi all, I’m having some issues with my overclock;
I have a Q9550, XFX 780i board, 2x2Gb Dominator 1066 and I am trying to stabilise 3.83Ghz on air with a TRUE (450x8.5, RAM 1:1 at 900 5-5-5-15). The chip is prime95 small FFTs 'stable' at 3.83 - it did a 9h15m run no errors with 1.3685v vcore in BIOS (1.352v actual - I have pencil-modded the board for vdroop so there is no droop.)
The rest of the voltage settings for that run were all the 'auto' settings and the actual values are as follows:
Summising from that run that the CPU is quite happy at 3.83 I then went about trying to stabilise the prime blend test and ran into a whole world of pain! Fairly extensive testing has brought me to another set of values which are nearly there:
The above values ran for 2h 3m then then all cores simultaneously error-ed. This was an improvement over the previous run at 1.3815v (1.360v) which went for around 1h 5m before generating a core error. The test that generated the simultaneous errors after 2hours was 6500x 640k FFT length. I re-ran a custom test with 640k length and it passed straight afterwards so it’s not related to that particular test.
I've read since that if the length is bigger than 512k this indicates a RAM/Northbridge fail - although this was from an article that's a few years old now – presumably relating to chips with smaller cache so I don't know if that still holds true for my CPU?
However as all cores failed together and it didn't freeze/bsod I'm guessing this could point towards RAM-N/B? My RAM is running within spec though at the correct voltage, so technically it should be fine. I really don't want to push any more volts through them; I'm thinking 2.15v would be my absolute max. That only leaves the N/B. There is one more increment I can go to on the NB which maxes it; 1.55v. So anyway, I am a little stuck here at the moment.
However, my biggest issue so far seems that from a cold boot, the settings that ran for one or two hours the previous evening will fail within 10mins, and it will be three or four re-boots before the same settings get close to the times they did the night before! This is obviously causing me problems because I think I'm getting somewhere with testing, start again the next day and I appear to be back to square one again! I have tried upping the vcore, the N/B to 1.55, the VTT to 1.4, but nothing seems to make any difference. Once its warm the old settings work fine again! As I type I’m just nearing the first hour of prime blend with the settings as above, the same settings that locked up three times when I first turned the machine on this evening! Anyone else run into this cold boot phenomenon?
A quick update on this - I was playing around a bit more last night, getting nowhere once more after about 2 weeks of testing on and off, basically going around in circles! I looked back through my log of testing and realised that I'd ended up with quite high values on all the chipset voltages, but had not tried to lower these values once I'd got to the new higher Vcore. So as I was having no luck going up anymore I went back to the auto voltages from the small FFTs run. They froze on blend within a minute, so I upped N/B to 1.45, that froze within a minute, dropped it back to 1.4, upped the S/B to 1.55 and it ran for 2h 18m before simulataneous core errors - nearly equalling my 2h 24m best run using the much higher voltages. I ran the test again from a cold boot this morning and it ran for a half hour no errors before I stopped it. So it would seem I was overvolting the chipsets quite a bit and this was causing the strange cold booting issue. So hopefully I will get there soon with a couple of minor tweaks!
The thing about blend tests is that they stress the RAM while large and small ffts are mostly just CPU stress. So if only blend are getting errors then it's something to do with the RAM or how the RAM connects to the CPU, IE if it goes through the NB or SB - I'm not sure how it's connected in your system.
Another thing is it can depend what options you have enabled for power saving. Try disabling all power saving features (speed stepping/EIST, C States etc), as well as spread spectrum if you have that option.
I have all the spread spectrums disabled, and all the power features turned off. As I say I think my mian issue was that during initial testing I didn't have the Vcore as high as it needed to be and I pushed all the chipset voltages up much higher instead. After going back to the Auto volatges and just tweaking the S/B I have better stability than before as it doesn't crash on a cold boot anymore, and with 0.1v less on the northbridge and southbridge and 0.2v less on the HT link.
I did also try raising the VTT up to 1.4v which is the max on my board, but this didn't help in any of the testing I've done so far. So I'm leaving that at 1.3v at the moment as 1.2v tends to generate core errors. (The XFX 780i only allows for VTT adjustment in 0.1v increments for some reason!)