Large FFT/Blend Stability

marco_polo

I don't have details to hand just yet but trying to get a stable 4.0GHz on a phenom II 965 BE(C3), up until now had been happy enough with a slight, stock voltage, multiplier only bump to 3.7GHz.

So far I can get reasonable stablity on small FTTs @4.0GHz, 1.475v (just for a couple of hours so far haven't done a longer test yet), and that seems to be about the max core limit that will be attainable with the 212 Hyper, as it is definatley not stable at 4.1GHz, 1.5v, and beyond that voltage temperatures start to go off the charts. So while I haven't got a premium piece of silicon its doesn't seem absolutely terrible either.

Anyway my problem is pretty much an instant BSOD @4.0GHz when running largeFFT or blend, whereas on 3.9GHz everything is perfectly stable for a few hours or so with the exact same settings bar the CPU bump. So it seems clear I have some sort of RAM/NB/IMC problem @4.0GHz.

A few loops of memtest seems fine as well @3.9 GHz (haven't ran it at 4.0 GHz yet as I figure a memory problem would give errors rather than an instant BSOD). - Just to add it is 4 dimms totaling 8GB, so it may be that this is the problem.

It took a while to get it stable at 3.9 as well, bumping up the NB to 2400MHz amongst other things.

Will post up the full settings later, but any ideas on the best place to start attacking the problem in the interim are most welcome (CPU-NB voltage, NB Freq, loosen memory timings etc)

marco_polo

Nevermind I bumped it up to 1.525 and large FFTS runs ok at least for a few minutes, but my cooler cannot sustain those sorts of voltages for more than about 10 miuntes without straying into the >62 degrees phenom danger zone.

I suppose large FFT must have marginally greater power consumption than small FFT which tips the chip over the stabilty cliff. So it looks like my CPU just aint that great a sample and I'll have to remain a confirmed member of the sub 4 GHz club for now. I suppose I should have suspected as much when I needed to 1.475 just to get to 3.9 GHz:)

I might attack 4.0 again from a bus speed angle in the near future, but that'll have to do for now.

Serephucus

It might be your memory controller being strained. I can't tell what voltage you should be screwing with to fix this as I'm not at all familiar with AMD systems, but as a simple test, pull out two of the DIMMs and run the tests again. If they run ok, find whatever voltage corresponds to the memory controller and bump it up a notch or two.

marco_polo

Cheers for the reply it's called the CPU NB on AMD, it was a little bump up on this that got 3.9 running stable, but more doesn't seem to help beyond that unfortunately.

Although it doesn't seem to make much sense that large FFT would need more vCore than small FFT from what I understand, it definitely seems to be the case here, as it is the only setting I have found that that gets around the insta BSOD (Stable as in for as long as I dare to run it at least, which is not very long a few tests at most). Its probably just not the greatest chip as I had to go from stock voltages (1.375) all the way up to 1.475 just to go from 3.7 to 3.9 GHz which from reading around definitely seems on the high side relative to what other can do with the c3 stepping.

Although since I haven't ran small FFT for longer than an hour or two @4.0GHz, it may well be the case that it isn't really all stable either on a longer test. I had planned to do some a longer blend tests when I had gotten small and large FFT going for at least an hour or two but hit this wall in the meantime.

Kinda done with it for now tbh but I might play around with just two dimms some evening I have nothing more productive to do. I've put alot of time for what will be a measly 100Mhz gain and of no noticeable benefit whatsoever, it is only dented pride that will motivate me at this point .