Prime95 Questions

[Frown66]

Got some questions here, since Prime95 errors out after 4+ hours, figured I'd get some advice before making a change and waiting 4+ hours to see if it worked.

Setup:
Venice 3200+ on DFI nforce 4 ultra d (latest bios)
xp-90 w/ AS5 + Panaflo M1A fan
OCZ modstream 450 (26A on 12v, voltages don't jump all over the place on smartguardian during Prime95, so I doubt that power will be an issue)
TwinMos TMII400 (I was unlucky, they aren't the good kind, but still rated at 2.5 3 3 8 at ddr400)

I've got the chip to 2.7ghz (270x10) with voltages at 1.5x110%. HTT at 3x. Actual smartguardian and cpuid reading in XP is 1.61v. Memory is set at 150 (202mhz, but cpuid indicates 193mhz). Memory is set to auto, so its just 2.5 3 3 8. Smartguardian reports idle cpu temps at 28C, with load at 42C. I'm not sure if this is accurate as it seems kind of low and maybe too good to be true. Either way, I've touched the xp-90 and its just slightly warm under load. I doubt it could be heat.

I've run prime95 for over 4 hours before it stops on 1 error. No freezes or lockups. I'm confident (or am I fooling myself?) that I can get it stable at 2.7 as I feel I am merely inches away from a stable oc. If prime takes that long to error out, is it usually due to memory timings, cpu voltage or something else? I'm thinking the memory, and I'll be swapping it with some OCZ soon, but I'm not sure what to change first as letting prime run for hours is rather annoying.

Also, what would be better performance-wise, 270x10 or 300x9 (mem at 133, for an even 200mhz on ram)

 

[exar333]

hmm, what voltages do you have for your chipset and RAM? maybe try inching those up a tad...this board dosnt seem to mind a little higher than stock volatge...also, try upping the drive strength a bit for the RAM as well as playing with the TRAS (of the top of my head, i think this is the name of it....)...just try a bunch of combos, see what works (usually around 2000-3500 seems to be good to ppl, depending on RAM type)

 

[Frown66]

: hmm, what voltages do you have for your chipset and RAM?

All stock. Since RAM isn't getting overclocked, I didn't feel a need to touch the voltages there. How much should I increase the chipset voltage?

At this point, any change can take hours to see if it works. Hoping to get one of those, 'oh yea, had something similar, this is what I did.'

 

[Thor86]

I'm going to guess on your power supply. 26A on the 12v line is pretty low. Get a 520 Powerstream with 33A on the 12v line. Any minor flux, and it usually results on Prime erroring out after an hour+.

I'm assuming you are using Large FTT/max heat tests?

 

[Duvie]

I recommend you test the small FFT or Large FFT only and avoid using the blended test...Sometimes I have noticed huge ram drains as it is doing it resulting in possible page file and errors....I would run small or large FFT for 8 hours minimum...The run memtest for the ram for 10-20 passes.....This has worked for me and if it passes I have never got an error....


Also that is really incorrect on the vdimm (not to raise) since you are not ocing the ram....It still may have an effect on overall system power (drain etc.) I would say bump it up a notch to 2.75....PPL dont realize that the memory built into the cpu has very strange effects. Sometimes raising mem speed at same clock speed of cpu results in more vcore being needed so I can see why it may go both ways here....Also as you climb in ocing A64 chips you start to lose the top end of your memory speeds.....I can run 600ddr cas3 with 1t stable at 7x300 but cannot do it at 8x300 same timings...no matter the vcore or vdimm....Heck I cannot even run 9x280 1:1 for 560 ddr at any cas timing and 1t or 2t.....It just starts fading and you are shite out of luck or need to drop the memory divider to something lower. I notice as the pcu speed increase you gain bandwidth so the drop in timings doesn't effect the overall bandwidth being reported by as much as one would think....

EDIT: By the way this system can do 2.7ghz at 9x300 with 133 divider cas 2.5-4-4-10 with 1.61-1.62v...So 2.1ghz and 2.4ghz is not much of a stress...It was just to show how the ram will really start to limit you...

You need to try all sorts of things....

 

[Frown66]

I recommend you test the small FFT or Large FFT only and avoid using the blended test...

Ok, semi-dumb question. What are FFT tests and how do they differ (compared to the blended tests as well)?

I'll try to increase the ram voltage as well. Also, Duvie, do you any comments about increasing the chipset voltage (yea or nay)?

I'm assuming you are using Large FTT/max heat tests?

I'm using the blended test. I'll try different tests tonight.

 

[Frown66]

Another related question, does the motherboard bios and overall system stability differ on these settings (if so, how)?

CPU voltage: 1.55v
CPU voltage: 1.4v x 110%

Both result within .1v of each other, but would one be more stable generally?

 

[Duvie]

Originally posted by: Frown66

I recommend you test the small FFT or Large FFT only and avoid using the blended test...

Ok, semi-dumb question. What are FFT tests and how do they differ (compared to the blended tests as well)?

I'll try to increase the ram voltage as well. Also, Duvie, do you any comments about increasing the chipset voltage (yea or nay)?

I'm assuming you are using Large FTT/max heat tests?

I'm using the blended test. I'll try different tests tonight.

When you check torture test and it ask what kind of test it tealls you the differences...Bottom line the blended test test the ram along with cpu but is not necessarily as stessful to the cpu in terms of power, heat and cache testing.....I think it is better to isolate the memory with memory related test...ONe thing is like i stated the newer test with memory will take 912mb of my ram eventhough when i get into windows I dont have that much left. Some sites have reported that this caused temp paging of the HDD resulting in errors (mainkly due to speed)...PPL if they reduced the amount of available ram to use to say 256-512 (old prime test used like 48k or something small like that at default)....i would stick with programs better suited for testing the ram....


If you have optiopns to raise chipset vcore you should if you are raising the HTT..that is still stressful for the chipset as well...

 

[Dadofamunky]

Originally posted by: Duvie

When you check torture test and it ask what kind of test it tealls you the differences...Bottom line the blended test test the ram along with cpu but is not necessarily as stessful to the cpu in terms of power, heat and cache testing.....I think it is better to isolate the memory with memory related test...ONe thing is like i stated the newer test with memory will take 912mb of my ram eventhough when i get into windows I dont have that much left. Some sites have reported that this caused temp paging of the HDD resulting in errors (mainkly due to speed)...PPL if they reduced the amount of available ram to use to say 256-512 (old prime test used like 48k or something small like that at default)....i would stick with programs better suited for testing the ram....

He's right, I ran into the same thing with page swapping.... although I had 2 GB, so I have no idea why, but the disk was getting thrashed....

 

[Frown66]

I recommend you test the small FFT or Large FFT only and avoid using the blended test...Sometimes I have noticed huge ram drains as it is doing it resulting in possible page file and errors....I would run small or large FFT for 8 hours minimum...The run memtest for the ram for 10-20 passes.....This has worked for me and if it passes I have never got an error....

Ok, so I was able to run over 10 passes of memtest86 with no errors. I increased the chipset and LDT voltage by .1. I also turned off cpu throttling (recommended elsewhere, not sure what this does). I ran small FFT with no errors for roughly 8 hours. I'll run large FFT tonight. If both tests and memtest are good, is it safe to conclude the system is stable? Anything else I should test?

 

[Snowice]

Originally posted by: Frown66

Ok, so I was able to run over 10 passes of memtest86 with no errors. I increased the chipset and LDT voltage by .1. I also turned off cpu throttling (recommended elsewhere, not sure what this does). I ran small FFT with no errors for roughly 8 hours. I'll run large FFT tonight. If both tests and memtest are good, is it safe to conclude the system is stable? Anything else I should test?

cpu thermal throttling is for controlling your cpu fan speed. your cpu fan will run at full speed if it's turned off so that your fan won't stop to freak you out

since you have a dfi board, i recommend you to stop by here to take a look.

good luck!

 

[Thor86]

Originally posted by: Frown66
Ok, so I was able to run over 10 passes of memtest86 with no errors. I increased the chipset and LDT voltage by .1. I also turned off cpu throttling (recommended elsewhere, not sure what this does). I ran small FFT with no errors for roughly 8 hours. I'll run large FFT tonight. If both tests and memtest are good, is it safe to conclude the system is stable? Anything else I should test?

10 full passes of memtest is a starter. You really should use Tests 5 and 8 for at least 100+ passes each to make sure your memory controller/memory is error-free. Does this mean Windows will be stable? Absolutely not. I've had the OS crash on me even after memtest error free tests.

 

[Concillian]

Originally posted by: Frown66

If prime takes that long to error out, is it usually due to memory timings, cpu voltage or something else?

Prime is testing things on statistics, it tests, then retests. You may pass (literally) a million times, but then fail. Barring thermal influence, the fail has an equal chance of happening on the third iteration.

But on blend, prime failures can be due to anything. Very early errors are usually RAM issues, but several hour failures can be due to anything.