Need Help with Prime95 Stability

[wingman99]

It just Failed again with everything on stock settings.
8 Hours 9 minutes it said (Prime95 Application has stopped working)
A problem cause the program to stop working correctly. Windows will close the program and notify you if a solution is available.

It stopped on the first test 2100 Lucas-Lehmer iterations of M34230145 using FFT length 1792K

Windows didn't notify me!
Should I run the new Prime95 V28.9 or just RMA my CPU?

Run the new Prime95. I was looking at the version you were running and it is old and has bugs for the updated HT P4 and plenty of new updates for the new CPU's. I did not look through all the updates on the list V25.6 is from 2007.

 

[oobymach]

Are you doing folding or seti or some other cloud computing where your cpu is going to be stressed continually for over 8 hours at a time? I do 2 tests with p95 or about 13mins of testing to make sure it passes the second hurdle and doesn't freeze, if so it's good to go for gaming and that's why I stop usually around 15mins in.

OCCT is another good stress testing environment however it is more likely to fail quickly than p95 if your oc is too high. I generally fail after about 1min:30sec of occt and my cpu is fine for long periods of gaming and video encoding, no major glitches in anything.

 

[RJARRRPCGP]

Linpack is the quickest for core testing...

 

[H2mike]

I used Linpack on Maximum and it passes 20 loops just fine.

I am going to use this for gaming, but mainly for heavy video editing of 4k footage. It will run at 100% for several hours a day, and I can't afford for it to be locking up during some complicated rendering.

I'm running the new Prime95 Blend right now and will have the results in the morning! Maybe with some luck, it only fails the old Prime95 V25.6 Blend and the CPU is ok!

 

[mackerel]

It does not mater what calculations are run first. The small FFT calculations fit in the L1 L2 L3 cache not much ram needed. Small FFTs run fast because of the faster cache speed and smaller calculations. The large FFT calculations run L1 L2 L3 some memory. Blend switches between both calculation sets every 3 minutes with system memory 2GB.

It really does not matter what part of the CPU is failing with large number calculations or small number calculations. The failure could be any one CPU transistor out of billions in any Cores with L1, L2, also the L3 IMC memory controller could be the problem.
I have seen prime95 fail or the OS running it. It makes no difference where the CPU Error comes from and from what I explained how can a person tell where the CPU failure came from, with running software.

I just built a skylake 3 months ago and it would have infrequent freezing problem every 1-3 weeks and it passed all stress testing software. Then I RMA'ed the CPU no more problems.

The point of small vs. large FFT is exactly about fitting in CPU cache or not. Failing small FFT is indicative of unstable CPU itself. Failing large FFT while passing small points towards the ram system. I had funnies with my 1st 6700k build. It would appear to pass short runs (hours), but in real work I'd get detected errors once every few days. Long story short, it was the ram. BIOS updates and settings improved it to an average of one error per month, but that dropped to zero only when I swapped the ram.

 

[H2mike]

New Prime95 V28.9 Build 2 Just passed Blend 14 hours, unbelievable!!
I'm thinking now that I don't have a problem!
I would have never believed an old version of Prime95 would have been the problem.
Now I'll get overclocking and keep my fingers crossed.
Thanks so much for the help everyone, I would not have figured this one out on my own.

 

[wingman99]

I'm glad it passed the V28.9 Prime95.

 

[H2mike]

Prime95 V28.9 build 2 works great!
My CPU is stable now. The older Prime95 I was running was causing all the problems.

I now have it overclocked and running great thanks to everyones help.
6850K 4200Mhz 1.3 volts max temp on Prime95 for 12 Hours of the hottest core was 89C
DDR 3200 running at 3200 16-18-18-36 2T

The new Prime95 does run hot!!!
I can get to 4300Mhz but the CPU turns into a space heater at 1.36 v
These temps are with a very good EKWB Performance 280 water cooler.

 

[wingman99]

Prime95 V28.9 build 2 works great!
My CPU is stable now. The older Prime95 I was running was causing all the problems.

I now have it overclocked and running great thanks to everyones help.
6850K 4200Mhz 1.3 volts max temp on Prime95 for 12 Hours of the hottest core was 89C
DDR 3200 running at 3200 16-18-18-36 2T

The new Prime95 does run hot!!!
I can get to 4300Mhz but the CPU turns into a space heater at 1.36 v
These temps are with a very good EKWB Performance 280 water cooler.

The newer Pirme95 V28.9 uses FMA3 (FMA) FMA3 instruction set for floating point scalar and (SIMD) Single instruction, multiple data operations, FMA3 instructions have three operands.

If your not going to use math calculations or scientific math calculations also distributed computing projects you can disable FMA3 also AVX2 with this command CpuSupportsAVX=0 in Pime95 local notepad. The instructions will run slower through the CPU, in turn a cooler CPU.

 

[mackerel]

http://www.mersenneforum.org/showthread.php?t=21691

Beware there's a problem with 28.9 and disabling instructions as detailed above. I guess it wasn't found until recently as no one actually did this...