Prime95 stress test BSOD instantly but not in other test.

[_Hustler_One_]

Hi guys..

I always use Realbench and Aida64 to stress test my CPU they mostly completed the stress test to show me that my system is stable.

 

Lately I installed Prime95, when I 1st time run stress test on it, instant BSOD Whea_Uncorrectable_Errors, 2nd try same, 3rd try same even I disable my CPU OC and turn off the turbo boost. Stock 3.7GHz still BSOD only with Prime95, with Realbench and Aida64 not even single BSOD.

 

Could someone point me out what happen here?

[xg32]

if there's absolutely nothing else that crashes ur pc, then just stop running prime

[Streetguru]

6 minutes ago, _Hustler_One_ said:

 

There's really no reason to run Prime95, Cinebench or Blender are pretty much the hardest stress tests you'd want to run

[_Hustler_One_]

13 minutes ago, xg32 said:

if there's absolutely nothing else that crashes ur pc, then just stop running prime

Which one between Prime95 vs Aida64 result is more reliable to trust?

[xg32]

Just now, _Hustler_One_ said:

Which one between Prime95 vs Aida64 result is more reliable to trust?

prime is just simple integars but spammed, aida is closer to real world load, i've had weird bugs on prime before on different builds,  but no way to be sure whats happening here, if prime is the only thing ur pc doesnt run, i wouldnt worry about it.

[_Hustler_One_]

11 minutes ago, Streetguru said:

There's really no reason to run Prime95, Cinebench or Blender are pretty much the hardest stress tests you'd want to run

Oh well. I also use Cinebench R20, it never give me BSODs..

Which one between Cinebench vs Blender is more reliable to trust the result?

[scottyseng]

I usually use Intel Burn Test (Maximum level at 10 passes) myself and haven't really tried prime95 too much. It does get to the point where you're just pushing over extreme loads.

[_Hustler_One_]

12 minutes ago, scottyseng said:

I usually use Intel Burn Test myself and haven't really tried prime95 too much. It does get to the point where you're just pushing over extreme loads.

Not even in the 1st minute my PC cant passed the test. Run Small FFTs, just for the 1st few seconds my PC went down with "Whea_Uncorrectable_Errors" BSOD every time I tried to run the test.

[Streetguru]

5 minutes ago, _Hustler_One_ said:

Oh well. I also use Cinebench R20, it never give me BSODs..

Which one between Cinebench vs Blender is more reliable to trust the result?

They're both basically the same, you don't need to run any power virus/burn in tests to test an overclock. What's stable for daily use or gaming won't be stable in those unrealistic workloads

[scottyseng]

2 minutes ago, _Hustler_One_ said:

Not even in the 1st minute my PC cant passed the test. Run Small FFTs, just for the 1st few seconds my PC went down with "Whea_Uncorrectable_Errors" BSOD every time I tried to run the test.

Well, you need to up the voltage for these extreme tests (Intel Burn Test as well). They push the CPU to crazy loads. I haven't had any issues with the builds that I've done that have survived Intel Burn Test. But yeah, if you're just gaming, I wouldn't try any crazy tests.

[_Hustler_One_]

Then I think I'll just get rid the Prime95.

 

Thanks guys..

[Bitter]

If it crashes just doing math work in Prime95 then it's not stable. If it crashes running the stress tests then you're still not stable but closer to being stable. Are you running, smallest, small, blend, or which set? if it's crashing on small then it's a CPU problem, if it's crashing on blend or large then it's more likely a RAM issue. Temperature can also cause instability if you're on the edge of stability, and Prime95 small fft gets things very very hot compared to any other CPU stability test I've ever used. I always go to Prime95 first and make it stable there then run other tools like realbench to verify stability.

How are your temps? Have you tried to increase voltage to the CPU or adjusting the RAM speed, voltage, or timings depending on which it's crashing for?  Give some more detailed info and I think we can get your system to be actually stable not just 'good enough'. Crashing at stock speeds likely means the voltage is too low or some associated voltage is too low for the CPU to operate properly when at real 100% load, again assuming this is the small fft testing.

 

Perfect example I recently had. I have an undervolted CPU that was stable at -0.145 offset. It ran Prime95 in all the different stress test modes for 24 hours each without a crash of any kind, it ran realbench for several 8 hour sessions without a crash. I changed the CPU cooler. It crashed in Prime95 small fft. Repeatedly would crash, sometimes soon sometimes after hours but it would crash. The ONLY change was the CPU cooler that was smaller and the CPU ran about 20C warmer, peaking out at near 70C instead of 50C. As temps rose how the CPU worked with that -0.145 offset changed and it became unstable. Adding back 0.005 volts made it stable again. Maybe I'm on the fringe here but when I say something is stable I mean it's stable as close to 100% as I can make it, even if it means I don't hit speeds that others say are stable. Data loss, crashing during a game, etc are just not acceptable to me.

[_Hustler_One_]

41 minutes ago, Bitter said:

if it's crashing on small then it's a CPU problem

Yes, I run a small FFTs, but what I cant understand is even I disable my OC, and let it run stock clock without Turbo boost, it still crashing. It crashed before the temps are high enough, still at 60C and it crashed.

[Falkentyne]

9 minutes ago, _Hustler_One_ said:

Yes, I run a small FFTs, but what I cant understand is even I disable my OC, and let it run stock clock without Turbo boost, it still crashing.

At what exact volts and loadline calibration level in bios?

Disable your OC please, set 1.25v in your bios, set Loadline Calibration to Mode 2, then test small FFT again.

[_Hustler_One_]

12 minutes ago, Falkentyne said:

At what exact volts and loadline calibration level in bios?

Disable your OC please, set 1.25v in your bios, set Loadline Calibration to Mode 2, then test small FFT again.

I set 1.1v for stock 3.7GHz, and auto LLC

[Bitter]

Sure sounds like a lack of voltage under load, the above advice sounds solid.

[_Hustler_One_]

15 minutes ago, Bitter said:

Sure sounds like a lack of voltage under load, the above advice sounds solid.

But with that volts it doesnt crash with other test software

[Falkentyne]

48 minutes ago, _Hustler_One_ said:

But with that volts it doesnt crash with other test software

1.1v with auto LLC?
No wonder it crashes.

Auto LLC, if that's set to 2.1 mOhms, and assuming prime95 is pulling 100 amps on your system is:

1100 - (2.1 * 100) = 890mv = 0.890v.  Even if it's using the mOhms value for default loadline for 8 core processors (1.6. mOhms) is still going to be 1100 - (1.6 * 100)=0.940v.  

Trying to run prime95 at 0.890v or 0.940v and complaining that it's crashing?  Good luck man.

(Then again I have no idea what mOhms setting MSI uses for its "Auto" LLC.  Gigabyte uses the same value for Auto/Normal/Standard (1.6 mOhms for 8 core processors, 2.1 mOhms for 4 and 6 core CFL, but it's possible other programs or Multi Core Enhancement can change the LLC if it's set to auto)

 

Please do what I said instead of making excuses first.  Other software doesn't pull the same amps as prime95.  I'm 99.95% sure if you downloaded OCCT 5.0.1 and ran the 90% linpack test, you would get an error after awhile.

[Bitter]

45 minutes ago, _Hustler_One_ said:

But with that volts it doesnt crash with other test software

Other test software don't stress the CPU the same or the the same level. If you watch the CPU package TDP when running other tests and running Prime95 you can see a big difference in how much work the CPU is actually doing.

[_Hustler_One_]

31 minutes ago, Falkentyne said:

1.1v with auto LLC?
No wonder it crashes.

Auto LLC, if that's set to 2.1 mOhms, and assuming prime95 is pulling 100 amps on your system is:

1100 - (2.1 * 100) = 890mv = 0.890v.  Even if it's using the mOhms value for default loadline for 8 core processors (1.6. mOhms) is still going to be 1100 - (1.6 * 100)=0.940v.  

Trying to run prime95 at 0.890v or 0.940v and complaining that it's crashing?  Good luck man.

(Then again I have no idea what mOhms setting MSI uses for its "Auto" LLC.  Gigabyte uses the same value for Auto/Normal/Standard (1.6 mOhms for 8 core processors, 2.1 mOhms for 4 and 6 core CFL, but it's possible other programs or Multi Core Enhancement can change the LLC if it's set to auto)

 

Please do what I said instead of making excuses first.  Other software doesn't pull the same amps as prime95.  I'm 99.95% sure if you downloaded OCCT 5.0.1 and ran the 90% linpack test, you would get an error after awhile.

I'll try to tweak the settings.. Im not expert in OC so, thanks for pointing me on this one..