Question Prime95 - worker failing

[milutin]

Hi everybody!

In January I built a new PC consisting of:

CPU: Intel Core i7-9700K CPU @ 3.60GHz, not overclocked
M/B: AsRock Z390 Taichi
GPU: NVIDIA GeForce RTX 2070 (slightly overclocked)
RAM: Corsair Vengeance RGB Pro DDR4 3000MHz 2x8Gb
Cooling: NZXT Kraken X62

Now the issue: Ever since I've been running the Prime95 torture test (Blend, incl. AVX and AVX2 instructions - temperature never exceeds 85 degrees Celsius) one of the workers, and it's never the same, keeps failing after approx. 30 min. showing the message "FATAL ERROR: Rounding was 0.5, expected less than 0.4 Hardware failure detected".

I have run RealBench for hours and IntelBurn stress tests on maximum settings and the CPU hasn't failed so far. The problem persists only with Prime95.

Since the CPU is brand new and on stock settings, I am surprised that it can not pass the Prime95 blend test.

Do I have a problem?

Any suggestions?

Thenks and regards!
Milutin

 

[mgallo848]

Prime95 by default also stress tests your memory so that may be what's failing.

Download memtest and check your memory. It's been a few years since I ran it but I believe it will even tell you what stick of memory is bad.

MemTest86 - Official Site of the x86 Memory Testing Tool

MemTest86 is the original self booting memory testing software for x86 and ARM computers. Supporting both BIOS and UEFI, with options to boot from USB.

www.memtest86.com

 

[milutin]

Thanks, I'll try that one.

 

[milutin]

I've just ran MemTest86 for about 3 hours with no errors. The RAM sticks are OK. So what else could be the cause?

 

[mgallo848]

I've Googled and read several different threads from people with the same exact problem as you. They could pass memtest but fail prime95. In all of the cases I read it turned out to be not enough voltage for either the CPU or the memory. I would try to go in your bios and turn on XMP profile number one. It's possible that your memory may be running at the proper speed but isn't getting the proper voltage it needs. XMP should fix that if that's the case.

 

[milutin]

OK, I will give it a try. I have been running the XMP profile since I built the PC but I'll increase the voltage. See if this helps

 

[mdd1963]

Just out of curiosity, what simultaneous core clock speeds do you see (HWmonitor) while running P95? (And are you running v26.6? OTher versions are bit more stressful, and Intel specs V26.6/ small FFTs as being the 'thermal/TDP test standard' (Worry about more RAM intensive 'blended mode' afterward....

Regarding core clock speeds....Is one core higher than the others by 100-300 MHz, with the remainder at 4.5-4.6 GHz or so, is there evidence of default MCE enabling? (where MCE has all cores at max spec turbo speed, or. 4.9 GHz in the case of a 9700K)

 

[milutin]

Thanks for your help.

I've carefully increased the memory voltage by .005 and am now running the test. Really don't know if this increase is enough, I'm no expert 🆒

My Prime95 version is the latest one, 29.5. Is this the stressful version you're talking about?

And now that you're mentioning it, the simultaneous core clock speed of all cores seems to be stuck at a constant 4.6GHz and won't clock higher. My CPU has obviously turned hot since I've been tweaking around all day...

...and right after the 30 minute mark, one worker has failed again...

How much do I actually have to increase the DRAM voltage to get any results?

 

[ayazam]

im not an expert myself at memory thingy

but im curious, what is your goal having prime95 running fine?

if i were you i'd just ignore that i cant run prime95 flawlessly 🤣
its either my lazyness or my ignorant "ah its just 1 test that i dont complete"

since you run any other test flawlessly... and you dont even overclocked the CPUs...

 

[milutin]

You know what, you're absolutely right...

It's just that I was wondering why I can't run this freaking Prime95 properly with a brand new CPU.

No, you're right, I'll leave it, my PC runs just fine without that torture test...

 

[superninja12]

Because you shouldn't run the latest version , you should run prime95 version 26.6

 

[milutin]

Ok, I got it

Still, what's the use of releasing stress test versions that keep flunking brand new CPUs on factory settings...

 

[mdd1963]

It should not fail/crash....it's only an excessive heat/TDP issue for later versions with small FFTs selected....

All cores at 4.6 GHz is normal operation, I was mainly looking to see if all cores would attempt 4.9 GHz, which would be indicative of MCE-equivalent being enabled in the BIOS, sort of a common trait with many mainboards in semi-overclocking by running all cores at higher max turbo at expense of increased power/heat...

You may need an AVX offset enabled in the BIOS (when AVX workloads enabled), where the max turbo clocks are lowered in the presence of an AVX (torturous) workload, very common these days....(offset of 2 would be 200 MHz less max turbo when running AVX workload, etc..)

What peak core/package temps were being hit in Blended mode?

 

[mdd1963]

I would at least make sure Intel's 64 bit diagnostic tool passes for your CPU, if nothing else...

https://downloadcenter.intel.com/download/19792/Intel-Processor-Diagnostic-Tool

 

[milutin]

So I ran the suggested Intel diagnostic tool - pass, and I ran Prime95 blend test, version 26.6 this time, for 5 hours - also pass ie. no failed workers. Temps never exceeded 70 C. I guess it was the latest version 29.5 that was causing problems. I think I am happy now.

Thanks everybody!

Btw, how do I mark this thread as solved?

 

[littleleo]

I think once you pick someone as giving the solution that might do it.