Need help with failed prime95 test

[joehesse]

Hello,

I ran ./prime with N to the first prompt and used the defaults for the other prompts.
I have done this a number of times and I always get the same error. Since you have the
following comment in stress.txt, I want to see if the error is due to a software bug.

"HOWEVER, if you are failing the torture test in the SAME SPOT with the SAME ERROR MESSAGE
every time, then ask for help at http://mersenneforum.org - it is possible that a recent
change to the torture test code may have introduced a software bug."

My computer:
Fedora 32 Linux
No overclocking
Gigabyte X570 Aorus Elite Wifi motherboard
AMD Ryzen 9, 3900X, EVGA liquid cooling with 2 fans
32G memory - passed MemTest 86
1TB nvme ssd for OS Fedora Linux 32
1TB mechanical drive for /home and swap
Corsair Gold 650 watt power supply
Bare minimum video Card - MSI GT710 1GB D3 PCIE LP
Computer powered by a heavy duty UPS

I used lm_sensors to monitor the CPU temperature.
$ sensors | grep --after-context=10 k10temp
The output of the above looks like

k10temp-pci-00c3
Adapter: PCI adapter
Vcore: 1.31 V
Vsoc: 1.07 V
Tdie: +69.9°C
Tctl: +69.9°C
Tccd1: +70.2°C
Tccd2: +70.0°C
Icore: 89.00 A
Isoc: 11.50 A

I'm not sure which of the temperatures is the correct one to monitor but all 4
were close to but under 70.0C.

Here is the file results.txt where lines of the form "Self-test xxK passed!" have been omitted.
In this test and previous ones, the error was always a "FATAL ERROR: Rounding was ..."

I would appreciate any help you can give me.
Thank you,
Joe

=================================================================================
[Mon Jun 22 13:25:22 2020]
[Mon Jun 22 13:31:06 2020]
[Mon Jun 22 13:37:14 2020]
[Mon Jun 22 13:43:41 2020]
[Mon Jun 22 13:50:17 2020]
[Mon Jun 22 13:57:03 2020]
[Mon Jun 22 14:03:38 2020]
[Mon Jun 22 14:09:16 2020]
[Mon Jun 22 14:14:53 2020]
FATAL ERROR: Rounding was 0.4998519667, expected less than 0.4
Hardware failure detected, consult stress.txt file.
[Mon Jun 22 14:22:02 2020]
FATAL ERROR: Rounding was 0.5, expected less than 0.4
Hardware failure detected, consult stress.txt file.
[Mon Jun 22 14:27:58 2020]

[paulunderwood]

This is tough problem. What version of mprime are you using?

[axn]

Originally Posted by joehesse

Here is the file results.txt where lines of the form "Self-test xxK passed!" have been omitted.
In this test and previous ones, the error was always a "FATAL ERROR: Rounding was ..."

You'll need to include those "passed" lines so that we can understand if the error is always happening at the same point in the test. Add it as an attachment to the post, rather than as inline text.

My first impression is that you have a flaky hardware. People have run mprime on Ryzen 3000 series without any software bugs being uncovered; not impossible but unlikely.

[moebius]

Maybe you could try a different XMP profile with less timing and latency for your RAM?

[kruoli]

Have you tried the stress test feature of y-cruncher, too?

[joehesse]

My bios has only 1 xmp profile and it is either enabled or disabled. When I did the original prime95 test it was enabled. I disabled it and prime95 went for 2.5 hours without an error. I will try again and if prime95 runs for 10 hours or so without error I will assume I solved it.


Thank you all for your responses.

[moebius]

Originally Posted by joehesse

My bios has only 1 xmp profile...

I have an MSI X470 mainboard with 3700X.
You can also set in the BIOS that the memory runs with an specific XMP profile and e.g. 3200 instead of 3000 MHz, which is prime95 stable for me.
Sometimes round off errors first occur at e.g. 50% of the LL test, but then quite a few times in succession.