Processors

[ivan-filippov]

We have a lot of people with WHEA errors - see 

"https://community.amd.com/t5/processors/ryzen-5900x-system-constantly-crashing-restarting-whea-logge...

I think it would be nice to find the most stable and quick way to reproduce such issue. Please share your findings

Environment:
Default settings in BIOS with XMP on (Core Performance Boost enabled, Performance Boost Overclock disabled)

AGESA 1.1.9.0 

My memory was on 3600 by default I checked 3200 and got the same error also.

For me the fastest way to reproduce was: 

Run benchmark in Shadow of Tomb Raider (I used highest settings with RTX ultra 2560*1440 Vsync off)

1. Start benchmark 

2. Wait sometime as it runs - at this time CPU will be pretty heavy loaded on all cores wait ~ 20-30 seconds

3. press Esc.  

4. Wait until results of benchmark appear. At this point we are switching from multi core load to single load to display this menu with results. In MSI Afterburner I saw that core was boosted to max 4950 for my CPU

6. Wait ~ 5-10 seconds - you can click some tabs here just to give a bit more load for that core

7. If there is no crash - go to 1

In this case I ususally got an error in less than 10 minutes
here you can find the recording of such test
https://youtu.be/9ZVOLd6nD_0?t=441

P.S.
1. I was able to reproduce the same error in RDR2 also - but it takes a bit more time because of longer loadings

2. I was unable to reproduce this issue running different synthetic CPU benchmarks (cinebench, geekbench, occt, aida64 and so on)

3. If I turned off Core Performance Boost - everything works rock solid

[rumple]

Try this. Determine what your preferred  cores are on CCD1 in particular, using Ryzen Master. (The cores with the stars). Open Prime95. Select "Large FFT", Disable AVX2 and AVX.  Select 1 Thread.  Click start. Open Task Manager. Click on the DETAILS TAB.  Find the Prime95 worker in the list and right-click and choose Set Affinity to CPU CORE (preferred #).  Now only the one thread is running on your preferred core  Stop and start the test about 6 times.  You'll see at the beginning of the restart, the core clocks as high as possible.  You will see a hardware error if the cores are unstable when attempting to boost.  Note: You have to set affinity, run just one or two threads, and stop and start the test.  

If you set your PBO2 curve off, or at least don't go with negative 30 on all, you might reduce the errors.  I'm running -14 on the preferred cores, -22 on the non-preferred cores on ccd1, and negative 30 on the ccd2 non-preferred cores.

[NanoExplorer]

Hey there! Unfortunately I don't have Shadow of the Tomb Raider, so I can't test that one out super easily.

Is there a trick to starting and stopping the P95 torture test with the same settings? Every time I stopped it and went to restart it I would have to re-disable the AVX and reset the number of threads. I'd really be interested to see this crash under Prime95 but I don't have the patience to change all those settings every time.

My go-to way of reproducing the WHEA error has been to run Unigine's free "Superposition" benchmark. If I set it to "game" -> preset: custom; resolution: System; shaders/textures: high, then after 10 minutes I get a very reliable crash.

[ivan-filippov]

Guys, thx for the hints!

"Is there a trick to starting and stopping the P95 torture test with the same settings"
I think some script can be written (bat or PowerShell) to automate this.

There is a simple utility on Github - it generates load per core in a cycle
https://github.com/jedi95/BoostTester/releases/download/1.1/BoostTester.exe
in my case, I did not get crash with it - but for some people, it helps to get this crash fast

[ivan-filippov · Accepted solution]

Here is the excellent script to reproduce the issue

Just run it on stock CPU settings - and if you get crashes\errors in log - that's a good reason for RMA

https://www.overclock.net/threads/single-core-prime95-test-script-for-zen-3-curve-offset-tuning.1777...