PC instant hardware shutdown if rendering with hyperthreaded Epyc 7773X CPU

[Torstein]

Hey all,

We recently switched out a cpu from Epyc 7373X to the 7773X, which has 64 cores / 128 Threads. It passed all cpu stress tests, but during Corona rendering - and it seems particularly during denoising with High Quality denoiser - the machine panics and shuts off. I don't mean blue screen or Windows performing a shutdown, the power goes straight out, dead.

After trying this multiple times with the same shutdown results, we turned off hyperthreading in the BIOS which stopped the crashing. However, the lack of Hyperthreading loses about 16 percent performance as verified by Vraybenchmark and Passmark tests.

We'd love to switch the hyperthreading back on. Is there any way?

[maru]

Software cannot cause a system to fully shut down. It has to be some hardware issue like overheating, wrong voltage, insufficient power supply.

[Torstein]

Surely that's not correct. The makers of the machine suggested kernel panic, for instance. Cpu cores were monitored and never exceeded standard values. The only change in the machine was swithing the cpu itself, RAM and PSU was identical.

[TomG]

It is correct - software can't make a machine Blue Screen or shut down. Shut downs are generally protective mechanisms, which is based on the hardware getting into some state where it might get damaged, e.g. overheating.

EDIT - as a note, Corona DOES use a CPU to its maximum capacity, for an extended period. This means that if there is some hardware issue like a cooling problem, it is more likely to happen with Corona than many other pieces of software (it's especially true during Denoising, if using Corona Denoising, which you are). This is not anything wrong with Corona though, it's the hardware not being able to run at its max for an extended period because of e.g. cooling.

EDIT 2 - remember CPU cores are not the only thing that can overheat. e.g. if it is drawing a lot of voltage at full usage, this could cause other components on the motherboard to get too hot if there is insufficient cooling.

[TomG]

I see lots of ways in which the new processor can consume more power than the old, according to e.g. https://versus.com/en/amd-epyc-7373x-vs-amd-epyc-7773x , for instance the tdp https://versus.com/en/amd-epyc-7373x-vs-amd-epyc-7773x/cpu-tdp

[Torstein]

Hi,

Thank you for the quick responses, I appreciate it.

In this case, we ran OCCT to stress test the CPU and it endured 20 minute ordeals just fine, full blast on all threads. But opening a scene and rendering for under 10 minutes + denoising was instant crash, repeatedly and consistently. Any ideas why the two situations produce different results, and how turning off hyper threading solved it?

[maru]

Denoising is a special case:
https://forum.corona-renderer.com/index.php?topic=32396.msg181824#msg181824

So the CPU may be running hotter, there may be some voltage fluctuations, etc. But it's still not Corona's "fault". Otherwise we would be getting lots of such reports from pretty much all users using the Corona denoiser.