Motherboard: MSI B550 Unify running BIOS version 7D13v132 (beta)
Problem: I am unable to get a stable system using PBO and Curve Optimizer with this system even setting the offset to a positive value for cores that throw MCE (machine check errors).
If I disable PBO and CO entirely, the machine is stable so I am thinking that either PBO/CO is buggy or I do not have it set up correctly.
I initially set all cores to -15 and all other settings left alone. It runs stable doing CPU intensive things like prime95 or compiling with gcc, but when I do very light work loads, I get MCE errors. I have been reducing the negative offset of affected CPUs but ultimately changed from negative to positive and am still getting MCE errors on light workloads.
Is there some other setting I need to manually define beyond:
A-XMP profile 1 setting for the memory.
PBO to Advanced and defined the PBO Limits to Manual with following:
The only modification beyond that was setting Curve Optimizer to Per Core and settings all cores to -15 to start with. Again, I have been giving the affected CPUs +15 values but am still getting MCEs.
Would be nice to have some light on this , because I've been facing the same issue with my 5900x.
System runs very stable doing CPU intensive tasks prime95, CB23 or playing games @ -15 or -20 but crash at idle or on very light work like simply whatching Youtube.
corecycler to test the stability ..... you cut voltage and turbo boost get not enough power will bsod........ i am having this random crash while gaming as well .. tweak and try only... disable curve optimiser directly if want stable
From what I read, it seens that corecycler wont catch iddle or light work load crashes.
Hwinfo monitor if your power deviation drop 6x or not... some mobo bios need update
Using X570 Xtreme mobo with latest release bios F34 AGESA ComboV2 22.214.171.124 B
I have since increased the PPT and TDC to larger numbers (500 and 245). Now, I am able to run with an EDC of 160 and curve values of -15 with apparent stability. Successfully gone through 6 cycles of low level stress updating openwrt patchsets which in the past, has been a good measure for stability here. See: https://wiki.archlinux.org/title/Stress_testing#Low_load_example
How do you tell that particular core caused MCE? You use some application that shows that or ...?
corecycler only can do... not perfect but at least can find the unstable core and increase more voltage for it... disable the curve optimizer just in case fail to stable it.......... and use the pc...
curve optimizer is consider unstable and bring problem only but only good if lucky tweak well
What this video, he talks about how on windows: https://youtu.be/dU5qLJqTSAc
On Linux, journalctl keeps track, see: https://wiki.archlinux.org/title/Stress_testing#Discovering_Errors