System crashes after sustained CPU load

[porina]

System:

• 8086k stock clocks, delid + LM
• Noctua D15 (temps typically ~60C under P95 load)
• Asrock Z370 Pro4, bios P1.20, defaults other than XMP
• PSU: Fractal Edison 550W can't remember, at worst Corsair CX550M, if not HXi series. Have to recheck when I'm back home
• Ram: tried various, see below
• Optane 900p 280GB boot, various SATA SSD for storage
• Asus 1080Ti Strix OC

This is something I've noticed on several occasions but as it doesn't affect day to day use I kinda ignored it up to now. If I run some CPU intensive workloads the system may crash after several hours. The workload is comparable to Prime95 as it uses the same core math library. If I run one task with 6 threads on 672k FFT work, it is fine and keeps going. This work is small enough to fit in the L3 cache, so doesn't significantly impact ram. The problematic work is 1792k - 1920k FFT. These tasks exceed L3 cache and hit up ram bandwidth.

 

So, is it the ram? Probably not. Originally the system had G.skill TridentZ 2x8GB 3000C14 in it. After the first time the problem happened, I tried replacing it with Ripjaws 4 4x4GB 3333C16 @ 3000 as I knew from past experience it can be fussy ram. Happened again. Last night, went in Corsair Vengeance RGB 2x8GB 3000C15, and same crash overnight.

 

When it crashes, there is no screen output and the system is completely unresponsive even to reset button. I have to hard cycle to get it back. I haven't tried running ram at base speed yet, but this work is ram bandwidth limited so I'd really like to keep at the higher speed, just without the crashes.

 

System is stable for gaming, I can stay in it all day without problem. Well, not a system stability problem anyway, my mental stability is another matter...  Anything I might have missed?

[GER_T4IGA]

Are you sure you got the whole die covered when deliding? if not you might have hotspots causing a crash. Unlikely but possible

Definitely run memtest to exclude ram from possibe variables. run it for 24h if you do not get immideate faults in the first few runs.

[Bananasplit_00]

you could run memtest86 for a while, could be something like the IMC is crap and dosent want to run the RAM at that speed or it could be three slightly bad sets of RAM or something else entierly

[porina]

Looking at individual core temps they seem consistent, certainly nothing is beyond the 60's.

 

Haven't done memtest, but could try it overnight or when at work. It seems unlikely three different sets of ram are "bad" so leaning more towards either mobo or IMC. I was also wondering if something else on mobo could be overheating but the available sensors look ok.

 

Oh, just saw I have that system in my sig! PSU is a Fractal Edison 550W. Will update 1st post.