EPYC Rome ES CPU temperature

[RolloZ170]

Maybe, but originally (at least for Ryzen) this value was said to exclude IOD.

I suspect the processor adds the temperatures of all 4 CCDs and then incorrectly divides them by 2 instead of 4.

 

[chraac]

HWinfo sensors: CPU Die (average) is IMHO nonsense. in the IPMI sensor page is nothing hot.

IPMI sensors

Thanks for the important picture, may cpu is B0 stepping and the temp in ipmi is not as high as the die average, from you picture I see your CPU also high in this die average temp, so I think it's probably because of the ES stage CPU.

 

[chraac]

I suspect the processor adds the temperatures of all 4 CCDs and then incorrectly divides them by 2 instead of 4.

Maybe it's calc by adding all the temp and then divide by some factor

 

[RolloZ170]

Maybe it's calc by adding all the temp and then divide by some factor

would be helpful to see a 7452/7502 production units CPU die (average) value.

 

[chraac]

would be helpful to see a 7452/7502 production units CPU die (average) value.

it's not so easy to find a 7452/7502 with windows os and run hiwnfo

 

[chraac]

I found some pic about this temp of TR 3970x which is the threadripper countpart of epyc 7452.

https://www.reddit.com/r/Amd/comments/ebbqjn

 

[RolloZ170]

I found some pic about this temp of TR 3970x which is the threadripper countpart of epyc 7452.

7452 ???
this part has only 155 Watts.
the 7542(225W) is more near the 3970x

 

[chraac]

7452 ???
this part has only 155 Watts.
the 7542(225W) is more near the 3970x

Yep, I think the 7452 and 7542 are just different in frequency but same in silicon die, so they are same in the physical chip...

 

[Martin]

Still a Threadripper and not EPYC. Running a different BIOS class, AGESA, SMU, etc. than EPYC.

 

[chraac]

Still a Threadripper and not EPYC. Running a different BIOS class, AGESA, SMU, etc. than EPYC.

Ok, but EPYC with windows and running hwinfo is hard to find.

 

[Chozo4]

Ok, but EPYC with windows and running hwinfo is hard to find.

If it helps... I'm running a Production Sample EPYC 7252 on bare metal here under Windows 10 1809. Noticing a similar trend with the CPU Die (Average) temp accordingly which may point to this not being an ES exclusive issue or even specific to 7452/7502 for that matter. There is barely any load on the CPU at the moment.

I would be happy to lend myself to any further testing needed.

 

[Martin]

Well, then this seems to be an issue of the SMU which calculates such a high average value.
There's a new feature in works now, which is currently just waiting for AMD's approval: per-core temperature monitoring. If this gets approved (which I expect/hope to happen in the next couple of days), it would give you a better insight into individual temperatures to confirm what's correct and what isn't.

 

[Tad863]

Hi, I know this topic has been quit for a while but I have run into the same issue with an AMD Epyc 7302p that I purchased from Ebay. This is from running Prime 95 with only 5 cores (10 threads) (of 16/32). This only causes 82W consumption and the highest core temp is 52.7C, yet the average is 95.3C. Since there aren't that many Epyc systems running on Windows apparently, I thought I'd share a screenshot.

Has there been any news on this since 2021?
Is this average reported by the CPU or this calcualted within HWiNFO?

 

[Martin]

Sorry, no news. Looks like a problem in the CPU firmware (SMU).
Try to upgrade the BIOS to the latest version if that will help.