EPYC Rome ES CPU temperature

[chraac]

I recently purchase a 7452 QS 57-04 on EPYCD8, 8x16GB 2400.
And I found that some temperature in HWInfo64 was ridiculous high:

As you can see that "CPU Die (average)" was around 72°C and when I run the Prime95 torture test it'll quick raise up to 100°C (TjMax), but the CCD temp seems pretty low (30-40°C).
Can someone tell me what this temp actually is and it's ok to ignore it?
Thanks

 

[Martin]

"CPU Die (average)" is actually the reference average temperature preferred by AMD and this value is also reported by Ryzen Master.
The difference from other values is indeed too big, so I suspect this might be due to the ES stage. Perhaps a BIOS update with newer AGESA and SMU firmware might fix this.

 

[chraac]

Thanks for replying, and is there any doc about this temp from AMD? I upgraded to latest BIOS and it still too high

 

[Martin]

Sorry, no documentation about this. Do you observe throttling when it reaches high levels?

 

[chraac]

There are no throttling when I run the Prime 95, so maybe is false alarm?

 

[Martin]

Is Ryzen Master perhaps running there and showing temperatures?

 

[chraac]

Is Ryzen Master perhaps running there and showing temperatures?

The Ryzen Master reports unsupported hardware and exits, unable to read the temp ...

 

[RolloZ170]

There are no throttling when I run the Prime 95, so maybe is false alarm?

be careful, AMD ES have no limiter/fuse. they burn out without warning.

 

[Martin]

be careful, AMD ES have no limiter/fuse. they burn out without warning.

No limiter on what, temperature? I doubt that.

 

[chraac]

be careful, AMD ES have no limiter/fuse. they burn out without warning.

Thanks for kindly reminder, is no limiter mean that the Rome ES will not throttling or just the CPU * Limit will be 0? as you can see the Throttling is always "No" and CPU * limit have some number.

 

[chraac]

I google about the CPU Die (average), it says that it's the average temp of all parts in cpu, we know that Rome has a separated IO die, maybe the temp is from IO die buildin sensor?

 

[Martin]

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

 

[Zach]

...and I seriously doubt that IOD can be hotter than CCD(s).
I know that this is no Ryzen, and the IOD has to deal with more CCDs but still...

 

[chraac]

...and I seriously doubt that IOD can be hotter than CCD(s).
I know that this is no Ryzen, and the IOD has to deal with more CCDs but still...

I think it's not the IOD, I ran some cpu only test program to force the cpu at max power but the IOD idle, and the temp was still 100 degree. maybe it's the false alarm.

 

[Zach]

Yes this doesnt make sense... Though on the other hand I find all the others a little low (~50s) when running Prime95.
And I dont know if you can run anything on these CPUs without stressing (low or high) the IOD, as its the connection to the system. Everything goes through the IOD.
You can see the IOD separate power consumption on HWiNFO (CPU SoC Power).

 

[chraac]

Yes this doesnt make sense... Though on the other hand I find all the others a little low (~50s) when running Prime95.
And I dont know if you can run anything on these CPUs without stressing (low or high) the IOD, as its the connection to the system. Everything goes through the IOD.
You can see the IOD separate power consumption on HWiNFO (CPU SoC Power).

The SoC max power on my system is 20w, I think it's because the 8ch memory controller, so I ran some CPU heavy but no memory usage program, with this program the SoC only consume 9w, so I said it's idle

 

[Zach]

Ok, 9W is down to “simple” Ryzen levels of idling or low load on the IOD. Mine fluctuates between 9W and 13W with an avg of 11-12W mostly because I have DRAM and UMC/IF oc to 1900MHz along with all IOD/Fabric power savings disabled.

I also saw some Ryzen (single and dual CCD, dual channel Ram) CPUs to draw 20-25W for the IOD alone, but that’s rare.

 

[RolloZ170]

No limiter on what, temperature? I doubt that.

i meant it is not a good idea to stress a engineering samples(unlocked) with temperature and load.
i don't know what stepping his cpu has, early ones just take to much power on high temps and burn out, that's what the seller say'd.

 

[Martin]

It should be B0 stepping and quite unlikely that it would be lacking a catastrophic thermal protection.

 

[RolloZ170]

As you can see that "CPU Die (average)" was around 72°C and when I run the Prime95 torture test it'll quick raise up to 100°C (TjMax), but the CCD temp seems pretty low (30-40°C).
Can someone tell me what this temp actually is and it's ok to ignore it?

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

IPMI sensors