So I just got an Epyc 7713P system in the mail, and one thing I have noticed is how it screams. I checked the temperatures using sensors on a stock Fedora server install, and even though each of the dies was sitting at a frosty 35 degrees C, the tctl (which I read is a non-physical temperature) is sitting at a toasty 90C. Is this normal? Is this part of the reason why my server (which I was fighting to not get a 1U…) is so vocal? If there’s a way to make it chill out and be happy with running a little above 40C, that’d definitely help save my sanity.
Alternatively, if anyone has done a fan mod on a recent ASUS 1U server, I’d be all ears to hear what’s been done.
Definitely don’t think so, this system is fresh from the manufacturer, and every single die is showing as barely ever hitting 40 while idle. The fact that it seems to always be a 50C offset between tctl and tccd (the actual die temperature) matches with the expected behavior, other than the insane magnitude. Kernel driver k10temp — The Linux Kernel documentation
I really just think I need to find a way to either control that offset (don’t think so) or find a way to change the fan ramp patterns (unlikely but more)
I previously owned a water cooled 7713P system and the tCTL temperature being reported was outright wrong. It’s a known issue. Nothing is wrong with the hardware.
Luckily my IPMI was not using that temperature for fan control, so it wasn’t reflected in higher fan speeds. It’s also worth noting that this incorrect tCTL value won’t cause undue throttling, so it can be safely ignored.
Modern BMCs like the AST 2500 & 2600 support manual fan speed and fan curves on a per-fan-header basis, unless the manufacturer has opted to hide or remove the feature. But frankly, a 240 watt CPU is not going to be quiet in a 1U server. Epyc processors idle in excess of 75 watts.