Been real busy but I've migrated my server to the new MS-01 with the PTM7950 and TGP8000PT installed.
I've been monitoring my temps over the last few days and its pretty shocking, actually.
Considering that
this was what I was looking at to begin with, even with case off:
And then after adjustments with
powersave script and fan curve by
@wadup, the temps dropped quite well:
Now I've migrated to a brand new box, with PTM7950 and TGP8000PT installed, and I'm sending my "old" system off today.
Temps I'm now getting are significantly lower again. Now this is what I'm seeing under "normal load".
For me normal load is AT MINIMUM running 18 dockers, 2 Windows VMs:
From what I've seen the
CPU temps do not rise above ~48c at this point, and this is with
THE CASE BACK ON.
I monitored it for over a day with
the case off as before, and
CPU temps didn't get above about ~43c.
As for the MB/MotherBoard temp, that still seems pegged in the ~114-117c region, which is still alarming, but then
@PigLover pointed out that this is some kind of sensor error, and I'm inclined to agree with him after seeing this between 2 separate units. I'm not sure how to adjust this, but I will need to chase this up inside Unraid.
Something especially interesting, is when I reboot and go into BIOS and look there, I see these temps.
Left, with normal stock fan curve: Right, with adjusted fan curve:
What's interesting to observe is that the MB temp is lower yes, but the also the CPU fan speed is ramped up.
So at least in Unraid, the fan speed on the right would appear to be the CPU fan.
This of course has no effect from the powersave script function, as that is applicable from inside Linux OS (No idea how to apply this in Windows yet). Seems a bit off that temps would be higher outside OS without load, but sensor readings are questionable at this point.
Lastly, I've noticed a SIGNIFCANT power usage drop between the two MS-01 units, I don't have exact numbers for individual items, because I measure "total system power" from my MS-01 and all attached storage devices as my measurement. The drop in power is hard to explain.
Suffice to say, I'm running 8 external disks and the MS-01 itself, and the power usage has dropped from ~102-112w to ~83-91w, and again this is total system power. I have changed no other hardware, this is just trading MS-01 unit itself to new one with PTM7950 and TGP8000PT installed:
In conclusion, I would call the MS-01 unit swap and fresh PTM7950 and TGP8000PT installation a success.
I think its important to state I am now using a
new unit, with paste change BEFORE I USED IT,
from my 4-month old MS-01 with heat issues.
I'd highly recommend others do the same, I hope this helps to explain to others.
Just wanted to chime in with my two cents. I've been struggling with overheating issues on two MS-01 servers for a while. I tried a bunch of different thermal pastes and application methods, and even used liquid metal at one point (but I ditched that because it started corroding the aluminum). The heat profile was really odd—after the CPU hit 80 degrees under load, the temperature wouldn't drop even after the load was gone. Surprisingly, disabling the E-cores solved the problem. Not only did it boost the performance of my virtual machines, but it also fixed the overheating issue. Now, as soon as the load is gone, the CPU temperature drops right back to a stable 50 degrees.
To add something to this, in Unraid under my use case, I DO NOT use E-cores in VMS. I make sure to pin any VMs to 6 P-cores only, and all Dockers use the 8 E-cores basically exclusively. I've played around with this setting over time, and I have 1 VM at this point that uses 6-P-Cores and 4 E-cores, and a couple dockers that use 4 P-cores and all 8 E-cores (making sure that P-cores are not shared between VMs and Docker containers).
I have 5 VMs in total, but mainly 2 are running and 30 Docker containers, but mainly 18 are running.
Optimising P/E core usage is good for performance, and I find it lowers temps overall.
This is hard to describe perhaps, but here is sample images to help from Unraid OS.
Note that there is a much longer list of Docker containers, MOST of which use no P-cores at all, just these top few:
