How trustworthy is the IPMI temp readout on an X8SIL-F?

[Metaluna]

I'm currently in the middle of a large (multi-terabyte) rsync between two ZFS servers. The source server is using a Supermicro X8SIL-F / Xeon X3440 and is in a Norco 4020 case with the stock fans exhausting out the back. It's not the best ventilated case in the world, but the two rear fans plus the PSU give pretty decent airflow and the overall system draws probably around 150W when it's loaded like this.


With an rsync server running, the Supermicro IPMIView application is reporting a system temperature of 75C(!), and a CPU temp of "medium". The other server, which is only a few feet away, has almost the same hardware specs, except it's an X8SIA-F motherboard, and it's in an Antec P180 case with a Supermicro 5-in-3 holding most of the drives. This system only reports 37C.

Should I be concerned? What exactly is the System Temp measuring? Is it the northbridge?

Another difference is that the primary server is running OpenIndiana virtualized on top of ESXi with two LSI controllers passed through. The backup server is running FreeBSD 9 on bare hardware.

Could the virtualization overhead (VT-d and all that) be putting a lot of stress on the motherboard chipset? As I've been noticing this for a while, I actually rigged a slow 80mm fan into position over the northbridge heatsink, but it doesn't seem to help much, which makes me wonder if the temperature is accurate.

 

[dswartz]

Does the freebsd box have the same two LSI controllers? If not, maybe they are the heat contributors?

 

[Patrick]

Let me ask a dumb question, are the temperatures rising or holding steady? Do you have at least decent airflow over the board? I'm a bit slammed, but sometimes locations of temp sensors are found in the motherboard manuals.

 

[Metaluna]

The freebsd box has the same LSI 2008 controller, but it has only one of them. The Norco box has about 6 more drives in it than the freebsd box, but they are green drives. I'd guess 6 green drives plus another LSI 2008 under moderate load would add maybe 50W? Also I don't have the midplane fan board installed. Like I said this box doesn't have superb airflow, but the system temp seems way out of proportion to the ambient air temp inside the case. The air coming out the back is warm, but not roasting hot. But then again maybe the temp just is what it is. I guess it's time to finally install my old cavediver 120mm fan board in this case and see if it helps.

I've been observing the box a little more, and the temperature seems to peak when rsync is transferring large files like ISOs (where I'm getting about 80MB/s over my LAN). I have a few databases that have huge numbers of small files, and when those are transferring, the temps drop to around 50C or so. Rsync CPU usage also drops way down. Rsync really hammers the CPU on large fast transfers.

 

[Metaluna]

I think I solved my problem.

First, I tried installing my fan board with 3 120mm fans. This didn't help the System Temp one bit, but at least my hard drives were running cooler. Then, based on the observation that whenever the System Temp shot up to 75C+, the CPU temp also would report "Medium", I reasoned that maybe the CPU and System Temp were linked more closely than I had assumed. So, I tried replacing the stock Intel cooler with a nice tower cooler, in this case an Arctic Cooling Freezer 13 CO (CO = "Continuous Operation", i.e. it comes with a supposedly high duty cycle 92mm fan).

Bingo. Both my CPU and System Temp readouts immediately dropped. The temperature now stays <=50C during the same operations that were previously sending it up near 80C.

So either the System Temp on the X8SIL-F really is just the CPU temp, or the CPU dumps so much heat into the motherboard when it has an inadequate HSF that it causes all the other components to get alarmingly hot.

 

[Patrick]

Good to know. Motherboards do generally have at least one TS near the CPU area so that is very plausible. Good to hear you figured it all out.