the Xeon Phi will shut down above 95 degrees C so your application is running fine.
You can set thresholds using micsmc (I believe) so you could set a cut-off value at 90 degrees - but normally the Xeon Phi will "protect itself" and will shut down.
In my case they throttle down rather than shut down completely. Has anyone been successful in running 2 mics at full blast with the passively cooled versions?
One more question related to that: Does mpss tie MIC thermals into temperature control for the host? If hosts' fan speeds are not a function of MICs temperature (just the temperature of the CPU) then that would explain why my MICs start literally boiling at heavier loads...
A passive unit does not mean no fan needed. Rather it means no fan supplied. You need to add a high capacity fan and some duct work. There are a few postings on this forum illustrating fan setups. Use the forum search, and search for "phi fan". There is even one with my posting of a dual 5110P setup. It is important that you have
high capacity fan (at relatively high air pressure differential)
a fan controller that is capable of driving the fan or fans at desired capacity
a fan controller that has temperature sensors
ducting that you will need to fabricate
Read the specifications on the fans. Terms with "high capacity" without numbers are meaningless. It took me three tries to get the appropriate fans, and two tries with fan controllers.
Yes I have (had) such a setup running. It all depends on the server hardware, I guess. I'm using a Supermicro server but I did have to set the system board fans to HIGH (using the BMC). With that setting the fans blowing over the two 5110P Phis are continually on and the temperature is steady at ~ 65 degrees during load.
thanks for the sound information.
I did notice that the system fans on the c612 where not spinning at full RPM, hence the reason for this post.
Some 1st hand info I have experienced with the passive cooling:
on a z97 workstation, a single 31S1P executed full steam for ~24 hours before I do a routine power cycle.
The 31S1P was cooled using a rinky dink EVERCOOL SB-F2 with ~42 CFM
on c612 based FT77CB7079 workstation, 8 5110P with (5+1) 120*120*38 system fans: this does not frequently power cycle.
Temps for both systems stayed around 75-80 degrees Celc.
My setup uses a tower, hand build duct work, and a temperature controlled fan controller. The initial fans I selected had the required CFM, but not under the requirements of the static pressure developed pushing that CFM through the 5110P. The second set of replacement fans came close but reached the point of having the 5100P's throttle back. The third set I chose could meet the CFM and static load, however the temperature controlled fan speed controller did not have sufficient watts per fan channel to get the fans up to required RPM's. I replaced the fan controller with one with sufficient output capacity and the system has been running fine since. I wish I had a 3D printer to build the duct work. What I did works. I selected some polystyrene "No Parking" signs for material, cut to shape, then taped together using electrical tape. It works, its ugly, its out of sight.