Counteracting Too Cold

[EdH]

i grew up with plastic insulation procedures quite common, but I'd rather attack the issue in a manner that doesn't include seasonal intervention with con/destruction requirements.

[masser]

Do all of the shed desktop computers have GPUs?

Adding a small, cheap GPU to each could be:
1. Fun
2. Interesting
3. additional heat
4. more continuous work: if/when unmonitored CPU tasks conclude, GPU tasks continue and vice versa.

[EdH]

Originally Posted by masser

Do all of the shed desktop computers have GPUs?

Adding a small, cheap GPU to each could be:
1. Fun
2. Interesting
3. additional heat
4. more continuous work: if/when unmonitored CPU tasks conclude, GPU tasks continue and vice versa.

They are all headless, although most have onboard video. No current GPUs.

[VBCurtis]

I ran peltier cooling between my CPU and heatsink back in the Celeron-300 days (stable at 504mhz for Prime95 for months!). When a peltier has no heat to pump, the cold side gets quite cold. The CPU likes the cold, but with extreme cold comes condensation- at least, when the ambient air is warm compared to the components. I also ran a machine in a chest freezer at 0-5F to see how much bigger an overclock I could manage- it was less than 10%, so that was a short experiment. Cold didn't crash the machines.

Anyway, cold on its own should not cause a machine malfunction. So, I wonder if any dampness in the shed condenses on a cold motherboard when it idles, or if BIOS just has a "shutdown if temp outside of X range" safety. Is it the same machine or couple of machines that crash when idle? If so, could be shed location -> condensation, or BIOS.

It shouldn't take much heat within the case to keep components warm enough to not crash. Something as simple as reversing the power-supply fan to exit into the case, or unplugging all intake case fans, each winter may maintain case temps high enough to avoid shutdowns (or at least delay them long enough for the next task to be fired up).

Even a half-hearted attempt to reduce airflow in/out of the shed should be enough- I wouldn't insulate it, but closing off big (like cat-sized, or maybe rat-sized) airgaps in the structure should hold in enough heat to avoid the problem.

[EdH]

It does seem only certain machines have suffered this type of failure. I must pay more attention during the upcoming winter. I can also surround the whole set of machines with a tarp, but I'm not sure of the potential for trouble during warmer swings. I might look into such.

Quite some time ago, I wrote a security program for a local club's entrances and had an issue with hard drives becoming corrupted during extreme lows in unheated buildings. I did enclose those machines for the winters.

[Mark Rose]

Originally Posted by EdH

My "farm" consists of recycled machines of all sorts. A common temperature specification is operation close to within human comfort. That works for those in the house, but my more capable Desktop machines are located in a shed that is temperature controlled by nature. The higher extremes seem to be handled adequately by throttling on those few really hot days during our two weeks of summer.

However, I find that many of them do not survive the sub-zero (Fahrenheit) temperatures of the long months of winter, unless they are kept busy, so they don't think about the cold.

I have tried running a secondary factoring routine with nice -n19 to keep one that recently crashed "warm" during its slack times of the main process(es) it is running, but even with nice, the secondary process detracts from the primary one(s) more than I would like.

Would anyone care to offer any ideas for a solution?

Rename one mprime to something else, then use the PauseWhileRunning= configuration option in a second. That will keep the second mprime paused.

Or run mprime in a loop: while true; do mprime -d ; sleep 5 ; done # the sleep 5 is to allow you to control-c it before the loop restarts.

[EdH]

Originally Posted by Mark Rose

Rename one mprime to something else, then use the PauseWhileRunning= configuration option in a second. That will keep the second mprime paused.

Or run mprime in a loop: while true; do mprime -d ; sleep 5 ; done # the sleep 5 is to allow you to control-c it before the loop restarts.

I'm guilty of posting in the wrong forum, this being the only "Hardware" sub-forum readily apparent. Although I only run mprime for testing, I could certainly use mprime for the purpose I'm seeking. But what would I do with all the money I'll get from the results?

Thanks for the suggestion.

Originally Posted by VBCurtis

I ran peltier cooling between my CPU and heatsink back in the Celeron-300 days (stable at 504mhz for Prime95 for months!). When a peltier has no heat to pump, the cold side gets quite cold. The CPU likes the cold, but with extreme cold comes condensation- at least, when the ambient air is warm compared to the components. I also ran a machine in a chest freezer at 0-5F to see how much bigger an overclock I could manage- it was less than 10%, so that was a short experiment. Cold didn't crash the machines.

Anyway, cold on its own should not cause a machine malfunction. So, I wonder if any dampness in the shed condenses on a cold motherboard when it idles, or if BIOS just has a "shutdown if temp outside of X range" safety. Is it the same machine or couple of machines that crash when idle? If so, could be shed location -> condensation, or BIOS.

It shouldn't take much heat within the case to keep components warm enough to not crash. Something as simple as reversing the power-supply fan to exit into the case, or unplugging all intake case fans, each winter may maintain case temps high enough to avoid shutdowns (or at least delay them long enough for the next task to be fired up).

Good elements in the above post. I've overclocked several machines successfully with each machine running three or four 970 GPU's. These machines would be running in our unheated garage during the winter where the temperature outside would be negative double digits. Condensation was an ever present concern so aeration must be considered. Since you have a shed (and thinking old-school economics) you may consider having a bed of decomposing fertilizer covered by plastic (venting appropriately) around your machines. We're surrounded by farmland so access to this (cheap) material is plentiful. Think of yourself as the professor on Gilligan's island..

[tServo]

This is an FYI rather than a possible solution for EdH's problem.
Some machines have sensors in them to enforce the documented operational temperature range.
The ones I have seen are mostly higher end machines like workstations.
I have a few Dells in my unheated garage and if I wait too long to fire them up when it starts to get cold,
they will refuse to start with a POST code indicating "too cold".
I believe the documented lowest operational temperature is 50 degrees F ( 10 degrees C).

HP might put checks like this in their workstations.

[EdH]

Originally Posted by tServo

This is an FYI rather than a possible solution for EdH's problem.
Some machines have sensors in them to enforce the documented operational temperature range.
The ones I have seen are mostly higher end machines like workstations.
I have a few Dells in my unheated garage and if I wait too long to fire them up when it starts to get cold,
they will refuse to start with a POST code indicating "too cold".
I believe the documented lowest operational temperature is 50 degrees F ( 10 degrees C).

HP might put checks like this in their workstations.

This is of interest. I have seen operating specs for some of my hardware giving a low value of 32F. I've "assumed" that if I keep the CPU cranking, it will take care of the rest. Perhaps this is a(nother) bit of evidence against assumptions.

[chalsall]

Originally Posted by EdH

This is of interest.

If you want to get serious about determining the problem(s) (and have a bit of fun)... I highly recommend the Cacti monitoring system.

The graphs are pretty! And very useful to assist with correlation determination.