Remember, they weren't mining or even connected to the pool at this point. The temps just started rising past 80c without shutting down. The firmware seems not to shut the rig down when it is over 80c. That really needs to be fixed.
Well that goes against everything I've tested, pulled the ethernet and nothing exciting happens. And I've previously tested the 80C limit, it 'freezes' and stops mining.
I agree - it doesn't make sense. My S3's properly honored that 80C safety shutoff. But on my pair of S5's, it does not shutoff over 80C. BTW, the miners really do heat up and it's not just an errant report from the firmware.
A few weeks ago when Ghash went down, my Nicehash limit for return value wasn't met and so it stopped hashing (no rollover) and I did catch this screenshot.
There is a conflict in that screenshot, a fan speed of 3600 RPM yet a high temp. At this point I'm leaning towards it reporting silly temperatures rather than reaching silly temperatures. On a 200W load, even with a 1000 RPM fan it should still be okay. Heatsink temps of 75C over heatsink temps of 50C provides DOUBLE the heat removal in a 25C ambient.
Ah! I think you have landed on an issue. Thank you.
The physical truth is, at the time of the screenshot, the fan was not at a high rpm (judging from startup/reboot speeds it was in the 300 to 500 rpm range); and the heat sink fins were too hot to keep my fingers on. I believe the heat reported, but not the fan speed.
But that still makes me wonder why the 80c safety didn't kick in. This also happened about 3-4 weeks before this screenshot. That time it reported 106C, but I didn't grab a screenshot, I was freaked out by the high temp and hurried to power down and move air over them with a box fan.
So the bits of information we have:
a) Status reported high temps
b) Heatsinks were too hot to keep fingers on are ~55C for aluminium, which puts chip temps not much higher.
c) Status reported high fan
d) Fan was low in reality
e) Status did not respond to these 'high' temps
f) Status previously showed 110C, but S5 still works.
From the very last one alone we could conclude that its a display / calculation issue, and the chips aren't reaching the temperature its displaying and / or the measurement isn't even what its displaying. If that type of temp sensor was 110C, the chips would have been way past the "melt into goop" region.
So I just ran a test, by killing my internet. Very interesting to watch. The fans spin slow (at an inaudible level). The GUI seems to be stuck with the last valid stats from while hashing.
I'm getting the feeling that the GUI only reports when the machine is actually hashing. I watched the temps of the heatsinks rise by ~8C during the lull of no hashing (with an infrared thermometer).
After connecting the internet, and the miners quit beeping, then the GUI would report an updated status. After the beeping stopped (internet connected) the fans also did spin up again (full rpm).
After restarting the internet, 1 miner had blades at 96C & 83C reported. It would not start hashing again, without a reboot. The other miner had blades at 93C & 78C reported. Based on my infrared temp readings, both heatsinks were within 1C of each other. This miner did start hashing again on its own, after restarting my internet connection.
Here's some absolutes I know after my test of an internet failure:
1) fan speed drop to nothing and the temperature rises during the failure
2) the windows GUI does not get updated status from the miner during the failure
So why don't the fans stay up to speed to keep the chips cooling during an internet failure?
Beyond that, I only have assumptions (though educated guesses, they are still guesses). It's as though the system is not monitoring its own temps during the internet failure, to adjust fan speed and maintain an operating temperature.
If I was a manufacturer, or had a miner that I didn't have to pay for, I would have let the my internet failure run for more than the 3 or 4 minutes I dared to (watching the temps go up). I would be interested to see how high the temps would really go after 30 minutes or more (but not on my out-of-pocket equipment).
