CPU Overheats during Memtest86

[arthur19]

Running the default series of RAM tests, I've noticed that reported CPU temperature only ever changes in between tests, not during any one test. For large amounts of installed memory, a single test can take a long time, and the CPU temperature can spike significantly before the end of the test when it is next updated. Even more troubling than this behavior, I've found that, at least with several Haswell & Skylake E3 machines, the CPU fan speed does not change during a test, even when the CPU is getting very hot. If a CPU-intensive test (like Test is interrupted, the CPU fan speed rapidly increases only AFTER the test is stopped and the CPU is mostly idled. This behavior seems to imply that during testing Memtest86 is somehow blocking even the EC's / SMBIOS's / low-level hardware's ability to monitor CPU temperature and control CPU fan speed. I don't know how this is possible, but the fact remains that the CPU is overheating during testing, which is very frustrating.

On some of my systems I can work around the problem by temporarily setting a minimum fan speed in BIOS. In other systems (notably Dell Precision 3420) the only choice other than automatic fan control is "full speed fans" which is VERY loud.

Is there some way that Memtest86 could update CPU temperature every x minutes or even (say) 10 times per test, rather than only between tests? More importantly, is there a way to allow automatic fan speed control to work during RAM testing?

[David (PassMark)]

We aren't aware of any fan control problem.
MemTest86 isn't actually all that CPU heavy. Displaying the CPU temperature more frequently isn't going to cool the machine down, or change the way the fans work. MemTest86 doesn't do anything to control the fans.
Maybe a BIOS bug? Or the BIOS algorithm is slow to respond? Or only works on system temperature and not CPU temperature?

[arthur19]

Thank you very much for your quick response. Good to know (and makes sense) that MemTest86 doesn't actively do anything to control the fans. It could be a BIOS bug. I was thinking not, because two different models of HP (Z230 & Z240) and the Dell 3420 show the same behavior. But perhaps they all share some upstream BIOS code.

While displaying CPU temperature more frequently won't solve this, it would allow me to better monitor and abort the test when CPU temp gets too hot, instead of letting the CPU cook itself for up to 20-30 min until reaching the end of a test and alerting me to the problem.

Your idea of BIOS algorithm being slow to respond is interesting. Could this explain why fans ramp up shortly after aborting a test, while they never do if testing is simply allowed to continue? I don't know enough about details of MemTest86 and BIOS fan algorithm operation. If both are operating in real mode (?) is it possible that MemTest86 could "crowd out" BIOS fan control algorithm, so it's effectively inoperative until RAM testing is stopped?

Also interesting that you say that MemTest86 isn't that CPU heavy. The machine does draw close to maximum wall power during RAM testing, and heats up accordingly. Maybe this is because it's running in real mode or otherwise doesn't make use of various power saving states that Linux and Windows do.

Thanks again for your help.