Prime95 lock-ups on AMD under Linux

airsquirrels · 2016-01-17, 05:02

I'm having some trouble with prime95 when running under Debian 8 (Jessie) linux. All of my Intel machines can run all day long without issues, however every single AMD machine I have will deadlock after 24-48 hours if prime95 is running.

This has happened on a Phenom II X6 chip, as well as FX8350, 8320, and 9370 chips. Different motherboards, different RAM brands.

The only other piece of information I have is that multiple GPUs are installed in each of these systems. If I only run mfakto / mfaktc / cudaLucas I never get the lockups. It is only when running prime95. Admittedly AMD prime95 performance isn't amazing anyway so it hasn't been a priority, but I recently built several more AMD GPU hosts and I would love to keep those cores busy at least with DC work.

Has anyone else experienced this problem?

Dubslow · 2016-01-17, 06:46

Could you be more specific about what's causing the lockups? In place tests? Large FFTs? All of the above?

Can you boot one of the problem machines to Windows or BSD to eliminate the OS variable?

Mark Rose · 2016-01-17, 06:46

It sounds like the common factor here is the AMD driver.

Does it still happen if you don't use all the cores?

airsquirrels · 2016-01-17, 14:38

I'll gather more information, this is using the latest mprime running double checks usually (too slow for LL). Currently happening with six of the 40,250,XXX range exponents. AMD K10 FFT 2240K FFT.

It happens on both systems with the AMD GPUs/driver installed and systems with the NVIDIA driver installed, so it's not specific to any GPU or driver.

I can try booting my windows live disk, I've also noted that once with a monitor connected that the kernel spit out a message about a CPU deadlock which I can not precisely recall. These machines are usually headless, but I will connect a monitor to one to see if I can reproduce.

I am configured for the full 6/8 workers, however I know AMD cores aren't exactly 1-1 with Intel cores, so I'm not sure if this is the best configuration.

Temps are not an issue, especially right now as one machine is essentially sitting 30* F in a window.

I can try using less cores, if memory serves it still occurred just less often. This hasn't been a very high priority because it is difficult to investigate, that's why I thought I would check if anyone else had experienced it before sinking too much time into experimentation.

2016-01-17, 05:02	#1
airsquirrels "David" Jul 2015 Ohio 1000000101₂ Posts	Prime95 lock-ups on AMD under Linux I'm having some trouble with prime95 when running under Debian 8 (Jessie) linux. All of my Intel machines can run all day long without issues, however every single AMD machine I have will deadlock after 24-48 hours if prime95 is running. This has happened on a Phenom II X6 chip, as well as FX8350, 8320, and 9370 chips. Different motherboards, different RAM brands. The only other piece of information I have is that multiple GPUs are installed in each of these systems. If I only run mfakto / mfaktc / cudaLucas I never get the lockups. It is only when running prime95. Admittedly AMD prime95 performance isn't amazing anyway so it hasn't been a priority, but I recently built several more AMD GPU hosts and I would love to keep those cores busy at least with DC work. Has anyone else experienced this problem?

2016-01-17, 06:46	#2
Dubslow Basketry That Evening! "Bunslow the Bold" Jun 2011 40<A<43 -89<O<-88 3×29×83 Posts	Could you be more specific about what's causing the lockups? In place tests? Large FFTs? All of the above? Can you boot one of the problem machines to Windows or BSD to eliminate the OS variable? Last fiddled with by Dubslow on 2016-01-17 at 06:46

2016-01-17, 14:38	#4
airsquirrels "David" Jul 2015 Ohio 11×47 Posts	I'll gather more information, this is using the latest mprime running double checks usually (too slow for LL). Currently happening with six of the 40,250,XXX range exponents. AMD K10 FFT 2240K FFT. It happens on both systems with the AMD GPUs/driver installed and systems with the NVIDIA driver installed, so it's not specific to any GPU or driver. I can try booting my windows live disk, I've also noted that once with a monitor connected that the kernel spit out a message about a CPU deadlock which I can not precisely recall. These machines are usually headless, but I will connect a monitor to one to see if I can reproduce. I am configured for the full 6/8 workers, however I know AMD cores aren't exactly 1-1 with Intel cores, so I'm not sure if this is the best configuration. Temps are not an issue, especially right now as one machine is essentially sitting 30* F in a window. I can try using less cores, if memory serves it still occurred just less often. This hasn't been a very high priority because it is difficult to investigate, that's why I thought I would check if anyone else had experienced it before sinking too much time into experimentation. Last fiddled with by airsquirrels on 2016-01-17 at 14:40

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Prime95 for PPC Linux	frolyar	Software	4	2015-07-27 14:10
Prime95 on Puppy Linux?	Rodrigo	Software	17	2011-01-30 16:55
Prime95 (32-bit) on 64-bit Linux OK?	fes016	Software	1	2007-10-29 01:50
Torture Lock	SWR_DMaster	Software	5	2006-04-06 21:55
prime95 linux	stealthaxe	Software	1	2003-05-16 17:55

2016-01-17, 06:46	#3
Mark Rose "/X\(‘-‘)/X\" Jan 2013 2951₁₀ Posts	It sounds like the common factor here is the AMD driver. Does it still happen if you don't use all the cores?