mprime fails on linux (debian, 5.3 kernel) when tty is switched

IgorG · 2019-11-15, 10:35

Hi,
mprime is failing immediately when i switch from xsession to tty1 and back to xsesion tty7 (ctrl+alt+f1) but it could run hours without any errors if i ran it initially from tty1. At the same time on Windows 10 prime95 has no any issues.

To test, I've replaced 5700 XT with my old rx 460 and mprime ran without any problems on linux/windows.

What could it be? Bad 5700 XT, motherboard or bug in linux kernel (as i know, 5.3 is the fist kernel version where navi is supported)?

Thanks for any hints.

hardware:

ryzen 3900x
Asus ROG Crosshair VIII Hero (WI-FI)

Seasonic Prime 1300W Gold
Sapphire PCI-Ex Radeon RX 5700 XT NITRO+

paulunderwood · 2019-11-15, 12:46

Quote:

Originally Posted by IgorG

Hi,
mprime is failing immediately when i switch from xsession to tty1 and back to xsesion tty7 (ctrl+alt+f1) but it could run hours without any errors if i ran it initially from tty1. At the same time on Windows 10 prime95 has no any issues.

Is it only mprime that behaves like this? Does it so behave when running the command yes instead?

IgorG · 2019-11-15, 17:40

Sorry for not clear wording. "mprime fails" - not segfaults but shows something like: "FATAL ERROR: Rounding was 0.5, expected less than 0.4" when switching ttys. And only after I've downgraded to rx 460, mprime runs without errors.

Prime95 · 2019-11-15, 18:30

No idea as to your problem. My gut says the AMD driver is not saving and restoring the floating point state properly.

Once you get the time could you provide us with some mfakto and gpuowl benchmarks? There has been keen interest in how well the RX5700 performs.

2019-11-15, 10:35	#1
IgorG Nov 2019 10₂ Posts	mprime fails on linux (debian, 5.3 kernel) when tty is switched Hi, mprime is failing immediately when i switch from xsession to tty1 and back to xsesion tty7 (ctrl+alt+f1) but it could run hours without any errors if i ran it initially from tty1. At the same time on Windows 10 prime95 has no any issues. To test, I've replaced 5700 XT with my old rx 460 and mprime ran without any problems on linux/windows. What could it be? Bad 5700 XT, motherboard or bug in linux kernel (as i know, 5.3 is the fist kernel version where navi is supported)? Thanks for any hints. hardware: ryzen 3900x Asus ROG Crosshair VIII Hero (WI-FI) Seasonic Prime 1300W Gold Sapphire PCI-Ex Radeon RX 5700 XT NITRO+

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Windows Subsystem for Linux v2 gets real Linux kernel	tServo	Software	0	2019-05-07 16:59
Linux Kernel and GPU optimizations	SELROC	GPU Computing	3	2018-11-15 09:52
Howto start mprime automically for (Debian) Linux	joblack	Software	26	2018-11-05 11:18
Linux kernel.shmmax value	pinhodecarlos	Linux	2	2012-06-09 12:55
8 GByte on 32-Bit Linux Kernel with PAE + mprime?	joblack	Hardware	6	2009-07-09 23:18

2019-11-15, 17:40	#3
IgorG Nov 2019 2 Posts	Sorry for not clear wording. "mprime fails" - not segfaults but shows something like: "FATAL ERROR: Rounding was 0.5, expected less than 0.4" when switching ttys. And only after I've downgraded to rx 460, mprime runs without errors.

2019-11-15, 18:30	#4
Prime95 P90 years forever! Aug 2002 Yeehaw, FL 7,949 Posts	No idea as to your problem. My gut says the AMD driver is not saving and restoring the floating point state properly. Once you get the time could you provide us with some mfakto and gpuowl benchmarks? There has been keen interest in how well the RX5700 performs.