PC Drivers & Software

[fractalyse]

I'm running ubuntu 18.04 with latest drivers 19.10 and kernel 4.15.0-48-generic.

I'm experiencing hard reboot while heavy Load on gpu. There is absolutely no System log since it look limes NMI.

Reboot happends with a gpu Power draw of 80w or 200w either (with Radeon VII).

This only happends under linux, I've tried to reproduce under windows but this never happends.

I have also tried 18.50 drivers with exactly same results.

I've also tried kernel 4.18, and 5.1 same behavior.

Before Radeon VII I was using 2x Rx580 running a 100w of constant power draw(each) never experience reboot like this before switching to new Radeon VII.

Thanks in advance

EDIT :

Finaly I got same behavior on windows running openCL stress test. Same while playing The Division 2. Seems to be the GPU the faulty.

EDIT Bis:

I've tried on differents computers, GPUs seems to be faulty.

I've bought 4 radeon VII,  One never works at all (computer won't post), 2 were defectives. It looks like there is QC at all.

700€ per card, 3 weren't working correclty.. I'm a bit disappointed.

I will send them back, again, and hope to have new one which are working this time (I hope..)

Thanks to all for helping

[andreluizbarbieri]

continue ... problems are to be reported.

[qwixt]

Can you compare voltage in wattman between Linux and Windows? Perhaps there is a difference in defaults. The only time I have experienced an reboot with the VII like you mention is when I went a little aggressive with the undervolting at 990mv.

[fractalyse]

I'm running at stock clock, 1009mv @ 1801Mhz.

I have more informations

First it's direclty depending of opencl version pulled by amdgpu-pro driver.

- 18.50-725072 luxmark(ball 50830 pts) and superposition (1080p medium 9486 pts) can run correclty and complete, reboot still observed while long opencl compute

- 18.50-756341 luxmark(ball 51304 pts) and superposition (1080p medium 9487 pts) can run correclty and complete, reboot still observed while long opencl compute

- 19.10-785425 luxmark or superposition, computer reboot few seconds after starting benchmark.