I have 2 Pascal GPUs under heavy load (MSI Afterburner says 80% power) : GTX1070 xtreme gaming (G0, PCIe x16, 4k video out) and GTX1070 g1 (G1, PCIe x4, unconnected) setup and they work fine in Windows 10 x64 Nvidia driver 376.19, motherboard z170-HD3P.
Well, apart from Google Chrome kills G0’s core clock from over 2000 to 1600. When Chrome stopped and after 5 minutes wait, clock went up again. No power or temp problem. Strange. G1 worked fine all time (I think I even swapped cards, same).
But under Linux (I think I even swapped cards, same), Fedora 24 x64, kernel 4.8.11-300.fc25.x86_64,
nvidia-smi shows G0’s power.draw only 64.26 W and clocks.mem only 810 MHz.
Using latest Nvidia driver version 375.20.
nvidia-smi --query-supported-clocks=timestamp,gpu_name,gpu_bus_id,gpu_serial,gpu_uuid,memory,graphics --format=csv
timestamp, gpu_name, gpu_bus_id, gpu_serial, gpu_uuid, memory [MHz], graphics [MHz]
2016/12/13 17:26:40.652, GeForce GTX 1070, 0000:01:00.0, [Not Supported], GPU-5154a925-2ee2-4075-76fe-bed257ef642b, [Not Supported], [Not Supported]
2016/12/13 17:26:40.653, GeForce GTX 1070, 0000:07:00.0, [Not Supported], GPU-314a701d-a3a2-5698-210e-d9004f0b2df8, [Not Supported], [Not Supported]
The problem is the same with only one card GTX 1070 xtreme installed.
- Is it possible that MB (or GPU) bios upgrade helps (note Win works, Linux does not) ?
- Is there any tool like MSI Afterburner for Linux to change fan / clocks (sudo nvidia-smi in persistence mode does not allow that for me) ?
- Would you give any general advice please, what shall I try / debug / etc. ?
Please note my time with G0 will be over in couple days, unless I am able to run it flawlessly.
Please note I am using unsupported config, but in the same time please note CUDA supported Fedora 23 will officially enter End Of Life (EOL) status on December 20th, 2016 (in 7 days).
( I posted a link to this in Linux section as well )