GPU temperature increases up to 30° after a connection

[waelti_de]

Hello,

We have the phenomenon that the temperature of the GPU increases up to 30° when connecting from a Windows client to a Linux client (CentOS 7.10), even if you do nothing on the Linux machine. When you disconnect again, the temperature stays that high. When I restart the nxserver on the Linux side the temperature will normalize again until someone reconnects. The graphics cards are NVidia K4000 and K4200. Is there a way to prevent this with a setting?

Kind regards

Walter

[fra81]

Hi Walter,

please show us the output of these three commands:

nvidia-smi --query-gpu=timestamp,temperature.gpu,utilization.gpu,utilization.memory,memory.total,memory.free,memory.used --format=csv -l 1

nvidia-smi --query-compute-apps=timestamp,process_name,pid,used_memory --format=csv -l 1

nvidia-smi --query-accounted-apps=timestamp,pid,gpu_util,mem_util,time --format=csv -l 1

Run each the command, reproduce the issue with high GPU temperature, the collect the output.

Also gather logs from the linux machine as shown in https://www.nomachine.com/AR10K00697.

[waelti_de]

nvidia-smi –query-gpu=timestamp,temperature.gpu,utilization.gpu,utilization.memory,memory.total,memory.free,memory.used –format=csv -l 1

2020/03/26 22:01:07.010, 81, 1 %, 0 %, 4036 MiB, 3713 MiB, 323 MiB

2020/03/26 22:01:08.011, 81, 0 %, 0 %, 4036 MiB, 3713 MiB, 323 MiB

2020/03/26 22:01:09.011, 81, 0 %, 0 %, 4036 MiB, 3713 MiB, 323 MiB

2020/03/26 22:01:10.011, 81, 0 %, 0 %, 4036 MiB, 3713 MiB, 323 MiB

2020/03/26 22:01:11.012, 81, 0 %, 0 %, 4036 MiB, 3713 MiB, 323 MiB

The other two commands produce no output

Attachments:

1. NoMachine-log.tar.gz

[fra81]

Hi Walter,

sorry for delay. At the moment we know that only Nvidia driver versions 4xx are affected and only with some video card models, versions 3xx are not. But it’s still not clear what is triggering the issue. Investigation is ongoing. At the present stage, it looks like NoMachine can cause this behaviour only indirectly and there is no evidence of a problem in our software.

[Author]

Posts