Issue with GPU load

[Ahmed-E-86]

GPU load = 0% all the time, and it doesn't change at all.

System information:
OS: Kubuntu 19.10 - kernel: 5.3.0-29
GPU: AMD Radeon HD 8750M
CPU: Intel core I5 4200M
Driver: Mesa 20-dev (obiaf ppa)

[flightlessmango]

The problem seems to be that /sys/class/drm/card1/device/gpu_busy_percent can't be read.
cat /sys/class/drm/card1/device/gpu_busy_percent
cat: /sys/class/drm/card1/device/gpu_busy_percent: Invalid argument

[NerosTie]

In my case, it doesn't even exist.

cat: /sys/class/drm/card1/device/gpu_busy_percent: No such file or directory

I think it's a tool for AMD GPUs.

I have a Nvidia GPU + Optimus.

edit: tested with the commands from here and I can have the gpu load. Maybe it can help.

[jackun]

@NerosTie where did you get MangoHud? It shouldn't go so deep that it complains about gpu_busy_percent, hopefully, and there is no cat .../gpu_busy_percent anymore

[NerosTie]

From AUR on Arch.

[jackun]

Strange, did you install it just today?

[NerosTie]

Yes, like 1 hour ago.

[jackun]

Oof, which one? This https://aur.archlinux.org/packages/mangohud ?

[NerosTie]

This one and lib32-mangohud.

[jackun]

Bizarre. Only cat usage is

src/overlay.cpp:831:      ram =  exec("cat /proc/meminfo...
src/overlay.cpp:833:      cpu =  exec("cat /proc/cpuinfo...
src/overlay.cpp:837:      os = exec("cat /etc/*-release...

[NerosTie]

Because it's a fopen command in a cpp code:
amdGpuFile = fopen((path + "/device/gpu_busy_percent").c_str(), "r");

But it's the case you have an AMD GPU.

For Nvidia:

            if (device_data->gpu_stats) {
              // get gpu usage
              if (device_data->properties.vendorID == 0x10de)
                 pthread_create(&nvidiaSmiThread, NULL, &getNvidiaGpuInfo, NULL);

But I don't know how it works.

edit:

I think it's that thing in cpu_gpu.h that doesn't work:

void *getNvidiaGpuInfo(void *){
    if (!nvmlSuccess)
        checkNvidia();

    if (nvmlSuccess){
        getNvidiaInfo();
        gpuLoad = nvidiaUtilization.gpu;
        gpuTemp = nvidiaTemp;
        gpuMemUsed = nvidiaMemory.used / (1024 * 1024);
    }

    pthread_detach(nvidiaSmiThread);
    return NULL;
}

[jackun]

Check if libnvidia-ml.so.1 exists. It should be somewhere in /usr/lib/x86_64-linux-gnu/. Could be installed by libnvidia-compute-[some version] package.

[NerosTie]

whereis libnvidia-ml.so.1                                                
libnvidia-ml.so: /usr/lib/libnvidia-ml.so.1 /usr/lib/libnvidia-ml.so

[flightlessmango]

@NerosTie what nvidia GPU is it?

[NerosTie]

GeForce GT 730M

[flightlessmango]

@NerosTie I think your problem is actually related to #3. The gpu is too old and does not play well with NVML, the solution is probably to implement a separate branch that uses nvctrl instead of NVML

[flightlessmango]

I think this issue is more or less 'resolved' as the GPU is not supported. Feel free to open up the issue again if that's not correct