Inconsistent performance with nvidia prime

Below image is the output of nvtop running 9 instances of prime-run mangohud glxgears. We can see that gpu usage fluctuates anywhere between 20-98%. This leads to drops frame rates and laggy experience.

The ouput of neofetch is

[sk@endeavour ~]$ neofetch
                     ./o.                  sk@endeavour 
                   ./sssso-                ------------ 
                 `:osssssss+-              OS: EndeavourOS Linux x86_64 
               `:+sssssssssso/.            Host: TravelMate P259-G2-MG V1.51 
             `-/ossssssssssssso/.          Kernel: 5.15.11-arch2-1 
           `-/+sssssssssssssssso+:`        Uptime: 1 hour, 28 mins 
         `-:/+sssssssssssssssssso+/.       Packages: 1201 (pacman) 
       `.://osssssssssssssssssssso++-      Shell: bash 5.1.12 
      .://+ssssssssssssssssssssssso++:     Resolution: 1366x768 
    .:///ossssssssssssssssssssssssso++:    DE: Plasma 5.23.4 
  `:////ssssssssssssssssssssssssssso+++.   WM: KWin 
`-////+ssssssssssssssssssssssssssso++++-   WM Theme: Sweet-Dark-transparent 
 `..-+oosssssssssssssssssssssssso+++++/`   Theme: Breeze Light [Plasma], Sweet-Dark-v40 [GTK2/3] 
   ./++++++++++++++++++++++++++++++/:.     Icons: candy-icons [Plasma], candy-icons [GTK2/3] 
  `:::::::::::::::::::::::::------``       Terminal: konsole 
                                           CPU: Intel i5-7200U (4) @ 3.100GHz 
                                           GPU: NVIDIA GeForce 940MX 
                                           GPU: Intel HD Graphics 620 
                                           Memory: 2167MiB / 7818MiB 


The output of nvidia smi is

[sk@endeavour ~]$ nvidia-smi
Thu Dec 30 11:20:44 2021       
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 495.46       Driver Version: 495.46       CUDA Version: 11.5     |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|                               |                      |               MIG M. |
|===============================+======================+======================|
|   0  NVIDIA GeForce ...  Off  | 00000000:01:00.0 Off |                  N/A |
| N/A   43C    P8    N/A /  N/A |      6MiB /  2004MiB |      0%      Default |
|                               |                      |                  N/A |
+-------------------------------+----------------------+----------------------+
                                                                               
+-----------------------------------------------------------------------------+
| Processes:                                                                  |
|  GPU   GI   CI        PID   Type   Process name                  GPU Memory |
|        ID   ID                                                   Usage      |
|=============================================================================|
|    0   N/A  N/A       533      G   /usr/lib/Xorg                       2MiB |
+-----------------------------------------------------------------------------+

Let me know if i can provide more info.

What if you get rid of mangohud? Are these glxgears all running in the background, or do you move them around?

I didn’t had mangohud before, i just installed them to check why my csgo was lagging. I checked it with mothergunship also, same issue.

I used glx gears to see what the gpu is actually doing under load and to be able to take a good screenshot of the scenario.

1 gear was not loading GPU enough so i took 9.

I’m not sure that multiple instances of glxgears is a reliable way to determine an issue here. For example, the games could be running the CPU and/or GPU at 100% and causing thermal throttling - but glxgears isn’t showing that.

Here is the issue shown in csgo.

no, its not thrrotling, as shown in above screemshot the temp of gpu is around 68. and cpu never got above 80.

manjaro is downgrade version drivers nvidia to series 495-44 ( Testing )
can you downgrade from your side, if possible use in cache version ( lastest 495-44 series )

Hmm.

Could you try GPU-Burn and see how the laptop behaves?

Also, check whether you have the Plasma compositor enabled or not, or any additional NVIDIA options like Pipeline Compositing.

Same results

I tried both enabled and disabled. Disabled had slightly good response but gpu was still jumping in usage.

I havent configured any additional nvidia options other that what EOS already does.

If GPU-Burn can’t keep it at 100% utilisation then it can only imply power management of some sort. I’d normally say thermal throttling, but if all temperatures are below 80C then that’s unlikely to be happening… though it would still be interesting to monitor system temperatures and see if there is any relation to GPU utilisation.

It might also be worth making sure fans and vents are clear of dust, just to make sure cooling is as good as it can be.

2 Likes

Re-ran the Gpu-burn test, this time the performance was consistent at full 100% utilisation. I left it to see if it throttles and at 90c it did. then i thought maybe when cpu+gpu both are stressed then it might cause issues. So i went to use stress and gpu burn at same time but still performance was consistent. Its only the games and glx gears that suffers this problem. I tried with plasma compositor both on and off.

1 Like