1

I want to see the GPU usage of my graphic card but its showing N/A!. I use Windows 10 x64, with an Nvidia GeForce GTX 1650.

I am getting the GPU availability status when executing my custom code on the Jupyter-Notebook. But after running Nvidia-smi it shows N/A for all kind of processes. In the Task-Manager my Python process is running on GPU 1 and the power consumption shows that GPU 1 is used.

Why is the GPU utilization N/A and how to fix or circumvent the issue?

Here is the output of nvidia-smi command

Here's the output of nvidia-smi command

Here is the Task-Manager process view

task-manager output

thatguy
  • 21,059
  • 6
  • 30
  • 40
kush
  • 11
  • 1
  • 6
  • Not sure whether this is useful for your use case, but you could try TechPowerUp GPU-Z to monitor GPU activity. – njuffa Jun 29 '20 at 23:05
  • Thanks for the response, but Is there a problem with my GPU Graphic card(GeForce 1650) that does not support Nvidia-smi monitoring?? – kush Jun 30 '20 at 08:00
  • I have never used a GTX 1650 and I cannot remotely diagnose hardware-related issues. You asked "how to ... circumvent the issue?" thus my suggestion. I have been using GPU-Z on Win10 x64 for a while for continuous monitoring of my GPU. It's a free download, so there seems to be no harm in giving it a try. – njuffa Jun 30 '20 at 08:35
  • 2
    Per-process memory usage is not typically referred to as GPU utilization. If you run `nvidia-smi -q`, you should be able to see why `N/A` is displayed: `Not available in WDDM driver model`. Under WDDM, the operating system is in control of GPU memory allocation, not the NVIDIA driver (which is the source of the data displayed by `nvidia-smi`). – njuffa Jul 03 '20 at 10:32
  • @njuffa when I run ```nvidia-smi -q``` I see exactly what you said. Does that mean `nvidia-smi` in Windows is currently not able to fill in values for `GPU Memory Usage` column? – user3731622 Aug 09 '20 at 22:31
  • 1
    @user3731622 While I cannot be sure, because I don't have such a system configuration in front of me, I *think* that per-process GPU memory usage can be display by `nvdia-smi` when a GPU is operated with a `TCC` (originally, **T**esla **C**ompute **C**luster) driver instead of a `WDDM` driver. Use of consumer GPUs with a `TCC` driver is not supported. There are no such limitations under Linux. – njuffa Aug 09 '20 at 23:03
  • I hit this problem? Is there any solution to this? – Eduardo Reis May 27 '22 at 23:22

0 Answers0