Monitor GPU usage via nvidia-smi on Thor

As previous threads has mentioned that nvidia-smi is available but it doesn’t log anything, making it hard to monitor the gpu usage for my application. Is there any news on if support is coming? And also if there is any alternatives to use instead.

I run the docker container:

nvcr.io/nvidia/deepstream:8.0-triton-multiarch

Kind Regards

You should be able to see the GPU utilization, but you can’t see a break down of the RAM usage.

+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 580.00                 Driver Version: 580.00         CUDA Version: 13.0     |
+-----------------------------------------+------------------------+----------------------+
| 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 Thor                    Off |   00000000:01:00.0 Off |                  N/A |
| N/A   N/A  N/A             N/A  /  N/A  | Not Supported          |     94%      Default |
|                                         |                        |             Disabled |
+-----------------------------------------+------------------------+----------------------+

+-----------------------------------------------------------------------------------------+
| Processes:                                                                              |
|  GPU   GI   CI              PID   Type   Process name                        GPU Memory |
|        ID   ID                                                               Usage      |
|=========================================================================================|
|    0   N/A  N/A            5243      C   VLLM::EngineCore                          0MiB |
+-----------------------------------------------------------------------------------------+

You can also run nvidia-smi dmon if you want a break out of the processing across the GPU (sm), video encoder (enc), video decoder (dec), jpg, and the optical flow accelerator (ofa)

# gpu    pwr  gtemp  mtemp     sm    mem    enc    dec    jpg    ofa   mclk   pclk 
# Idx      W      C      C      %      %      %      %      %      %    MHz    MHz 
    0      -      -      -      0      0      0      0      0      0      -      - 
    0      -      -      -      0      0      0      0      0      0      -      - 
    0      -      -      -      0      0      0      0      0      0      -      - 
    0      -      -      -      0      0      0      0      0      0      -      - 
    0      -      -      -      0      0      0      0      0      0      -      - 
    0      -      -      -      0      0      0      0      0      0      -      - 
    0      -      -      -     92      0     12      1      0      0      -      - 
    0      -      -      -     92      0     23      1      0      0      -      - 
    0      -      -      -     91      0     20      1      0      0      -      - 
    0      -      -      -     92      0      7      2      0      0      -      - 
    0      -      -      -     92      0     19      0      0      0      -      - 
    0      -      -      -     91      0     22      0      0      0      -      - 
    0      -      -      -     93      0     10      1      0      0      -      - 
    0      -      -      -     92      0     14      2      0      0      -      - 
    0      -      -      -     89      0     22      3      0      0      -      - 
    0      -      -      -     91      0     23      2      0      0      -      - 
    0      -      -      -     92      0     21      1      0      0      -      - 
    0      -      -      -     91      0     23      0      0      0      -      - 
    0      -      -      -     91      0     10      1      0      0      -      - 
    0      -      -      -     93      0     16      0      0      0      -      - 
    0      -      -      -     90      0     18      1      0      0      -      - 

I believe the jtop utility needs to be compiled for Thor? I don’t know if it provides any additional insights either. Another command: tegrastats will give you usage, power, & temperature readings.

sudo apt install nvtop

Does not show “gpu” memory. Does show gpu related process memory use.

Nvtop is a ncurses-based GPU monitoring interface which provides information
 on the GPU states (GPU and memory utilization, temperature, etc) and well as
 information about the processes executing on the GPUs.

Here’s instructions to install jtop for Thor which shows some gpu memory information.

Hi,

GPU usage is supported in nvidia-smi.
Could you share more information about the feature you need?

Thanks.