GPU Stats #262

eximo84 · 2024-11-05T17:51:16Z

Would be nice to be able to see GPU stats using intel_gpu_top or AMD equivalent.

Running the hub in docker but agent on containers I'm yet to install the agent on the Proxmox host.

henrygd · 2024-11-06T19:12:05Z

We may be able to pull data for Nvidia and AMD GPUs from nvidia-smi and rocm-smi.

It would probably only work with the binary version of the agent, though.

henrygd · 2024-11-09T01:19:06Z

Experimental support for Nvidia and AMD will be added in 0.7.4.

This works for the binary agent only and requires nvidia-smi (Nvidia) or rocm-smi (AMD) to be installed on the system.

To enable, set the environment variable GPU=true.

If you used the install script, you can do this by adding Environment="GPU=true" in the [Service] section of /etc/systemd/system/beszel-agent.service. Then reload / restart the service:

sudo systemctl daemon-reload
sudo systemctl restart beszel-agent

Any feedback is appreciated. If it works then I'll enable it by default in the next minor release.

I don't have a device using an Intel GPU, unfortunately, so I won't be able to add that.

Tip

Installing rocm-smi-lib on Arch and Debian places the rocm-smi binary in /opt/rocm. If this isn't in the PATH of the user running beszel-agent, you can fix by symlinking to /usr/local/bin:

sudo ln -s /opt/rocm/bin/rocm-smi /usr/local/bin/rocm-smi

Morethanevil · 2024-11-09T08:12:25Z

Tested it with my GTX 1660 Super, works fine on Fedora 41 with NVIDIA drivers

Did a short conversation with my LLM :)

eximo84 · 2024-11-09T08:16:19Z

Argh sad you can't get intel stats being most are no doubt using iGPUs 🙁

Amazing work for the Nvidia and AMD guys though.

eximo84 · 2024-11-09T08:30:40Z

If it's useful here is a JSON output from intel_gpu_top


{
        "period": {
                "duration": 16.488048,
                "unit": "ms"
        },
        "frequency": {
                "requested": 2244.049750,
                "actual": 1273.649858,
                "unit": "MHz"
        },
        "interrupts": {
                "count": 2971.849670,
                "unit": "irq/s"
        },
        "rc6": {
                "value": 0.000000,
                "unit": "%"
        },
        "engines": {
                "Render/3D/0": {
                        "busy": 0.000000,
                        "sema": 0.000000,
                        "wait": 0.000000,
                        "unit": "%"
                },
                "Blitter/0": {
                        "busy": 0.000000,
                        "sema": 0.000000,
                        "wait": 0.000000,
                        "unit": "%"
                },
                "Video/0": {
                        "busy": 99.270775,
                        "sema": 0.000000,
                        "wait": 0.000000,
                        "unit": "%"
                },
                "Video/1": {
                        "busy": 51.775195,
                        "sema": 0.000000,
                        "wait": 0.000000,
                        "unit": "%"
                },
                "VideoEnhance/0": {
                        "busy": 0.000000,
                        "sema": 0.000000,
                        "wait": 0.000000,
                        "unit": "%"
                },
                "VideoEnhance/1": {
                        "busy": 0.000000,
                        "sema": 0.000000,
                        "wait": 0.000000,
                        "unit": "%"
                },
                "[unknown]/0": {
                        "busy": 12.718176,
                        "sema": 0.000000,
                        "wait": 0.000000,
                        "unit": "%"
                }
        },
        "clients": {
                "4294005725": {
                        "name": "ffmpeg",
                        "pid": "961571",
                        "engine-classes": {
                                "Render/3D": {
                                        "busy": "0.000000",
                                        "unit": "%"
                                },
                                "Blitter": {
                                        "busy": "0.000000",
                                        "unit": "%"
                                },
                                "Video": {
                                        "busy": "2.226429",
                                        "unit": "%"
                                },
                                "VideoEnhance": {
                                        "busy": "0.000000",
                                        "unit": "%"
                                },
                                "[unknown]": {
                                        "busy": "0.000000",
                                        "unit": "%"
                                }
                        }
                }
        }
},

https://manpages.debian.org/testing/intel-gpu-tools/intel_gpu_top.1.en.html

henrygd · 2024-11-09T19:08:28Z

Maybe xpu-smi is a better option. intel_gpu_top isn't made by Intel and doesn't seem to have been updated in many years.

SonGokussj4 · 2024-11-10T13:00:34Z

How do I do this if I'm running beszel-agent as a Docker service through docker-compose.yaml?

 WARN GPU err="no GPU found - install nvidia-smi or rocm-smi"

# cat docker-compose.yaml
services:
 beszel-agent:
   image: "henrygd/beszel-agent"
   container_name: "beszel-agent"
   restart: unless-stopped
   network_mode: host
   volumes:
     - /var/run/docker.sock:/var/run/docker.sock:ro
   environment:
     PORT: 45876
     KEY: "ssh-ed25519 AAAAC3NzaC1lZ......adrfOpvRdFLD6p"
     GPU: "true"

# nvidia-smi
Sun Nov 10 13:59:38 2024
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 525.125.06   Driver Version: 525.125.06   CUDA Version: 12.0     |
|-------------------------------+----------------------+----------------------+
| 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 GeForce ...  Off  | 00000000:09:00.0 Off |                  N/A |
| 35%   38C    P8    14W / 215W |   1121MiB /  8192MiB |      0%      Default |
|                               |                      |                  N/A |
+-------------------------------+----------------------+----------------------+
|   1  NVIDIA GeForce ...  Off  | 00000000:0A:00.0 Off |                  N/A |
| 33%   35C    P8     2W / 215W |      3MiB /  8192MiB |      0%      Default |
|                               |                      |                  N/A |
+-------------------------------+----------------------+----------------------+

+-----------------------------------------------------------------------------+
| Processes:                                                                  |
|  GPU   GI   CI        PID   Type   Process name                  GPU Memory |
|        ID   ID                                                   Usage      |
|=============================================================================|
|    0   N/A  N/A   1472954      C   python                           1118MiB |
+-----------------------------------------------------------------------------+

Morethanevil · 2024-11-10T17:27:05Z

How do I do this if I'm running beszel-agent as a Docker service through docker-compose.yaml?

 WARN GPU err="no GPU found - install nvidia-smi or rocm-smi"

# cat docker-compose.yaml
services:
 beszel-agent:
   image: "henrygd/beszel-agent"
   container_name: "beszel-agent"
   restart: unless-stopped
   network_mode: host
   volumes:
     - /var/run/docker.sock:/var/run/docker.sock:ro
   environment:
     PORT: 45876
     KEY: "ssh-ed25519 AAAAC3NzaC1lZ......adrfOpvRdFLD6p"
     GPU: "true"

# nvidia-smi
Sun Nov 10 13:59:38 2024
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 525.125.06   Driver Version: 525.125.06   CUDA Version: 12.0     |
|-------------------------------+----------------------+----------------------+
| 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 GeForce ...  Off  | 00000000:09:00.0 Off |                  N/A |
| 35%   38C    P8    14W / 215W |   1121MiB /  8192MiB |      0%      Default |
|                               |                      |                  N/A |
+-------------------------------+----------------------+----------------------+
|   1  NVIDIA GeForce ...  Off  | 00000000:0A:00.0 Off |                  N/A |
| 33%   35C    P8     2W / 215W |      3MiB /  8192MiB |      0%      Default |
|                               |                      |                  N/A |
+-------------------------------+----------------------+----------------------+

+-----------------------------------------------------------------------------+
| Processes:                                                                  |
|  GPU   GI   CI        PID   Type   Process name                  GPU Memory |
|        ID   ID                                                   Usage      |
|=============================================================================|
|    0   N/A  N/A   1472954      C   python                           1118MiB |
+-----------------------------------------------------------------------------+

You can't at the moment, see above

May it will be possible in future. The image needs to include nvidia-smi or rocm-smi, then it should be possible to mount the gpu like this:

services:
 beszel-agent:
   image: "henrygd/beszel-agent"
   container_name: "beszel-agent"
   restart: unless-stopped
   network_mode: host
   volumes:
     - /var/run/docker.sock:/var/run/docker.sock:ro
   deploy:
     resources:
       reservations:
         devices:
           - driver: nvidia
             count: all
             capabilities:
               - gpu
   environment:
     PORT: 45876
     KEY: "ssh-ed25519 AAAAC3NzaC1lZ......adrfOpvRdFLD6p"
     GPU: "true"

NVIDIA container toolkit should be installed.

henrygd · 2024-11-10T19:07:09Z

I don't plan to include nvidia-smi or rocm-smi in the Docker image because it will increase the size many times over for a feature that not everyone will use.

But I can add example dockerfiles and compose configs. Maybe even build them automatically as separate images or tags.

Morethanevil · 2024-11-10T19:10:07Z

I don't plan to include nvidia-smi or rocm-smi in the Docker image because it will increase the size many times over for a feature that not everyone will use.

But I can add example dockerfiles and compose configs. Maybe even build them automatically as separate images or tags.

As separate images would be the best way 👍🏻
Latest-cuda, latest-rocm etc would be cool

SonGokussj4 · 2024-11-11T08:30:29Z

@Morethanevil This didn't work for me. I have container toolkit installed, we are using GPU in our other applications deployed by docker-compose and the deploy section is used the same way.
Here, it still prints the warning.

services:
  beszel-agent:
    image: "henrygd/beszel-agent"
    container_name: "beszel-agent"
    restart: unless-stopped
    network_mode: host
    volumes:
      - /var/run/docker.sock:/var/run/docker.sock:ro
    environment:
      PORT: 45876
      KEY: "ssh-ed25519 AAAAC3Nz.....fOpvRdFLD6p"
      GPU: "true"
      # FILESYSTEM: /dev/sda1 # set to the correct filesystem for disk I/O stats
      #FILESYSTEM: data
    deploy:
     resources:
       reservations:
         devices:
           - driver: nvidia
             count: all
             capabilities:
               - gpu
               
# docker compose down; docker compose up -d; docker compose logs -f
[+] Running 1/0
 ✔ Container beszel-agent  Removed                                                                                 0.1s
[+] Running 1/1
 ✔ Container beszel-agent  Started                                                                                 0.0s
beszel-agent  | 2024/11/11 08:29:40 INFO Detected root device name=nvme0n1p1
beszel-agent  | 2024/11/11 08:29:40 INFO Detected network interface name=enp5s0 sent=446766673405 recv=7705963243348
beszel-agent  | 2024/11/11 08:29:40 WARN GPU err="no GPU found - install nvidia-smi or rocm-smi"
beszel-agent  | 2024/11/11 08:29:40 INFO Starting SSH server address=:45876

So it can't be done if it's not build as separated images?

Morethanevil · 2024-11-11T09:15:02Z

@Morethanevil This didn't work for me. I have container toolkit installed, we are using GPU in our other applications deployed by docker-compose and the deploy section is used the same way. Here, it still prints the warning.

services:
  beszel-agent:
    image: "henrygd/beszel-agent"
    container_name: "beszel-agent"
    restart: unless-stopped
    network_mode: host
    volumes:
      - /var/run/docker.sock:/var/run/docker.sock:ro
    environment:
      PORT: 45876
      KEY: "ssh-ed25519 AAAAC3Nz.....fOpvRdFLD6p"
      GPU: "true"
      # FILESYSTEM: /dev/sda1 # set to the correct filesystem for disk I/O stats
      #FILESYSTEM: data
    deploy:
     resources:
       reservations:
         devices:
           - driver: nvidia
             count: all
             capabilities:
               - gpu
               
# docker compose down; docker compose up -d; docker compose logs -f
[+] Running 1/0
 ✔ Container beszel-agent  Removed                                                                                 0.1s
[+] Running 1/1
 ✔ Container beszel-agent  Started                                                                                 0.0s
beszel-agent  | 2024/11/11 08:29:40 INFO Detected root device name=nvme0n1p1
beszel-agent  | 2024/11/11 08:29:40 INFO Detected network interface name=enp5s0 sent=446766673405 recv=7705963243348
beszel-agent  | 2024/11/11 08:29:40 WARN GPU err="no GPU found - install nvidia-smi or rocm-smi"
beszel-agent  | 2024/11/11 08:29:40 INFO Starting SSH server address=:45876

So it can't be done if it's not build as separated images?

As I said before: it is not possible at the moment with docker, since there is no nvidia-smi included. You need to wait for a separate image. I just provided an example how to include the GPU into the compose.yaml (docker-compose.yml)

SonGokussj4 · 2024-11-11T09:23:54Z

I see, sorry for the misunderstanding. Haven't got a coffee yet. (cheap excuse 😅)

eximo84 · 2024-11-11T21:27:18Z

Maybe xpu-smi is a better option. intel_gpu_top isn't made by Intel and doesn't seem to have been updated in many years.

Interesting. Didn't know this existed. Most guides point you towards jntel_gpu_top.

I'm happy to test but appreciate its tricky to code it you don't have the hardware.

henrygd · 2024-11-12T22:45:55Z

I think I was wrong about intel_gpu_top not being updated actually.

Does anyone know if it works with newer iGPUs and Arc cards?

If someone with Intel can look into this further and compare intel_gpu_top / xpu-smi that would be very helpful.

We need JSON or CSV output and ideally all the same info as Nvidia / AMD -- GPU name, utilization, VRAM usage, power draw, and temperature.

Maybe next week I'll have some time to try doing it blind with sample output.

Obamium69 · 2024-11-14T16:33:36Z

I'm running a Intel Arc A750 and I can read out the following informations as a JSON:

{
        "period": {
                "duration": 1000.351522,
                "unit": "ms"
        },
        "frequency": {
                "requested": 141.950101,
                "actual": 140.950453,
                "unit": "MHz"
        },
        "interrupts": {
                "count": 187.933937,
                "unit": "irq/s"
        },
        "rc6": {
                "value": 74.878730,
                "unit": "%"
        },
        "imc-bandwidth": {
                "reads": 2758.395964,
                "writes": 1349.216517,
                "unit": "MiB/s"
        },
        "engines": {
                "Render/3D/0": {
                        "busy": 0.000000,
                        "sema": 0.000000,
                        "wait": 0.000000,
                        "unit": "%"
                },
                "Blitter/0": {
                        "busy": 0.000000,
                        "sema": 0.000000,
                        "wait": 0.000000,
                        "unit": "%"
                },
                "Video/0": {
                        "busy": 5.275372,
                        "sema": 0.000000,
                        "wait": 0.000000,
                        "unit": "%"
                },
                "Video/1": {
                        "busy": 1.311127,
                        "sema": 0.000000,
                        "wait": 0.000000,
                        "unit": "%"
                },
                "VideoEnhance/0": {
                        "busy": 2.374794,
                        "sema": 0.000000,
                        "wait": 0.000000,
                        "unit": "%"
                },
                "VideoEnhance/1": {
                        "busy": 0.561587,
                        "sema": 0.000000,
                        "wait": 0.000000,
                        "unit": "%"
                },
                "[unknown]/0": {
                        "busy": 0.000000,
                        "sema": 0.000000,
                        "wait": 0.000000,
                        "unit": "%"
                }
        }
},

intel_gpu_top can't give you the exact name of the GPU. When running intel_gpu_top -L the output looks like this:

card1                    8086:56a1                         pci:vendor=8086,device=56A1,card=0
└─renderD128

eximo84 · 2024-11-15T10:07:38Z

I think I was wrong about intel_gpu_top not being updated actually.

Does anyone know if it works with newer iGPUs and Arc cards?

If someone with Intel can look into this further and compare intel_gpu_top / xpu-smi that would be very helpful.

We need JSON or CSV output and ideally all the same info as Nvidia / AMD -- GPU name, utilization, VRAM usage, power draw, and temperature.

Maybe next week I'll have some time to try doing it blind with sample output.

#262 (comment) - my output here from an Intel ARC A310

Jamy-L · 2024-12-26T11:25:50Z

I have been trying to run xpu-smi to provide a sample, without success. Most links on the intel install guides are dead and the install sounds like a massive headache. It must also be noted that few cards appear to be supported, because it targets data centers. intel_gpu_top sounds more reasonable and works with igpus as well. With my 12500 cpu, intel_gpu_top -L gives me:

card0                    Intel Alderlake_s (Gen12)         pci:vendor=8086,device=4690,card=0
└─renderD128

which is totally expected.

henrygd · 2024-12-26T21:27:11Z

@Jamy-L Thanks, let's strike xpu-smi then. Unfortunately intel_gpu_top isn't a great option for this either. Too bad there's not a simple nvidia-smi / rocm-smi equivalent for Intel.

henrygd added the enhancement New feature or request label Nov 5, 2024

henrygd added the in progress We've started work on this label Nov 9, 2024

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

GPU Stats #262

GPU Stats #262

eximo84 commented Nov 5, 2024

henrygd commented Nov 6, 2024

henrygd commented Nov 9, 2024 •

edited

Loading

Morethanevil commented Nov 9, 2024

eximo84 commented Nov 9, 2024

eximo84 commented Nov 9, 2024

henrygd commented Nov 9, 2024

SonGokussj4 commented Nov 10, 2024

Morethanevil commented Nov 10, 2024

henrygd commented Nov 10, 2024

Morethanevil commented Nov 10, 2024

SonGokussj4 commented Nov 11, 2024

Morethanevil commented Nov 11, 2024

SonGokussj4 commented Nov 11, 2024

eximo84 commented Nov 11, 2024

henrygd commented Nov 12, 2024

Obamium69 commented Nov 14, 2024

eximo84 commented Nov 15, 2024

Jamy-L commented Dec 26, 2024

henrygd commented Dec 26, 2024

GPU Stats #262

GPU Stats #262

Comments

eximo84 commented Nov 5, 2024

henrygd commented Nov 6, 2024

henrygd commented Nov 9, 2024 • edited Loading

Morethanevil commented Nov 9, 2024

eximo84 commented Nov 9, 2024

eximo84 commented Nov 9, 2024

henrygd commented Nov 9, 2024

SonGokussj4 commented Nov 10, 2024

Morethanevil commented Nov 10, 2024

henrygd commented Nov 10, 2024

Morethanevil commented Nov 10, 2024

SonGokussj4 commented Nov 11, 2024

Morethanevil commented Nov 11, 2024

SonGokussj4 commented Nov 11, 2024

eximo84 commented Nov 11, 2024

henrygd commented Nov 12, 2024

Obamium69 commented Nov 14, 2024

eximo84 commented Nov 15, 2024

Jamy-L commented Dec 26, 2024

henrygd commented Dec 26, 2024

henrygd commented Nov 9, 2024 •

edited

Loading