update Nvidia device driver docs to link to list of supported cards and newer versions

[sofixa]

Description

NVIDIA Device Driver docs had a few older references (e.g. nvidia/cuda:11.0-base doesn't exist anymore), and didn't have a link to the list of compatible NVIDIA devices (e.g. Jetsons aren't compatible).

Also removed the x86_64 qualifier for Linux because the driver runs successfully (even if it can't fully fingerprint the unsupported card) on arm64 in my testing, and the underlying library and its Go bindings are arm64 compatible (NVML and nvml-go).

Reviewer Checklist

• Backport Labels Please add the correct backport labels as described by the internal
  backporting document.
• Commit Type Ensure the correct merge method is selected which should be "squash and merge"
  in the majority of situations. The main exceptions are long-lived feature branches or merges where
  history should be preserved.
• Enterprise PRs If this is an enterprise only PR, please add any required changelog entry
  within the public repository.

[aimeeu]

Thanks for the update! I left a few style suggestions.

[tgross]

Also removed the x86_64 qualifier for Linux because the driver runs successfully (even if it can't fully fingerprint the unsupported card) on arm64 in my testing, and the underlying library and its Go bindings are arm64 compatible (NVML and nvml-go).

I'm hesitant to remove this unless we've actually tested it on arm64 and shown it works end-to-end. I don't see any issues in the driver repo that show someone using it on arm64. Do you know of anyone who's done so?

[sofixa]

@tgross I've ran it on arm64, kind of, and it works enough to identify the NVIDIA GPU, but fails at getting a device handle to check power/memory because the card in question is an iGPU which isn't supported by NVML.

We can give it a spin in an AWS G5g (Graviton ARM CPU + NVIDIA T4 GPU, which is supported by NVML).

[tgross]

We can give it a spin in an AWS G5g (Graviton ARM CPU + NVIDIA T4 GPU, which is supported by NVML).

If you could, that'd be great. It'd at least give us a reasonable smoke test before we attest to it.

[sofixa]

@tgross gave it a spin on an AWS G5g.2xlarge, the driver compiles like a charm, and everything I could think of works:

root@ip-172-31-5-203:/home/ubuntu# nomad node status -self -verbose
ID              = 3187b260-20ae-6a1d-ffc4-e121a3a7298a
Name            = ip-172-31-5-203
Node Pool       = default
Class           = <none>
DC              = dc1
Drain           = false
Eligibility     = eligible
Status          = ready
CSI Controllers = <none>
CSI Drivers     = <none>
Uptime          = 9m34s

Drivers
Driver    Detected  Healthy  Message   Time
docker    true      true     Healthy   2025-03-27T16:39:45Z
exec      true      true     Healthy   2025-03-27T16:39:45Z
java      false     false    <none>    2025-03-27T16:39:45Z
qemu      false     false    <none>    2025-03-27T16:39:45Z
raw_exec  false     false    disabled  2025-03-27T16:39:45Z

Node Events
Time                  Subsystem       Message          Details
2025-03-27T16:25:05Z  Driver: docker  Healthy          driver: docker
2025-03-27T16:17:35Z  Cluster         Node registered  <none>

Allocated Resources
CPU          Memory       Disk
0/10000 MHz  0 B/7.6 GiB  0 B/3.2 GiB

Allocation Resource Utilization
CPU          Memory
0/10000 MHz  0 B/7.6 GiB

Host Resource Utilization
CPU           Memory           Disk
75/10000 MHz  367 MiB/7.6 GiB  (/dev/root)

Device Resource Utilization
nvidia/gpu/NVIDIA T4G[GPU-0c3f1dde-eb43-9134-a648-afe61c408cfa]  446 / 15360 MiB

....

Attributes
cpu.arch                                 = arm64
cpu.frequency.efficiency                 = 2500

Output of a docker job with nvidia-smi running in a cuda container (same image as the one I've put for the doc).

Thu Mar 27 16:39:57 2025       
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 570.86.15              Driver Version: 570.86.15      CUDA Version: 12.8     |
|-----------------------------------------+------------------------+----------------------+
| 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 T4G                     Off |   00000000:00:1F.0 Off |                    0 |
| N/A   43C    P8             15W /   70W |       1MiB /  15360MiB |      0%      Default |
|                                         |                        |                  N/A |
+-----------------------------------------+------------------------+----------------------+
                                                                                         
+-----------------------------------------------------------------------------------------+
| Processes:                                                                              |
|  GPU   GI   CI              PID   Type   Process name                        GPU Memory |
|        ID   ID                                                               Usage      |
|=========================================================================================|
|  No running processes found                                                             |
+-----------------------------------------------------------------------------------------+

Anything else you can think of that I should test?

[tgross]

Awesome, that's great @sofixa. Let's ship it. I'll mark this for re-review.