CUDA_ERROR_INVALID_HANDLE on rtx 5050

TensorFlow
, ,
1

Hi everyone,
I recently bought a new laptop with an RTX 5050 GPU. I installed Ubuntu 24.04.2 LTS and then installed the NVIDIA drivers from the official website. Here’s what I see in nvidia-smi:

+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 575.64.03              Driver Version: 575.64.03      CUDA Version: 12.9     |
|-----------------------------------------+------------------------+----------------------+
| 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 RTX 5050 ...    Off |   00000000:01:00.0 Off |                  N/A |
| N/A   39C    P0              9W /   42W |      15MiB /   8151MiB |      0%      Default |
|                                         |                        |                  N/A |
+-----------------------------------------+------------------------+----------------------+
                                                                                         
+-----------------------------------------------------------------------------------------+
| Processes:                                                                              |
|  GPU   GI   CI              PID   Type   Process name                        GPU Memory |
|        ID   ID                                                               Usage      |
|=========================================================================================|
|    0   N/A  N/A            2964      G   /usr/lib/xorg/Xorg                        4MiB |
+-----------------------------------------------------------------------------------------+

Then, I created a virtual environment and ran python3 -m pip install 'tensorflow[and-cuda].

TensorFlow installs and detects my GPU:

tf.config.list_physical_devices('GPU')
W0000 00:00:1752022359.118443    5335 gpu_device.cc:2430] TensorFlow was not built with CUDA kernel binaries compatible with compute capability 12.0. CUDA kernels will be jit-compiled from PTX, which could take 30 minutes or longer.
[PhysicalDevice(name='/physical_device:GPU:0', device_type='GPU')]

But when I try to run any model or even a simple operation, I get this error

    raise core._status_to_exception(e) from None  # pylint: disable=protected-access
    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
tensorflow.python.framework.errors_impl.InternalError: {{function_node __wrapped__Mul_device_/job:localhost/replica:0/task:0/device:GPU:0}} 'cuLaunchKernel(function, gridX, gridY, gridZ, blockX, blockY, blockZ, 0, reinterpret_cast<CUstream>(stream), params, nullptr)' failed with 'CUDA_ERROR_INVALID_HANDLE' [Op:Mul] name:

I’ve tried multiple TensorFlow versions. Version 2.16 doesn’t see the GPU. Only 2.19 detects the GPU, but I get that CUDA_ERROR_INVALID_HANDLE error. I suspect that full support for RTX 50xx series might not be available yet.

So:

  • Is there a known fix or workaround for this issue?
  • Should I just wait for TensorFlow 2.20? If so, does anyone know when it’s expected to be released?

I’d really appreciate any advice or a clear step-by-step guide to help me get this working.
Thanks in advance!

1 Like
4

Hi @toxa237,

Welcome to the TensorFlow forum!

Try installing the nightly version of TensorFlow or use NVIDIA’s Docker container that supports Blackwell GPUs. If the issue still persists, you’ll need to wait for TensorFlow release that supports your GPU. Thank you!

1 Like
6

thanks @Divya_Sree_Kayyuri
so, i will wait new TensorFlow.
Do you know when will this be?

1 Like
8

in short, I have installed windows and i found this one containers which work

and, i have to run it only from powershell, because when when i run from wsl ubuntu it doesn’t see GPU. Usual tensorflow containers also don’t see it.

1 Like
10

Hi @toxa, Ensure that you’ve installed the appropriate NVIDIA drivers for your GPU. To use your GPU with TensorFlow, you need compatible NVIDIA drivers. Thank you!

1 Like