failed to run cuBLAS routine: CUBLAS_STATUS_EXECUTION_FAILED

[yejiao0828]

When I try to migrate my code from a P4 GPU to an RTX 4090, I encounter an error in code that previously ran without issue:
failed to run cuBLAS routine: CUBLAS_STATUS_EXECUTION_FAILED.

My environment is as follows:

TensorFlow: 1.15.0

CUDA: 10.0

cuDNN: 7

The 8GB memory on my original P4 GPU was sufficient for running the code, but after switching to the RTX 4090, this error occurs.
According to online sources, there are two potential solutions: reducing the model’s batch_size(), or changing the software versions.
I have already tried reducing the batch size, but the same error still occurs.
Upgrading the software versions would require significant changes to my existing codebase, so I would prefer not to do that.

Is there a better solution? Ideally, one that does not require switching versions.

My log output is shown below:
1.txt

[yejiao0828]

Switching to another GPU is not feasible for me — my goal is specifically to migrate the code to the RTX 4090.
So I would like to ask: Are there any other solutions besides upgrading to a newer version of TensorFlow and CUDA?

[yejiao0828]

。你好，我已经收到邮件