Help Center/ ModelArts/ FAQs/ Notebook (New Version)/ Code Execution/ What Do I Do If cudaCheckError Occurs During Training?
Updated on 2024-06-15 GMT+08:00

What Do I Do If cudaCheckError Occurs During Training?

Symptom

The following error occurs when the training code is executed in a notebook:

cudaCheckError() failed : no kernel image is available for execution on the device

Possible Cause

Parameters arch and code in setup.py have not been set to match the GPU compute power.

Solution

For GP Vnt1 GPUs, the GPU computing power is -gencode arch=compute_70,code=[sm_70,compute_70]. Set the compilation parameters in setup.py accordingly.