does cudaThreadExit() clean up the memory allocated by cudaMalloc()

[pssuperZoro]

Hi guys, the reference manual only lists that cudaThreadExit() “explicitly cleans up all runtime-related resources associated with the calling host thread”, so i am wondering whether cudaThreadExit() implicitly cleans up the memory allocated by cudaMalloc()?

[pssuperZoro]

Hi guys, the reference manual only lists that cudaThreadExit() “explicitly cleans up all runtime-related resources associated with the calling host thread”, so i am wondering whether cudaThreadExit() implicitly cleans up the memory allocated by cudaMalloc()?

[YDD]

I would expect it to free up all memory on the GPU - after all, the GPU memory is managed by the host, so if the driver context goes bye-bye, then all record that the GPU memory is in use goes as well. However, relying on the driver to tidy up your messes might not be considered best practice.

[YDD]

I would expect it to free up all memory on the GPU - after all, the GPU memory is managed by the host, so if the driver context goes bye-bye, then all record that the GPU memory is in use goes as well. However, relying on the driver to tidy up your messes might not be considered best practice.

[pssuperZoro]

Thanks for replying.

I would expect it to free up all memory on the GPU - after all, the GPU memory is managed by the host, so if the driver context goes bye-bye, then all record that the GPU memory is in use goes as well. However, relying on the driver to tidy up your messes might not be considered best practice.

[pssuperZoro]

Thanks for replying.

I would expect it to free up all memory on the GPU - after all, the GPU memory is managed by the host, so if the driver context goes bye-bye, then all record that the GPU memory is in use goes as well. However, relying on the driver to tidy up your messes might not be considered best practice.