Subject: CUFFT_INTERNAL_ERROR on Specific GPU Models While Running WhisperX Model
Hello,
I've encountered a recurring CUFFT_INTERNAL_ERROR while running the WhisperX model for audio transcription on Runpod, specifically with certain GPU models. Below are the details:
Successful Execution:
RTX A5000: GPU Utilization 96%, Successful
RTX 3090: GPU Utilization 97-100%, Successful
RTX A4000: GPU Utilization 97%, Successful
RTX A4500: GPU Utilization 96%, Successful
Error Encountered:
L4: GPU Utilization 100%, Error: CUFFT_INTERNAL_ERROR
RTX 4000 Ada: GPU Utilization 31%, Error: CUFFT_INTERNAL_ERROR
The issue seems to occur irrespective of the GPU's memory capacity, as it affects both 24 GB (L4) and 16 GB (RTX 4000 Ada) GPUs. The same script and audio file work without issues on Google Colab using T4 or V100 GPUs.
P.S My endpoint ID: m3lzr86sanjxt5
0 Replies