Dear author：
I run Fastfold in a 4 GPU device， each GPU have an 24GiB memory。
I run inference.py with an fasta lenght 1805AA (without triton), with parameter --gpus 3
and the error prints like:
RuntimeError: CUDA out of memory. Tried to allocate 29.26 GiB (GPU 0; 23.70 GiB total capacity; 9.63 GiB already allocated; 11.79 GiB free; 10.65 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF
my questions is:
-
why there are only one GPU(GPU0 but not GPU0, GPU1,GPU2) used to calculate total memory? what should I do to get over this?
-
Is there a way to run an extremely long fasta files, like 4000AA?
appriciate your reply, thankyou.
Dear author：
I run Fastfold in a 4 GPU device， each GPU have an 24GiB memory。
I run inference.py with an fasta lenght 1805AA (without
triton), with parameter --gpus 3and the error prints like:
RuntimeError: CUDA out of memory. Tried to allocate 29.26 GiB (GPU 0; 23.70 GiB total capacity; 9.63 GiB already allocated; 11.79 GiB free; 10.65 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONFmy questions is:
why there are only one GPU(GPU0 but not GPU0, GPU1,GPU2) used to calculate total memory? what should I do to get over this?
Is there a way to run an extremely long fasta files, like 4000AA?
appriciate your reply, thankyou.