AlphaFold installed in HPCC
To use AlphaFold installed in HPCC, users can use the command:
to find all versions installed in HPCC and use the command:
to find how to load a specific AlphaFold version, where
the version of AlphaFold to load. All AlphaFold versions use the same
data structure and
/mnt/research/common-data/alphafold/database as mentioned
in Alphafold via Singularity.
Illegal memory address
If CUDA_ERROR_ILLEGAL_ADDRESS or an illegal memory access was encountered while running AlphaFold, this is mostly due to not enough memory in GPU cards (from the python package "jax"). Please try to request the high memory GPU card A100 (79GB) to run your AlphaFold jobs. You can also set the environment variable:
to see if the allocated memory is enough or not. Please see GPU memory allocation for more information.
Users can get an example of AlphaFold to run on HPCC nodes. After log into HPCC and ssh to a dev node with GPU cards, users can run the powertools command:
to copy the example directory
AlphaFold in the current directory.
After you cd to the directory, you should be able to see the files
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
The job script file
slurm_script.sb shows how to load the
AlphaFold module and run the command
alphafold. Since most of
the specifications and variables have been set in the
command script and the module file, 5 options:
1 2 3 4 5
for the job are specified in the command line. Users can look into the AlphaFold documentation or use the commands:
to find out how to use all options after you load the AlphaFold module.