AlphaFold installed in HPCC
The following command can be used to find all versions of AlphaFold installed on HPCC:
1 |
|
To find how to load a specific AlphaFold version, where <version>
is
the version of AlphaFold to be load, use:
1 |
|
All AlphaFold versions use the same data structure and
location /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 usually 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:
1 |
|
Please see GPU memory allocation for more information.
AlphaFold example
Users can get an example of AlphaFold to run on HPCC nodes.
Log into HPCC and ssh to a dev node with GPU cards, then run the
following command to copy the example directory AlphaFold
to the current directory:
1 |
|
(If the above command fails, you may need to manually load powertools with
module load powertools
)
After you cd to the directory, you should be able to see the files inside:
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 |
|
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 alphafold
command script and the module file, five options for the job are
specified in the command line:
1 2 3 4 5 |
|
Users can look into the AlphaFold documentation or use the commands:
1 |
|
to find out how to use all options after you load the AlphaFold module.