Differenze

Queste sono le differenze tra la revisione selezionata e la versione attuale della pagina.

--- calcoloscientifico:userguide:alphafold [24/01/2025 19:01] – fabio.spataro
+++ calcoloscientifico:userguide:alphafold [06/02/2025 19:46] (versione attuale) – fabio.spataro
@@ Linea 13: / Linea 13: @@
 <code>
 /hpc/share/containers/apptainer/alphafold/3.0.1/alphafold-3.0.1.sif
+</code>
+=== Alphafold3 GPU demo ===
+<code>
+mkdir -p demo/af_input
+cp -p /hpc/share/containers/apptainer/alphafold/3/af_input/fold_input.json demo/af_input
+cp -p /hpc/share/containers/apptainer/alphafold/3.0.1/slurm-alphafold-gpu-a100_40g.sh demo
+cd demo
+sbatch slurm-alphafold-gpu-a100_40g.sh
 </code>
@@ Linea 36: / Linea 46: @@
 </code>
-Script ''slurm-alphafold.sh'' to run ''alphafold'' on 1 node with 1 GPU (8 tasks per node):
+Script ''slurm-alphafold-gpu-a100_40g.sh'' to run ''alphafold'' on 1 node with 1 A100 (40 GB) GPU (8 tasks per node):
-<code bash slurm-alphafold.sh>
+<code bash slurm-alphafold-gpu-a100_40g.sh>
 #!/bin/bash --login
 #SBATCH --job-name=alphafold
@@ Linea 62: / Linea 72: @@
 test -n "$ALPHAFOLD_CONTAINER" || exit 1
+set -x
+ALPHAFOLD_JSON_INPUT_FILE='fold_input.json'
 ALPHAFOLD_INPUT_DIR="$PWD/af_input"
 ALPHAFOLD_OUTPUT_DIR="$PWD/af_output/${SLURM_JOB_NAME}.d${SLURM_JOB_ID}"
@@ Linea 68: / Linea 81: @@
 apptainer exec \
-    --bind '/opt/hpc/system/nvidia/driver:/usr/local/nvidia/bin' \
+    --nv \
-    --bind '/opt/hpc/system/nvidia/driver:/usr/local/nvidia/lib' \
     --bind "$ALPHAFOLD_INPUT_DIR:/root/af_input" \
     --bind "$ALPHAFOLD_OUTPUT_DIR:/root/af_output" \
-    --bind "$ALPHAFOLD_MODEL_DIR:/root/models" \
-    --bind "$ALPHAFOLD_DB_DIR:/root/public_databases" \
     "$ALPHAFOLD_CONTAINER" \
     python /app/alphafold/run_alphafold.py \
-    --json_path=/root/af_input/fold_input.json \
+    --json_path="/root/af_input/$ALPHAFOLD_JSON_INPUT_FILE" \
     --model_dir=/root/models \
     --db_dir=/root/public_databases \
+    --db_dir=/root/public_databases_fallback \
     --output_dir=/root/af_output
 </code>
 The processing result will be saved in the ''af output'' folder.
+Scripts for specific NVIDIA GPU models to run ''alphafold'' on 1 node with 1 GPU (8 tasks per node):
+^ GPU  ^ Path  ^
+| NVIDIA [[https://github.com/google-deepmind/alphafold3/blob/main/docs/performance.md#nvidia-p100|P100 (12 GB)]]  | ''/hpc/share/containers/apptainer/alphafold/3.0.1/slurm-alphafold-gpu-p100.sh''  |
+| NVIDIA [[https://github.com/google-deepmind/alphafold3/blob/main/docs/performance.md#nvidia-v100|V100 (32 GB)]]  | ''/hpc/share/containers/apptainer/alphafold/3.0.1/slurm-alphafold-gpu_guest-v100_hylab.sh''  |
+| NVIDIA [[https://github.com/google-deepmind/alphafold3/blob/main/docs/performance.md#nvidia-a100-40-gb|A100 (40 GB)]]  | ''/hpc/share/containers/apptainer/alphafold/3.0.1/slurm-alphafold-gpu-a100_40g.sh''  |
+| NVIDIA [[https://github.com/google-deepmind/alphafold3/blob/main/docs/performance.md#accelerator-hardware-requirements|A100 (80 GB)]]  | ''/hpc/share/containers/apptainer/alphafold/3.0.1/slurm-alphafold-gpu-a100_80g.sh''  |
 === Documentation ===
@@ Linea 119: / Linea 138: @@
   --db_dir: Path to the directory containing the databases. Can be specified multiple times to search multiple directories in order.;
     repeat this option to specify a list of values
-    (default: "['/home/sti_calcolo/public_databases']")
+    (default: "['/hpc/home/sti_calcolo/public_databases']")
   --flash_attention_implementation: <triton|cudnn|xla>: Flash attention implementation to use. 'triton' and 'cudnn' uses a Triton and cuDNN flash attention
     implementation, respectively. The Triton kernel is fastest and has been tested more thoroughly. The Triton and cuDNN kernels require Ampere GPUs or later.
@@ Linea 147: / Linea 166: @@
     (default: '${DB_DIR}/mgy_clusters_2022_05.fa')
   --model_dir: Path to the model to use for inference.
-    (default: '/home/sti_calcolo/models')
+    (default: '/hpc/home/sti_calcolo/models')
   --nhmmer_binary_path: Path to the Nhmmer binary.
     (default: '/hmmer/bin/nhmmer')