How to run ParaFold in the CPU mode on a machine with GPU. #45
Description
Hello,
First of all, thank you for all the work you've done on parallelizing alphafold2. I'm currently using ParaFold for MSA and template search on CPUs (with -f flag). However, when I tried export CUDA_VISIBLE_DEVICES="" and the -g flag (-g or -g False) in the run_alphafold.sh code, it still looks like the GPU is doing the work.
source $HOME/software/micromamba/etc/profile.d/micromamba.sh
micromamba activate $HOME/software/micromamba/envs/alphafold
export CUDA_VISIBLE_DEVICES=""
for file in ${PRJ_DIR}/01_fasta_dir/*.fasta; do
run_alphafold.sh -d $database -i $file \
-o ${PRJ_DIR}/02_AF2_search_output \
-m model_1 -p monomer_ptm \
-f # with "-g" or "-g False" (I'm not sure which one is the correct usage)
done
And I checked the usage of GPU.
$ nvidia-smi
Thu Apr 18 13:15:18 2024
+---------------------------------------------------------------------------------------+
| NVIDIA-SMI 535.113.01 Driver Version: 535.113.01 CUDA Version: 12.2 |
|-----------------------------------------+----------------------+----------------------+
| GPU Name Persistence-M | Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap | Memory-Usage | GPU-Util Compute M. |
| | | MIG M. |
|=========================================+======================+======================|
| 0 NVIDIA GeForce RTX 2080 Ti Off | 00000000:17:00.0 Off | N/A |
| 36% 34C P8 7W / 250W | 543MiB / 11264MiB | 0% Default |
| | | N/A |
+-----------------------------------------+----------------------+----------------------+
| 1 NVIDIA GeForce RTX 2080 Ti Off | 00000000:25:00.0 Off | N/A |
| 24% 30C P0 21W / 250W | 0MiB / 11264MiB | 0% Default |
| | | N/A |
+-----------------------------------------+----------------------+----------------------+
+---------------------------------------------------------------------------------------+
| Processes: |
| GPU GI CI PID Type Process name GPU Memory |
| ID ID Usage |
|=======================================================================================|
| 0 N/A N/A 216735 C python 156MiB |
+---------------------------------------------------------------------------------------+
And I checked the usage of CPUs. It seems only 5 CPUs was running.
$ cpu4user.sh
%cpu %mem user
======================
7361.8 7.5 fengxiao
527.4 0 liyulong # this is me
9.9 0 root
$ top -u liyulong
PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
220070 liyulong 25 5 721252 111572 1284 R 531.4 0.0 7:28.39 jackhmmer
185580 liyulong 20 0 923760 51784 20584 S 3.2 0.0 0:58.62 node
220092 liyulong 20 0 163708 4160 1588 R 2.6 0.0 0:01.12 top
188285 liyulong 20 0 130768 6624 4492 S 0.6 0.0 2:01.42 wget
185455 liyulong 20 0 1026756 123144 23964 S 0.3 0.0 0:55.26 node
218404 liyulong 20 0 988700 81968 23308 S 0.3 0.0 0:06.10 node
218686 liyulong 25 5 123.3g 1.9g 311612 S 0.3 0.1 0:23.86 run_alphafold.p
160206 liyulong 20 0 130244 1800 944 S 0.0 0.0 0:00.35 screen
160207 liyulong 20 0 116764 3364 1680 S 0.0 0.0 0:00.11 bash
160337 liyulong 20 0 116764 3344 1660 S 0.0 0.0 0:00.10 bash
160440 liyulong 20 0 127572 5020 2556 S 0.0 0.0 0:03.35 zsh
185443 liyulong 20 0 113184 1408 1212 S 0.0 0.0 0:00.00 sh
185563 liyulong 20 0 728112 34716 19720 S 0.0 0.0 1:05.09 node
185623 liyulong 20 0 128236 5812 2752 S 0.0 0.0 0:18.03 zsh
188637 liyulong 20 0 728112 35116 19756 S 0.0 0.0 1:07.38 node
207806 liyulong 20 0 127252 4560 2460 S 0.0 0.0 0:00.25 zsh
207961 liyulong 20 0 127660 5020 2512 S 0.0 0.0 0:00.48 zsh
214561 liyulong 25 5 113312 1632 1328 S 0.0 0.0 0:00.00 bash
218318 liyulong 20 0 160908 2488 1092 S 0.0 0.0 0:00.54 sshd
218319 liyulong 20 0 113316 1724 1416 S 0.0 0.0 0:00.08 bash
218455 liyulong 20 0 728112 33044 19548 S 0.0 0.0 1:01.50 node
218676 liyulong 25 5 113188 1524 1264 S 0.0 0.0 0:00.00 run_alphafold.s
220079 liyulong 20 0 107956 356 280 S 0.0 0.0 0:00.00 sleep
220158 liyulong 20 0 113184 1484 1264 S 0.0 0.0 0:00.00 cpuUsage.sh
220165 liyulong 20 0 107956 356 280 S 0.0 0.0 0:00.00 sleep
275815 liyulong 25 5 227996 19272 3352 S 0.0 0.0 0:36.77 pyth
According to the Quick-start of ParaFold (https://parafold.sjtu.edu.cn/docs/quick-start/). It should use 20 CPUs for the blasting.
在三个数据库的多序列比对计算中，两个 jackHMMER 各需要 8 核 CPU 达到最高性能，
HHblits 需要 4 核 CPU 达到最高性能，所以在并行加速过程中，
我们给整个任务提供足够的 20 核 CPU 以保证计算效率。
So I'm calling ParaFold in a wrong way?
Looking forward to your reply. Thank you.
Best regards,
Yulong Li