One GPU could handle for Users-tailored pairing docking?

HBioquant / DiffBindFR

Diffusion model based protein-ligand flexible docking method

BSD 3-Clause Clear License

90 stars 11 forks source link

One GPU could handle for Users-tailored pairing docking? #14

Open rytakahashi opened 4 months ago

rytakahashi commented 4 months ago

Dear Sir,

I was just wandering about 'Users-tailored pairing docking'. As in the description, "Run the job array with total 10 jobs if you have 10000 pocket-molecule pairs in your task table, limit to 5 jobs at a time:" means one GPU can handle 5 jobs? For sure it will depend which type of GPU is, but if you could give me a rough estimation about how long will take to finish all 10000 predictions by 16 CPUs with one GPU?

Many thanks,

HBioquant commented 1 month ago

This means that the docking task of 10,000 pocket-molecule pairs is divided into 10 tasks, which means that 10 slurm compute nodes (each node configured with 16 CPUs with one GPU), but a maximum of 5 tasks are allowed at one time. Our internal time estimate is about one day on the NVIDIA Tesla V100 card to complete this job.