RFdifussion and RFdiffusion2 hardware and GPU requirements

[rubenalv]

Hello, could you direct me to a resource with the minimal hardware requirements for inference and generative design? I am thinking on 200-500 aminoacid models.
In a laptop:

• how many GPU Gb and RAM would make it practical (e.g. setting up and configure, or run some experiments, and then burst to cloud or HPC)?
• laptops overheat quickly, so if you think that laptop GPU is not worth it please let me know
• for CPU-only tasks how much RAM would be reasonable? I have read 1-2Gb RAM/core, but what would be the timeframe to run the above experiments?

[HangboZhu]

Hello, I used the PDB file of the tau protein from the example on GitHub. This file contains 700 amino acids. When I ran the script named design_ppi_flexible_peptide_with_secondarystructure_specification.sh, on my local RTX 3090 graphics card, it only occupied less than 3GB of video memory. The memory usage was 16GB. After 10 iterations, it took about 10 minutes. The version of PyTorch is 1.13+cu117. If you want to conduct high-concurrency mining, it is recommended that the memory be above 128GB and the video memory of a single GPU card be above 24GB, which can meet your requirements. Of course, the larger the memory and the video memory of the GPU, the better. Hahaha, I hope this is helpful to you.

[neru1234]

It there open source link for RFdiffsuion2?