Hi! I would like to use RFdiffusion for binder design, but I wonder if there is any computer specifications requirement for running this deep learning model? Thank you for your help.
Category:
Post Situation:
The hub for Rosetta modeling software
For RFDiffusion by itself, the primary hardware requirement is a CUDA-enabled GPU. Most modern NVIDIA gaming GPUs will do fine for individual inference efforts.
However, your pipeline is also likely to require AlphaFold-2 or RoseTTAFold-2 as oracles. The hardware requirements are not much more significant for single-sequence mode, but the generation of an MSA as input to these is resource-intensive. The MMSEQS2 server is one alternative if your data doesn't need to stay proprietary (and your throughput is not so high you hit the IP-based job number limits). For a private server or box capable of generating an MSA within minutes you'll need a system with 1TB of RAM and a significant number of CPU cores.
Alternatively to a hardware investment, explore the ColabDesign notebooks and run them on Google CoLab.