Hi ！
First of all, thank you for open-sourcing Nesso-1 and publishing such a detailed technical report!
I'm very excited to try Nesso-1 in my own work. I noticed from the paper that the inference benchmarks were conducted on a single NVIDIA H100 , However, the paper doesn't seem to specify the minimum GPU memory (VRAM) required to run Nesso-1. Since H100s are not always accessible , I would like to ask:
- What is the minimum GPU memory required to run Nesso-1 for inference on a single protein-ligand pair? For example, would a consumer-grade GPU (e.g., RTX 4090 with 24GB) be sufficient?
- How does VRAM consumption scale with protein size? The paper mentions benchmarking on targets ranging from ~300 to ~1000 residues — is there a rough estimate of memory usage at each scale?
- During fine-tuning on continuous data streams (mentioned as one of Nesso-1's advantages), what is the recommended minimum GPU memory?
Understanding these requirements would be very helpful for planning our experiments and for many others in the community who may not have access to H100-class hardware.
Thanks again for your excellent work, and looking forward to your response!
Hi ！
First of all, thank you for open-sourcing Nesso-1 and publishing such a detailed technical report!
I'm very excited to try Nesso-1 in my own work. I noticed from the paper that the inference benchmarks were conducted on a single NVIDIA H100 , However, the paper doesn't seem to specify the minimum GPU memory (VRAM) required to run Nesso-1. Since H100s are not always accessible , I would like to ask:
Understanding these requirements would be very helpful for planning our experiments and for many others in the community who may not have access to H100-class hardware.
Thanks again for your excellent work, and looking forward to your response!