[emanuele-em]

This is seriously impressive for a single project. Rust + CUDA kernels for MD, fallback to SIMD thread pools, pharmacokinetics inference, docking, and it compiles to a standalone binary? How does the MD throughput compare to GROMACS on equivalent hardware?

[the__alchemist]

Ty! MD thoroughput is same OOM for both backends, but I haven't done detailed benchmarks across various molecule arrangements and config. SPME dominates the calculation for the built-in backend, with short-range Coulomb and LJ taking second place. MD accuracy is... not on the same level unfortunately, at least for proteins. Now that the backends are interchangeable with the similar config and FF params, I can hopefully narrow down what's going wrong. (I suspect SPME, again...)