A critical lesson that DeepSeek tells us is that competition is good and necessary for innovation. Sanction or export control doesn’t work well.

In a real time crunch writing these, sorry for any typos or anything isn’t well explained.

Performant Automatic BLAS Offload for Quantum Chemistry/Physics on Cache-Coherent Unified Memory architecture This work is done on NVIDIA Grace-Hopper when I'm at TACC. Today is my last day at TACC, and I'm joining AMD next week (because AMD yes!) .

Wrote a paper draft and linkedin post about doing performant automatic BLAS offload for quantum chemistry, the work was done on NVIDIA Grace-Hopper, not sure how this will go after I join AMD. If it can’t be officially published, I hope this post and draft can be helpful to some scientists.

I’ll join AMD next week, not much HPC related work initially, but hopefully can reconnect with HPC & quantum chemistry later.

Hi there.