Note: this is probably not an easy port, but the strategic value is immense.
This is the link: https://github.com/ROCm/ROCm
Motivation, benfits for freeBSD:
Nowadays, a big chunk of servers are doing machine learning compute. More so after LLMs (chatGPT and friends) came into the scene. Those run on nvidia's CUDA. CUDA is not available for freebsd, period. That blocks its usage for machine learning. ... or did. AMD, after ignoring that market for many years, finally woke up and released a library for machine learning (ROCm) that is not terrible as it used to be, but actually good (since v. 6.0).
Not to mention AMD cards are more performant at the same price, compared to Nvidia. And that nvidia had a monopoly.
This is why ROCm is important, and why freebsd supporting it would have a big impact. Nvidia on linux (binary blob) is utterly terrible, devs and admins would jump to freebsd ASAP once ROCm works there.
This is the link: https://github.com/ROCm/ROCm
Motivation, benfits for freeBSD:
Nowadays, a big chunk of servers are doing machine learning compute. More so after LLMs (chatGPT and friends) came into the scene. Those run on nvidia's CUDA. CUDA is not available for freebsd, period. That blocks its usage for machine learning. ... or did. AMD, after ignoring that market for many years, finally woke up and released a library for machine learning (ROCm) that is not terrible as it used to be, but actually good (since v. 6.0).
Not to mention AMD cards are more performant at the same price, compared to Nvidia. And that nvidia had a monopoly.
This is why ROCm is important, and why freebsd supporting it would have a big impact. Nvidia on linux (binary blob) is utterly terrible, devs and admins would jump to freebsd ASAP once ROCm works there.