Microsoft suddenly going to town again with ARM/qualcomm surely will be good for us in terms of more arm64 software compat and Linux mainlining efforts for Qualcomm hw.
also, people who are tired of all these AI anti-features can be sure that the MNT Reform family will be free of that stuff.
@mntmn
Thanks for this! That was actually one of my main concerns when I was reading the review — all these NPUs…
It's not about moral implications of where the data comes from or about sending my data to the cloud — I just don't need any of that and now my computer is to have something that I will likely never use 😩
@m0xee @mntmn NPUs are glorified coprocessors for matrix multiplication. They can be used. or purposes other than LLMs... Not that there's much of alternative uses going on, but in theory they should have other uses.
At any rate, I'm against most AI/LLM stuff, if only for the all the cringe and hiperinflated propaganda. But there are a few useful cases (like speech to text) which I'd rather run locally without connecting to any online service, and having a NPU helps.
@mntmn @aperezdc
Sure, we can probably find uses for them, I even used neural networks for OCR myself via libfann, before machine learning and big data became buzz words even. But I'm still not sure that I might need it often enough to benefit from performance boost from having dedicated units for that 🤷
@m0xee @mntmn I am looking forward to see OpenCL running on the CM4 via Etnaviv though. (Afaik Collabora is working on that: https://www.collabora.com/news-and-blog/blog/2022/12/15/machine-learning-with-etnaviv-and-opencl/)
I don't know how far the development is as of yet, but wouldn't it be great to be able to use the hardware for tools like LibreOffice or Darktable?