6,474 karma · joined January 30, 2016
https://scholar.google.com/citations?user=BrwGRLwAAAAJ&hl=en
https://www.maxg.io
https://blog.maxg.io
All posts are my own personal opinion
[ my public key: https://keybase.io/mtg; my proof: https://keybase.io/mtg/sigs/4PwbkRLWa26ARSIXs2TCRAFZaUTDQ2lsT8JGzJD72QM ]
(I don't think you can do this but I'm not a physicist...)
On the other hand, the lack of broad HW support means that my FreeBSD server burned 2x more power at low to mid usage levels than the same HW running Proxmox.
Edit - I just can’t read, original article was llama.c
Gotta push my powerpc llama.cpp fork now for sure!
I’ve been getting llama.cpp going on various weird, old systems as I can and qwen3.c where llama.cpp has no hope. So far, I’ve tried various sparc generations (IIi, IIIi, Fujitsu M10, and an Oracle M7), a C8900 PA-RISC, some riscv boards, an Alpha 21264, POWER 9, and many X86 and ARM systems of course.