Intel B70 Launch - Unboxed and Tested

[long.barn4823]

@sniperwhg

Temps seem a bit on the high-side despite high airflow (78-80C), but the card doesn’t seem to be throttling. Functionally seems to work out of the box with the SYCL variant of llama.cpp, but performance seems to be on the low-side when comparing to an RTX PRO 4000. GGML_SYCL_F16=ON seems to make a big difference in the prompt processing speed. Not sure if there are any OneAPI updates or firmware updates anyone knows of that are coming soon.

try the latest llama.cpp from b8685+ ([SYCL] Add Q8_0 reorder optimization for Intel GPUs (~3x token generation speedup) by PMZFX · Pull Request #21527 · ggml-org/llama.cpp · GitHub)

I just tested it on my B60 and seems to be a fair bit better - Arc Pro B60 for Local LLMs - #13 by long.barn4823