▲ local intelligence

can it run

Can the Intel Arc A770 16GB run qwen3-coder 30B-A3B MoE?

OFFLOAD PARTIAL

19.0 GB vs 16 GB usable — partial CPU offload, expect large speed loss

16GB · Alchemist · model file 18.6GB · 30B-A3B MoE

~60–150 tok/s (active-expert estimate)

computed roofline: 560 GB/s × 0.20–0.50 efficiency window / 1.9GB (Q4_K_M) — bands, never points.

MoE: file is 18.6GB (that decides fit) but only ~3B of experts activate per token (~1.9GB streamed) — band below is the active-expert estimate, optimistic; community MoE rows run several× the naive bandwidth math

Run qwen3-coder 30B-A3B MoE on instead

→ Tesla V100 32GB (used) — cheapest catalog machine that runs qwen3-coder 30B-A3B MoE fully ($700)

→ arc-b580 — the next step up from Intel Arc A770 16GB

→ every model the Intel Arc A770 16GB can run — full list

→explore the full catalog91 machines indexed · live prices · what each one can run
2026-10-01 · ← all models on the Intel Arc A770 16GB · fit = weights + 4k KV vs usable memory