can it run
26.5 GB vs 24 GB usable — partial CPU offload, expect large speed loss With a q8 (quantized) KV cache it comes to the edge (26.2GB) — possible with modest context, at some quality cost.
24GB · Ampere · model file 26GB · 8x7B MoE
computed roofline: 300 GB/s × 0.55–0.88 efficiency window / 3.9GB (Q4_K_M) — bands, never points.
MoE: file is 26GB (that decides fit) but only ~1.05B of experts activate per token (~3.9GB streamed) — band below is the active-expert estimate, optimistic; community MoE rows run several× the naive bandwidth math
→ Mac mini M5 Pro (64GB) — cheapest catalog machine that runs mixtral 8x7B MoE fully ($1669)
→ RTX 4090 (24GB, used) — upgrade path machine
→ every model the NVIDIA L4 can run — full list
→explore the full catalog84 machines indexed · live prices · what each one can run