can it run
Only with compromises. 48GB is below the 54GB working set — expect a lower quant, offloading, or heavy swap. Possible, not pleasant.
weights 43GB + KV cache 1.25GB at 4k context vs ~48GB usable → tight. 44.2 of 48 GB usable — keep context modest
computed roofline: 864 GB/s × 0.46–0.76 efficiency window / 43.0GB (Q4_K_M) — bands, never points.
| model file | 43GB (GGUF Q4_K_M) |
| minimum memory | 54GB working set |
| this machine | 48GB GDDR6, 864 GB/s |
| params | distill 70B |
| license | mit |
| price | $7450 |
easy run: ollama pull deepseek-r1:70b
runs with: ollama · llama.cpp · koboldcpp · lm studio
→ NIMO AI Mini PC (128GB) — cheapest machine that runs it ($1999)
→ ASUS ExpertCenter Pro ET900N G3 (GB300) — this model on the biggest machine
→ rtx-pro-6000 — the next step up from this machine's GPU
NVIDIA (used market) ↗ · Amazon ↗