can it run
Only with compromises. 24GB is below the 25GB working set — expect a lower quant, offloading, or heavy swap. Possible, not pleasant.
weights 20GB + KV cache 1GB at 4k context vs ~22GB usable → tight. 21.0 of 22 GB usable — keep context modest
computed roofline: 273 GB/s × 0.25–0.65 efficiency window / 20.0GB (Q4_K_M) — bands, never points.
| model file | 20GB (GGUF Q4_K_M) |
| minimum memory | 25GB working set |
| this machine | 24GB unified LPDDR5X (18-core CPU / 5,120-core GPU base config) |
| params | distill 32B |
| license | mit |
| price | $2599.99 |
easy run: ollama pull deepseek-r1:32b
runs with: ollama · llama.cpp · koboldcpp · lm studio
→ Mac mini M5 Pro (64GB) — cheapest machine that runs it ($1669)
→ ASUS ExpertCenter Pro ET900N G3 (GB300) — this model on the biggest machine
→ rtx-5090 — the next step up from this machine's GPU