local intelligence

can it run

Can the RTX 5090 (32GB) run DEEPSEEK-R1 distill 32B?

TIGHT

Yes, but close to the edge. 32GB against a 25GB working set: it runs, with little room for long context windows or other processes.

~49–72 tok/s

weights 20GB + KV cache 1GB at 4k context vs ~32GB usable → full. 21.0 GB needed of 32 GB usable — headroom for context

computed roofline: 1792 GB/s × 0.55–0.80 efficiency window / 20.0GB (Q4_K_M) — bands, never points.

Community check: ~67 tok/s · 30-33B Q4 · 13 runs · inside our computed band — localmaxxing.com

explore the full catalog84 machines indexed · live prices · what each one can run

Specs

model file20GB (GGUF Q4_K_M)
minimum memory25GB working set
this machine32GB GDDR7
paramsdistill 32B
licensemit
price$5000

Run it

model files: hugging face ↗

easy run: ollama pull deepseek-r1:32b

runs with: ollama · llama.cpp · koboldcpp · lm studio

Alternatives

Mac mini M5 Pro (64GB) — cheapest machine that runs it ($1669)

ASUS ExpertCenter Pro ET900N G3 (GB300) — this model on the biggest machine

every model the RTX 5090 (32GB) can run — full list

RTX PRO 6000 Blackwell (96GB) — upgrade path machine

Buy

B&H Photo ↗ · Amazon ↗ · Amazon ↗

2026-09-17 · ← all models on the RTX 5090 (32GB) · ← full hardware catalog · prices verified at source, may drift