local intelligence

can it run

Can the RTX 3090 (24GB, used) run PHI4 mini 3.8B?

YES

Comfortably. 24GB of memory against a 4GB working set leaves headroom for context, the OS, and anything else you're running.

~170–280 tok/s

weights 2.5GB + KV cache 0.5GB at 4k context vs ~24GB usable → full. 3.0 GB needed of 24 GB usable — headroom for context

computed roofline: 936 GB/s × 0.46–0.76 efficiency window / 2.5GB (Q4_K_M) — bands, never points. Small models (under ~4.5B) run BELOW these bands — kernel-launch overhead keeps them from saturating bandwidth (measured against community medians, 09-06).

Community check: ~130 tok/s · 3-4B Q4 · 11 runs · just below our band (22% under the low edge) — localmaxxing.com

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

Specs

model file2.5GB (GGUF Q4_K_M)
minimum memory4GB working set
this machine24GB GDDR6X, 936 GB/s
paramsmini 3.8B
licensemit
price$1100

Run it

model files: hugging face ↗

easy run: ollama pull phi4-mini

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

Alternatives

no alternatives — every machine in the catalog can run this one.

rtx-5090 — the next step up from this machine's GPU

Buy

NVIDIA (used market) ↗ · Amazon ↗

2026-09-17 · ← full hardware catalog · prices verified at source, may drift