Gemma 3 on NVIDIA GeForce RTX 5090 · 32GB VRAM

✅ Runs comfortably — Gemma 3 27B @ Q5_K_M
Gemma 3 27B at Q5_K_M needs ~23.0 GB (weights 18.2 GB + KV 3.9 GB + overhead 930 MB @ 8K ctx) of your 28.8 GB usable VRAM — plenty of headroom. Expect ~35–55 tok/s (fast).
context @ Q5_K_M: comfortable to 8K · runs to 32K · won't fit past that on this rig
quantneedsspeed
~ Q8_032.4 GB~7–12 tok/s (usable)
Q5_K_M23.0 GB~35–55 tok/s (fast)
Q4_K_M20.1 GB~35–60 tok/s (fast)
Q3_K_M17.0 GB~45–75 tok/s (very fast)
$runlocal install gemma3:27b
⚠ NVIDIA support is best-effort in v0.1 — verify before relying on it.
runlocal verdict card
download card share on x share on reddit drop it in your model-drop post
Gemma 3: Runs comfortably
for your own README — links back here

Same model, other machines: Apple M1 · Apple M1 Pro · Apple M2 · Apple M3 Pro · Apple M2 Pro · Apple M4 Pro

Also runs on a NVIDIA GeForce RTX 5090: DeepSeek-R1 (distill) · Qwen2.5 · Qwen3 · Llama 3.1 · Llama 3.3

Best models for a NVIDIA GeForce RTX 5090 · 32GB VRAM →

Architecturally similar

These share Gemma 3's KV-cache geometry — they size memory the same way, so here's how they fit on a NVIDIA GeForce RTX 5090 · 32GB VRAM too:

modelsharesverdict here
Gemma 4 8kv/256hd ✅ Runs comfortably
Gemma 2 16kv/128hd4kv/256hd8kv/256hd ✅ Runs comfortably
Falcon3 7B 4kv/256hd ✅ Runs comfortably

Check any combo yourself: open the checker →