RTX 3090 VRAM Benchmarks for Local LLMs

The RTX 3090 has 24GB VRAM. This page tracks model tags with measured local runs on RTX 3090 so you can compare tokens/s, latency, and model fit instead of relying only on baseline estimates.

Measured model tags

Model Tag Tokens/s Latency (ms) Test time Detail
gpt-oss:20b (tag-level reference; exact profile unverified) gpt-oss:20b 156.1 1524 2026-04-29T05:39:58Z -
qwen3:8b (tag-level reference; exact profile unverified) qwen3:8b 109.6 1442 2026-10-07T08:35:40Z -
qwen2.5:14b (tag-level reference; exact profile unverified) qwen2.5:14b 84.0 946 2026-04-29T05:39:58Z -
ministral-3:14b (tag-level reference; exact profile unverified) ministral-3:14b 70.0 2113 2026-10-07T08:35:40Z -
deepseek-r1:14b (tag-level reference; exact profile unverified) deepseek-r1:14b 66.5 2189 2026-10-07T08:35:40Z -
qwen3-coder:30b (tag-level reference; exact profile unverified) qwen3-coder:30b 61.8 1575 2026-10-07T08:35:40Z -
nemotron-3-nano:30b (tag-level reference; exact profile unverified) nemotron-3-nano:30b 57.0 2468 2026-04-01T11:53:50Z -
mistral-small:22b (tag-level reference; exact profile unverified) mistral-small:22b 48.9 2087 2026-10-07T08:35:40Z -
translategemma:27b (tag-level reference; exact profile unverified) translategemma:27b 41.3 3142 2026-04-01T11:53:50Z -
qwen3.6:35b (tag-level reference; exact profile unverified) qwen3.6:35b 36.6 3162 2026-10-07T08:35:40Z -
gemma3:27b (tag-level reference; exact profile unverified) gemma3:27b 36.6 3006 2026-10-07T08:35:40Z -
qwq:32b (tag-level reference; exact profile unverified) qwq:32b 31.1 3409 2026-10-07T08:35:40Z -
qwen2.5-coder:32b (tag-level reference; exact profile unverified) qwen2.5-coder:32b 17.3 6086 2026-10-07T08:35:40Z -
llama4:16x17b (tag-level reference; exact profile unverified) llama4:16x17b 16.8 4319 2026-09-16T07:39:43Z -
glm-4.7-flash:bf16 (tag-level reference; exact profile unverified) glm-4.7-flash:bf16 11.2 9291 2026-03-04T09:01:38Z -
qwen3.5:122b (tag-level reference; exact profile unverified) qwen3.5:122b 4.9 11915 2026-02-26T19:19:16Z -
qwen3.5:35b (tag-level reference; exact profile unverified) qwen3.5:35b 2.8 37551 2026-06-24T06:22:37Z -
llama3.3:70b (tag-level reference; exact profile unverified) llama3.3:70b 1.5 37261 2026-06-24T06:22:37Z -

Validation notes

Open benchmark changelog Open model catalog Estimate VRAM