vram.run Models Hardware Providers Cloud State of Inference
API provider data is live · Hardware & cloud pricing curated 2026-02-23

SmolLM3-3B

by HuggingFaceTB

3.1B params · text-generation · 906 likes · 137.5k downloads

SmolLM3-3B is a 3.1B parameter model. At Q4 quantization it requires 2GB of VRAM. It runs comfortably on GeForce RTX 4090 (426 tok/s), GeForce RTX 5090 (640 tok/s), M4 Max 128GB (156 tok/s).

GPU compatibility

GPUVRAMQ4 DecodeVerdict
GeForce RTX 409024GB426 tok/scomfortable
GeForce RTX 509032GB640 tok/scomfortable
M4 Max 128GB128GB156 tok/scomfortable
M4 Pro 48GB48GB78 tok/scomfortable
M4 Pro 24GB24GB78 tok/scomfortable
A100 PCIe 80 GB80GB782 tok/scomfortable
H100 SXM5 80 GB80GB1573 tok/scomfortable
GeForce RTX 309024GB377 tok/scomfortable
Radeon RX 7900 XTX24GB312 tok/scomfortable
GeForce RTX 408016GB303 tok/scomfortable
Install CLI [email protected] Raw data · MIT · API data: live · HW/Cloud data: curated 2026-02-23 · v0.6.0