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

MiniCPM-SALA

by openbmb

9.5B params · text-generation · 682 likes · 5.8k downloads

MiniCPM-SALA is a 9.5B parameter model. At Q4 quantization it requires 5GB of VRAM. It runs comfortably on GeForce RTX 4090 (138 tok/s), GeForce RTX 5090 (207 tok/s), M4 Max 128GB (50 tok/s).

GPU compatibility

GPUVRAMQ4 DecodeVerdict
GeForce RTX 409024GB138 tok/scomfortable
GeForce RTX 509032GB207 tok/scomfortable
M4 Max 128GB128GB50 tok/scomfortable
M4 Pro 48GB48GB25 tok/stight
M4 Pro 24GB24GB25 tok/stight
A100 PCIe 80 GB80GB253 tok/scomfortable
H100 SXM5 80 GB80GB510 tok/scomfortable
GeForce RTX 309024GB122 tok/scomfortable
Radeon RX 7900 XTX24GB101 tok/scomfortable
GeForce RTX 408016GB98 tok/scomfortable
Install CLI [email protected] Raw data · MIT · API data: live · HW/Cloud data: curated 2026-02-23 · v0.6.0