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

bitnet-b1.58-2B-4T

by microsoft

3.0B params · text-generation · 1.3k likes · 16.5k downloads

bitnet-b1.58-2B-4T is a 3.0B parameter model. At Q4 quantization it requires 2GB of VRAM. It runs comfortably on GeForce RTX 4090 (432 tok/s), GeForce RTX 5090 (648 tok/s), M4 Max 128GB (158 tok/s).

GPU compatibility

GPUVRAMQ4 DecodeVerdict
GeForce RTX 409024GB432 tok/scomfortable
GeForce RTX 509032GB648 tok/scomfortable
M4 Max 128GB128GB158 tok/scomfortable
M4 Pro 48GB48GB79 tok/scomfortable
M4 Pro 24GB24GB79 tok/scomfortable
A100 PCIe 80 GB80GB792 tok/scomfortable
H100 SXM5 80 GB80GB1593 tok/scomfortable
GeForce RTX 309024GB382 tok/scomfortable
Radeon RX 7900 XTX24GB316 tok/scomfortable
GeForce RTX 408016GB306 tok/scomfortable
Install CLI [email protected] Raw data · MIT · API data: live · HW/Cloud data: curated 2026-02-23 · v0.6.0