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

phi-2

by microsoft

2.8B params · text-generation · 3.4k likes · 1.7M downloads

phi-2 is a 2.8B parameter model. At Q4 quantization it requires 1GB of VRAM. It runs comfortably on GeForce RTX 4090 (472 tok/s), GeForce RTX 5090 (708 tok/s), M4 Max 128GB (172 tok/s).

GPU compatibility

GPUVRAMQ4 DecodeVerdict
GeForce RTX 409024GB472 tok/scomfortable
GeForce RTX 509032GB708 tok/scomfortable
M4 Max 128GB128GB172 tok/scomfortable
M4 Pro 48GB48GB86 tok/scomfortable
M4 Pro 24GB24GB86 tok/scomfortable
A100 PCIe 80 GB80GB865 tok/scomfortable
H100 SXM5 80 GB80GB1740 tok/scomfortable
GeForce RTX 309024GB417 tok/scomfortable
Radeon RX 7900 XTX24GB345 tok/scomfortable
GeForce RTX 408016GB335 tok/scomfortable
Install CLI [email protected] Raw data · MIT · API data: live · HW/Cloud data: curated 2026-02-23 · v0.6.0