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

Phi-3-vision-128k-instruct

by microsoft

4.1B params · text-generation · 971 likes · 91.5k downloads

Phi-3-vision-128k-instruct is a 4.1B parameter model. At Q4 quantization it requires 2GB of VRAM. It runs comfortably on GeForce RTX 4090 (316 tok/s), GeForce RTX 5090 (474 tok/s), M4 Max 128GB (115 tok/s).

GPU compatibility

GPUVRAMQ4 DecodeVerdict
GeForce RTX 409024GB316 tok/scomfortable
GeForce RTX 509032GB474 tok/scomfortable
M4 Max 128GB128GB115 tok/scomfortable
M4 Pro 48GB48GB57 tok/scomfortable
M4 Pro 24GB24GB57 tok/scomfortable
A100 PCIe 80 GB80GB580 tok/scomfortable
H100 SXM5 80 GB80GB1166 tok/scomfortable
GeForce RTX 309024GB279 tok/scomfortable
Radeon RX 7900 XTX24GB231 tok/scomfortable
GeForce RTX 408016GB224 tok/scomfortable
Install CLI [email protected] Raw data · MIT · API data: live · HW/Cloud data: curated 2026-02-23 · v0.6.0