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

Llama-2-7b-hf

by meta-llama

6.7B params · text-generation · 2.3k likes · 686.3k downloads

Llama-2-7b-hf is a 6.7B parameter model. At Q4 quantization it requires 3GB of VRAM. It runs comfortably on GeForce RTX 4090 (194 tok/s), GeForce RTX 5090 (292 tok/s), M4 Max 128GB (71 tok/s).

GPU compatibility

GPUVRAMQ4 DecodeVerdict
GeForce RTX 409024GB194 tok/scomfortable
GeForce RTX 509032GB292 tok/scomfortable
M4 Max 128GB128GB71 tok/scomfortable
M4 Pro 48GB48GB35 tok/scomfortable
M4 Pro 24GB24GB35 tok/scomfortable
A100 PCIe 80 GB80GB356 tok/scomfortable
H100 SXM5 80 GB80GB718 tok/scomfortable
GeForce RTX 309024GB172 tok/scomfortable
Radeon RX 7900 XTX24GB142 tok/scomfortable
GeForce RTX 408016GB138 tok/scomfortable
Install CLI [email protected] Raw data · MIT · API data: live · HW/Cloud data: curated 2026-02-23 · v0.6.0