NVIDIA GeForce RTX 4090 — Benchmarks & Specs

Bottom line: how fast is the NVIDIA GeForce RTX 4090?

SpecPicks median: 131 tokens/sec across 7 cited single-user runs of 7-9B dense models at 4-bit (Q4) under llama.cpp (range 113–189 tok/s), from llama.cpp GitHub Discussion #15013, llama.cpp GitHub, Hardware Corner, GPU-Benchmarks-on-LLM-Inference (2 runs), MyAIHardware, Awesome Agents LLM Leaderboard. At the $4,479.95 Amazon price of the cheapest linked listing (checked 2026-10-21 04:14 UTC; prices change), that is a SpecPicks Value Index of 29.1 tok/s per $1,000. At 1440p Ultra, the NVIDIA GeForce RTX 4090 averages 121 fps in ComputerBase raster performance rating (geomean) with a 95 fps 1% low, per ComputerBase. For local LLM inference it generates 38.9 tokens/sec running qwen3:32b at Q4_K_XL under llama.cpp, per Hardware Corner. In Deep Learning Benchmark (ResNet-50) it scores 128,000 images/sec, per Puget Systems. Its 24 GB of VRAM is the binding constraint for local inference: the qwen3:32b run above (32B parameters) is the largest model on file at 4-bit quantization on this card. A second published run of the same setup measured 139.7 tokens/sec, per Puget Systems, so expect 38.9-139.7 tokens/sec.

Every figure above is a row in the tables below, and each row links out to the review or public benchmark database the number was taken from; the SpecPicks median is computed here from the runs it names, each a row in the AI table, and the per-$1,000 figure divides it by the dated listing price. SpecPicks aggregates published measurements; it does not report first-party benchmark runs.