NVIDIA RTX 4000 Ada Generation — Benchmarks & Specs

Bottom line: how fast is the RTX 4000 Ada Generation?

SpecPicks median: 58.6 tokens/sec across 6 cited single-user runs of 7-9B dense models at 4-bit (Q4) under llama.cpp (range 44.4–61.8 tok/s), from KnightLi, llama.cpp GitHub Discussion #15013, GPU-Benchmarks-on-LLM-Inference (2 runs), LocalScore (2 runs). At the $1,995.00 Amazon price of the linked Amazon listing (checked 2026-10-20 16:16 UTC; prices change), that is a SpecPicks Value Index of 29.4 tok/s per $1,000. At 4K Ultra, the RTX 4000 Ada Generation averages 30 fps in Unreal Engine 4.26 (Audi Car Configurator), per Develop3D. For local LLM inference it generates 24.7 tokens/sec running qwen2.5:14b at q4_K_M under llama.cpp, per LocalScore. In Geekbench 5 OpenCL it scores 144,947 points, per technical.city. Its 20 GB of VRAM is the binding constraint for local inference: the largest model on file at 4-bit quantization on this card is qwen2.5:32b (32B parameters, IQ4_XS), generating 16.9 tokens/sec under llama.cpp, per Hardware Corner. That run's source does not report a context length; the Will-It-Run matrix sizes each model at its listed quantization with an ~8K context, and by that measure Gemma 2 27B at Q4_K_M is the largest that fits in 20 GB.

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.