NVIDIA GeForce RTX 3070 — Benchmarks & Specs

Bottom line: how fast is the NVIDIA GeForce RTX 3070?

SpecPicks median: 59.6 tokens/sec across 11 cited single-user runs of 7-9B dense models at 4-bit (Q4) under llama.cpp (range 46.0–78.7 tok/s), from llama.cpp GitHub Discussions, GPU-Benchmarks-on-LLM-Inference (4 runs), LocalScore.ai, LocalLLM.in (4 runs), InferenceRig. At the $469.99 Amazon price of the cheapest linked listing (checked 2026-10-14 04:13 UTC; prices change), that is 126.8 tok/s per $1,000. At 1440p Ultra, the NVIDIA GeForce RTX 3070 averages 61 fps in Crimson Desert, per Gamers Nexus. For local LLM inference it generates 59.4 tokens/sec running qwen3.5:9b at q4_K_M under ollama, per Patrick Hughes. In PassMark G3D Mark it scores 22,123 pts, per PassMark. Its 8 GB of VRAM is the binding constraint for local inference: the qwen3.5:9b run above (9B parameters) is the largest model on file at 4-bit quantization on this card.

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.