> ## Content Index
> Fetch the complete content index at: https://globalfeed.ai/llms.txt
> Use this file to discover other available public pages before exploring further.

# Inference speed doubles on a single AMD Radeon R9700
- URL: https://globalfeed.ai/en/inference-speed-doubles-on-a-single-amd-radeon-r9700/
- Published: 2026-09-17T18:49:45.000Z
- Updated: 2026-09-17T18:49:45.000Z
- Description: Benchmarks shared on LocalLLaMA report 153 tok/s decode on one card and 470 tok/s across eight concurrent requests after optimization work.
- Author: GlobalFeed Editor
- Tags: AMD, Qwen3, local inference, x-AMD, dil-en, denetim-gerekli, video

A set of benchmarks posted to the LocalLLaMA community reports that inference performance on a single AMD Radeon R9700 doubled across the board after targeted optimization work. The tests were run with Unsloth's Qwen3.8 27b NVFP4.

The author says questions about single-card R9700 setups kept coming up in comments and on Discord, which prompted the work on that specific configuration.

## The numbers

Median decode throughput by category:

- json: 153.1 tok/s
- summarization: 141.7 tok/s
- math: 140.0 tok/s
- file\_edit: 138.0 tok/s
- reasoning: 123.9 tok/s
- code: 120.5 tok/s
- prose: 69.2 tok/s
- chat: 67.1 tok/s

Update latency (p50) sits around 42 ms in most categories and drops to roughly 34 ms for reasoning and summarization.

Two further figures come from the post title: 470 tok/s aggregate across eight concurrent requests, and 3,619 tok/s on prefill.

## Why it matters

Most people running models locally default to NVIDIA hardware, and concrete single-card measurements on the AMD side are shared less often. These results give a direct data point on how fast a 27-billion-parameter model in NVFP4 quantization runs on one R9700.

The figures have not been independently verified; they come from the poster's own test rig. Full tables, prefill details and the setup discussion are in the [original thread](https://www.reddit.com/r/LocalLLaMA/comments/1wiws8e/153%5Ftoks%5Fon%5F1x%5Famd%5Fradeon%5Fr9700%5Frunning%5Fqwen38/?ref=globalfeed.ai).