153 tok/s on 1x AMD Radeon R9700 running Qwen3.8 27b NVFP4, 470 tok/s @ 8 conc requests, Prefill @ 3,619 tok/s
retained text
People kept commenting and asking about single AMD 1xR9700 cards in the comments and discord. Well, I finally had time to do some optimizations for 1xR9700 owners and performance has doubled across the board. You can see the results in [BetterBench ](https://github.com/GGZ14/BetterBench)above if you like visuals or below if you're more into text.
These results were measured running [Unsloth's Qwen3.8 27b NVFP4](https://huggingface.co/unsloth/Qwen3.8-27B-NVFP4).
Decode
┌───────────────┬───────────────┬──────────────────┐
│ category │ update p50 ms │ decode t/s (med) │
├───────────────┼───────────────┼──────────────────┤
│ chat │ 42.3 │ 67.1 │
├───────────────┼───────────────┼──────────────────┤
│ code │ 42.5 │ 120.5 │
├───────────────┼───────────────┼──────────────────┤
│ file_edit │ 42.5 │ 138.0 │
├───────────────┼───────────────┼──────────────────┤
│ json │ 42.4 │ 153.1 │
├───────────────┼───────────────┼──────────────────┤
│ math │ 42.5 │ 140.0 │
├───────────────┼───────────────┼──────────────────┤
│ prose │ 42.3 │ 69.2 │
├───────────────┼───────────────┼──────────────────┤
│ reasoning │ 34.3 │ 123.9 │
├───────────────┼───────────────┼──────────────────┤
│ summarization │ 34.2 │ 141.7 │
└───────────────┴───────────────┴──────────────────┘
Prefill
┌───────────────┬───────────────┐
│ prefill depth │ pp tok/s │
├───────────────┼───────────────|
│ 2000 │ 3552 │
├───────────────┼───────────────|
│ 8000 │ 3536 │
├───────────────┼───────────────|
│ 16000 │ 3619 │
├───────────────┼───────────────|
│ 32000 │ 3437 │
├───────────────┼───────────────|
│ 64000 │ 3192 │
├───────────────┼───────────────|
Concurrency
┌───────────────┬───────────────┐
│ level │ tok/s │
├───────────────┼───────────────|
│ 1 │ 120 │
├───────────────┼───────────────|
│ 2 │ 215 │
├───────────────┼───────────────|
│ 4 │ 322 │
├───────────────┼───────────────|
│ 8 │ 471 │
├───────────────┼───────────────|
**Links** (Both repo's updated as some users wanted Github)
[https://codeberg.org/ggz14/radiance-vllm-mxfp4](https://codeberg.org/ggz14/radiance-vllm-mxfp4)
[https://github.com/GGZ14/vllm-mxfp4](https://github.com/GGZ14/vllm-mxfp4)
[https://x.com/bkuyper](https://x.com/bkuyper)
**I hope you single R9700 card owners enjoy this release!**document fetched 2026-09-18 00:43 · canonical https://reddit.com/r/LocalLLaMA/comments/1wiws8e/153_toks_on_1x_amd_radeon_r9700_running_qwen38