Benchmark result
Dirk-Qwen3.8-27B-GSQ-RCO-IQ3_S on AMD Radeon RX 5600 OEM/5600 XT / 5700/5700 XT +1 more — 21.4 tok/s
Measured with llama.cpp b10770-bf300c61f on September 4, 2026.
What the model built
Open full screen →The coding scenario asks for a playable game in a single HTML file. This is exactly what the model returned, unedited; the frame only adds a line that tells this page how tall it is.
How this run compares
1st fastest of 2 runs of this model on AMD Radeon RX 5600 OEM/5600 XT / 5700/5700 XT +1 more · Multi-GPU · median 21.4 tok/s.
sort
- 121.4tok/s85.8llama.cppthis run
- 221.4tok/s83.2llama.cpp
Reproduce this run
toolllama.cpp b10770-bf300c61f
modelDirk-Qwen3.8-27B-GSQ-RCO-IQ3_S
context145,152
thinkingon
samplingtemperature 0.699999988079071 · top_p 0.949999988079071 · top_k 20 · min_p 0 · presence_penalty 0 · repeat_penalty 1
clientv0.4.55+97
Set those in llama.cpp, then:
llm-benchmark benchmark --model "Dirk-Qwen3.8-27B-GSQ-RCO-IQ3_S" --tool "llama.cpp"The client prompts for context length and thinking mode, and for the KV cache dtype on Unsloth Studio. Sampling is left at the model default — the values above are what the tool reported using, not overrides the client sent.