Qwen3.8-Flash-Next speed on every GPU
5 measured runs across 3 GPUs, fastest first.
| GPU | Quant | Engine | tok/s | Roofline | Share |
|---|---|---|---|---|---|
| RTX 5090 | GSQ-RCO IQ1_M | llama.cpp + mods | 226.1 | 528.5 | 43% |
| H200 | UD-Q4_K_XL | llama.cpp | 100.4 | 709.1 | 14% |
| RTX 5090 | GSQ-RCO IQ1_M | llama.cpp | 77.2 | 537.1 | 14% |
| 2× Radeon R9700 | GSQ-RCO IQ3_S | llama.cpp + MTP k=3 | 72.5 | – | – |
| 2× Radeon R9700 | GSQ-RCO IQ3_S | llama.cpp | 39.4 | – | – |