4× MI325X
14 records across 1 models (quants: MXFP8, unstated).
| hardware | model | quant | backend | tok/s | pp tok/s | W | ttft s | ctx/tg | date | class | flags | source |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 4× MI325X | DeepSeek-V4.1-Flash | MXFP8 | vLLM | 714.8 | – | – | – | 8192 | 2026-10-14 | sourced | – | source |
| 4× MI325X | DeepSeek-V4.1-Flash | MXFP8 | vLLM | 699.8 | – | – | – | 8192 | 2026-10-14 | sourced | – | source |
| 4× MI325X | DeepSeek-V4.1-Flash | MXFP8 | vLLM | 431.1 | – | – | – | 8192 | 2026-10-14 | sourced | – | source |
| 4× MI325X | DeepSeek-V4.1-Flash | MXFP8 | vLLM | 430.2 | – | – | – | 8192 | 2026-10-14 | sourced | – | source |
| 4× MI325X | DeepSeek-V4.1-Flash | MXFP8 | vLLM | 165.8 | – | – | – | 8192 | 2026-10-14 | sourced | contradiction | source |
| 4× MI325X | DeepSeek-V4.1-Flash | MXFP8 | vLLM | 165.7 | – | – | – | 8192 | 2026-10-14 | sourced | contradiction | source |
| 4× MI325X | DeepSeek-V4.1-Flash | unstated | vLLM | 155.5 | – | – | 1070.71 | – | 2026-10 | sourced | – | source |
| 4× MI325X | DeepSeek-V4.1-Flash | unstated | vLLM | 155.45 | – | – | 1101.5 | – | 2026-10 | sourced | – | source |
| 4× MI325X | DeepSeek-V4.1-Flash | MXFP8 | vLLM | 113.5 | – | – | – | 262144 | 2026-10-14 | sourced | – | source |
| 4× MI325X | DeepSeek-V4.1-Flash | MXFP8 | vLLM | 109.3 | – | – | – | 262144 | 2026-10-14 | sourced | – | source |
| 4× MI325X | DeepSeek-V4.1-Flash | MXFP8 | vLLM | 98.2 | – | – | – | 262144 | 2026-10-14 | sourced | – | source |
| 4× MI325X | DeepSeek-V4.1-Flash | MXFP8 | vLLM | 95.5 | – | – | – | 262144 | 2026-10-14 | sourced | – | source |
| 4× MI325X | DeepSeek-V4.1-Flash | unstated | vLLM | 33.58 | – | – | 36063.46 | – | 2026-10 | sourced | – | source |
| 4× MI325X | DeepSeek-V4.1-Flash | unstated | vLLM | 32.84 | – | – | 37310.48 | – | 2026-10 | sourced | – | source |
All rows cite their source. Provenance: community 40018 sourced 1085