GB10 (DGX Spark)
3 records across 3 models (quants: INT4 (AutoRound), Q4_K_M).
| hardware | model | quant | backend | tok/s | pp tok/s | W | ttft s | ctx/tg | date | class | flags | source |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| GB10 (DGX Spark) | Qwen3-Coder-Next | Q4_K_M | llama.cpp | 63 | – | – | – | – | 2026-05 | community | – | source |
| GB10 (DGX Spark) | Qwen3.5-122B-A10B 122B-A10B | INT4 (AutoRound) | vLLM | 59.6 | – | – | – | – | 2026-07 | community | – | source |
| GB10 (DGX Spark) | Qwen3.6-35B-A3B 35B-A3B | Q4_K_M | llama.cpp | 59 | – | – | – | – | 2026-05 | community | – | source |
All rows cite their source. Provenance: community 40018 sourced 1085