LLM Inference Benchmark Explorer
Qwen3.6-27B on 1× DGX Spark — FP8, vLLM, TP1, speculative decoding inference benchmark
Open this configuration in the LLM Inference Benchmark Explorer
| Model | Qwen3.6-27B |
|---|---|
| Parameters | 27B |
| Intelligence Index | 21.4 |
| Agentic Index | 18.5 |
| Device | 1× DGX Spark |
| Quantization | FP8 |
| Max C | 2 |
| Chat Capacity | 8 |
| Agentic Capacity | 3 |
| TP | — |
| DP | — |
| PP | — |
| Engine | vLLM |
| Speculative Decoding | Yes |
| C | TTFT (ms) | TPS (tok/s) | Status |
|---|---|---|---|
| 1 | 323 | 20.24 | PASS |
| 2 | 516 | 21.34 | PASS |
| 4 | 531 | 19.90 | FAIL |
| 8 | 621 | 17.87 | FAIL |
| 16 | 813 | 15.00 | FAIL |
In this measurement, Qwen3.6-27B (27B parameters), served in FP8 format with vLLM and speculative decoding on 1× DGX Spark, reached a generation speed of 20.2 tok/s per request and a time to first token (TTFT) of 323 ms with a single request.
Considering the speed targets and the available KV cache capacity, the estimated capacity is 8 users for chat use and 3 for agentic use. The realistic capacity will likely fall between these two values. In scenarios dominated by coding, tool use, long workflows and multi-agent use, capacity approaches the agentic estimate; where shorter interactions, standard conversations and lighter tasks dominate, it approaches the chat estimate.
Notes: MTP gives 2.4x TPS improvement
Intelligence Index and Agentic Index values are published by Artificial Analysis and are reproduced here with attribution.
In this measurement, Qwen3.6-27B (27B parameters), served in FP8 format with vLLM and speculative decoding on 1× DGX Spark, reached a generation speed of 20.2 tok/s per request and a time to first token (TTFT) of 323 ms with a single request.
Considering the speed targets and the available KV cache capacity, the estimated capacity is 8 users for chat use and 3 for agentic use. The realistic capacity will likely fall between these two values. In scenarios dominated by coding, tool use, long workflows and multi-agent use, capacity approaches the agentic estimate; where shorter interactions, standard conversations and lighter tasks dominate, it approaches the chat estimate.