LLM Inference Benchmark Explorer
GPT-OSS 120B on 2× DGX Spark — MXFP4, vLLM, TP2 inference benchmark
Open this configuration in the LLM Inference Benchmark Explorer
| Model | GPT-OSS 120B |
|---|---|
| Parameters | 120B |
| Intelligence Index | 11.6 |
| Agentic Index | 3.7 |
| Device | 2× DGX Spark |
| Quantization | MXFP4 |
| Max C | 8 |
| Chat Capacity | 32 |
| Agentic Capacity | 12 |
| TP | 2 |
| DP | — |
| PP | — |
| Engine | vLLM |
| Speculative Decoding | — |
| C | TTFT (ms) | TPS (tok/s) | Status |
|---|---|---|---|
| 1 | 168 | 69.86 | PASS |
| 2 | 228 | 51.52 | PASS |
| 4 | 255 | 37.14 | PASS |
| 8 | 292 | 26.92 | PASS |
| 16 | 320 | 19.08 | FAIL |
| 32 | 465 | 12.63 | FAIL |
In OpenZeka's measurement, GPT-OSS 120B (120B parameters), served in MXFP4 format with vLLM on 2× DGX Spark (TP=2), reached a generation speed of 69.9 tok/s per request and a time to first token (TTFT) of 168 ms with a single request.
Considering the speed targets and the available KV cache capacity, the estimated capacity is 32 users for chat use and 12 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: CUTLASS MoE, FlashInfer attn. +26% TPS over TP=1 at C1
Intelligence Index and Agentic Index values are published by Artificial Analysis and are reproduced here with attribution.
In OpenZeka's measurement, GPT-OSS 120B (120B parameters), served in MXFP4 format with vLLM on 2× DGX Spark (TP=2), reached a generation speed of 69.9 tok/s per request and a time to first token (TTFT) of 168 ms with a single request.
Considering the speed targets and the available KV cache capacity, the estimated capacity is 32 users for chat use and 12 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.