LLM Inference Benchmark Explorer
NVIDIA-Nemotron-3.5-Bolt-74B-A9B on DGX B300 — BF16, vLLM, TP1 inference benchmark
Open this configuration in the LLM Inference Benchmark Explorer
| Model | NVIDIA-Nemotron-3.5-Bolt-74B-A9B |
|---|---|
| Parameters | 74B |
| Intelligence Index | — |
| Agentic Index | — |
| Device | DGX B300 |
| Quantization | BF16 |
| Max C | 256+ |
| Chat Capacity | 236 |
| Agentic Capacity | 158 |
| TP | — |
| DP | — |
| PP | — |
| Engine | vLLM |
| Speculative Decoding | — |
| C | TTFT (ms) | TPS (tok/s) | Status |
|---|---|---|---|
| 1 | 58 | 167.03 | PASS |
| 2 | 112 | 147.90 | PASS |
| 4 | 114 | 130.85 | PASS |
| 8 | 135 | 107.23 | PASS |
| 16 | 166 | 84.50 | PASS |
| 32 | 196 | 61.84 | PASS |
| 64 | 271 | 47.81 | PASS |
| 128 | 439 | 36.16 | PASS |
| 256 | 730 | 20.74 | PASS |
In OpenZeka's measurement, NVIDIA-Nemotron-3.5-Bolt-74B-A9B (74B parameters), served in BF16 format with vLLM on DGX B300, reached a generation speed of 167 tok/s per request and a time to first token (TTFT) of 58 ms with a single request.
Considering the speed targets and the available KV cache capacity, the estimated capacity is 236 users for chat use and 158 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: 74B total / 9B active, 256K context. Model: nvidia/NVIDIA-Nemotron-3.5-Bolt-74B-A9B-BF16
Intelligence Index and Agentic Index values are published by Artificial Analysis and are reproduced here with attribution.
In OpenZeka's measurement, NVIDIA-Nemotron-3.5-Bolt-74B-A9B (74B parameters), served in BF16 format with vLLM on DGX B300, reached a generation speed of 167 tok/s per request and a time to first token (TTFT) of 58 ms with a single request.
Considering the speed targets and the available KV cache capacity, the estimated capacity is 236 users for chat use and 158 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.