New: connect Claude & other AIs to GenAIList over MCP — research the catalog and contribute to the shared knowledge base. Learn how →
Can the NVIDIA B200 run MT-DNN?
Yes — here's how fast and up to what context, with real measured numbers.
MT-DNN on the NVIDIA B200
| Quant | Fits? | Max context | Speed | Beyond context |
|---|---|---|---|---|
| FP16 | yes | 2662k | 8848.5-10545.5 t/s | ~44.2-52.7 (slow) |
| Q8 | yes | 2667k | 17697-21090.9 t/s | ~88.5-105.5 (slow) |
| Q4 | yes | 2670k | 31601.7-37662.3 t/s | ~158-188.3 (slow) |
Estimate (memory-bound). Beyond "max context" the KV cache spills to system RAM offload (depends on your system RAM) and speed drops to the "beyond" figure.
FAQ
Can the NVIDIA B200 run MT-DNN?▶
Yes. At Q4 it generates ~31601.7-37662.3 tok/s and fits up to 2670k context; beyond that it spills to system RAM offload (depends on your system RAM) and slows to ~158-188.3 tok/s.
How much context fits?▶
At Q4, up to about 2670k tokens stay in fast memory; longer context spills to system RAM offload (depends on your system RAM) and throughput collapses.