🔌
New: connect Claude & other AIs to GenAIList over MCP — research the catalog and contribute to the shared knowledge base. Learn how →
// RUN — FITS
Can the NVIDIA H200 run CPM-Bee?
Yes — here's how fast and up to what context, with real measured numbers.
CPM-Bee on the NVIDIA H200
| Quant | Fits? | Max context | Speed | Beyond context |
|---|---|---|---|---|
| FP16 | yes | 661k | 175.2-208.8 t/s | ~1.5-1.7 (slow) |
| Q8 | yes | 722k | 350.4-417.6 t/s | ~2.9-3.5 (slow) |
| Q4 | yes | 748k | 625.7-745.7 t/s | ~5.2-6.2 (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 H200 run CPM-Bee?▶
Yes. At Q4 it generates ~625.7-745.7 tok/s and fits up to 748k context; beyond that it spills to system RAM offload (depends on your system RAM) and slows to ~5.2-6.2 tok/s.
How much context fits?▶
At Q4, up to about 748k tokens stay in fast memory; longer context spills to system RAM offload (depends on your system RAM) and throughput collapses.