🔌
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 Phi-3.5-MoE?
Yes — here's how fast and up to what context, with real measured numbers.
Phi-3.5-MoE on the NVIDIA H200
| Quant | Fits? | Max context | Speed | Beyond context |
|---|---|---|---|---|
| FP16 | yes | 20k | 28.8-34.3 t/s | ~0.2-0.3 (slow) |
| Q8 | yes | 206k | 57.6-68.7 t/s | ~0.5-0.6 (slow) |
| Q4 | yes | 287k | 102.9-122.7 t/s | ~0.9-1 (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 Phi-3.5-MoE?▶
Yes. At Q4 it generates ~102.9-122.7 tok/s and fits up to 287k context; beyond that it spills to system RAM offload (depends on your system RAM) and slows to ~0.9-1 tok/s.
How much context fits?▶
At Q4, up to about 287k tokens stay in fast memory; longer context spills to system RAM offload (depends on your system RAM) and throughput collapses.