New: connect Claude & other AIs to GenAIList over MCP — research the catalog and contribute to the shared knowledge base. Learn how →
Can the NVIDIA H200 run LUKE?
Yes — here's how fast and up to what context, with real measured numbers.
LUKE on the NVIDIA H200
| Quant | Fits? | Max context | Speed | Beyond context |
|---|---|---|---|---|
| FP16 | yes | 1942k | 3627.3-4323 t/s | ~30.2-36 (slow) |
| Q8 | yes | 1949k | 7254.7-8646 t/s | ~60.5-72 (slow) |
| Q4 | yes | 1952k | 12954.7-15439.2 t/s | ~108-128.7 (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 LUKE?▶
Yes. At Q4 it generates ~12954.7-15439.2 tok/s and fits up to 1952k context; beyond that it spills to system RAM offload (depends on your system RAM) and slows to ~108-128.7 tok/s.
How much context fits?▶
At Q4, up to about 1952k tokens stay in fast memory; longer context spills to system RAM offload (depends on your system RAM) and throughput collapses.