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 DiT-XL/2?
Yes — here's how fast and up to what context, with real measured numbers.
DiT-XL/2 on the NVIDIA H200
| Quant | Fits? | Max context | Speed | Beyond context |
|---|---|---|---|---|
| FP16 | yes | 1936k | 2595.6-3093.3 t/s | ~21.6-25.8 (slow) |
| Q8 | yes | 1946k | 5191.1-6186.7 t/s | ~43.3-51.6 (slow) |
| Q4 | yes | 1951k | 9269.8-11047.6 t/s | ~77.2-92.1 (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 DiT-XL/2?▶
Yes. At Q4 it generates ~9269.8-11047.6 tok/s and fits up to 1951k context; beyond that it spills to system RAM offload (depends on your system RAM) and slows to ~77.2-92.1 tok/s.
How much context fits?▶
At Q4, up to about 1951k tokens stay in fast memory; longer context spills to system RAM offload (depends on your system RAM) and throughput collapses.