🔌
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 ruDalle: Kandinsky 3.0?
Yes — here's how fast and up to what context, with real measured numbers.
ruDalle: Kandinsky 3.0 on the NVIDIA H200
| Quant | Fits? | Max context | Speed | Beyond context |
|---|---|---|---|---|
| FP16 | yes | 637k | 147.2-175.5 t/s | ~1.2-1.5 (slow) |
| Q8 | yes | 710k | 294.5-350.9 t/s | ~2.5-2.9 (slow) |
| Q4 | yes | 742k | 525.8-626.7 t/s | ~4.4-5.2 (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 ruDalle: Kandinsky 3.0?▶
Yes. At Q4 it generates ~525.8-626.7 tok/s and fits up to 742k context; beyond that it spills to system RAM offload (depends on your system RAM) and slows to ~4.4-5.2 tok/s.
How much context fits?▶
At Q4, up to about 742k tokens stay in fast memory; longer context spills to system RAM offload (depends on your system RAM) and throughput collapses.