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 StyleGAN3-R?
Yes — here's how fast and up to what context, with real measured numbers.
StyleGAN3-R on the NVIDIA H200
| Quant | Fits? | Max context | Speed | Beyond context |
|---|---|---|---|---|
| FP16 | yes | 1956k | 1108860.8-1321519 t/s | ~9240.5-11012.7 (slow) |
| Q8 | yes | 1956k | 2217721.5-2643038 t/s | ~18481-22025.3 (slow) |
| Q4 | yes | 1956k | 3960217-4719710.7 t/s | ~33001.8-39330.9 (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 StyleGAN3-R?▶
Yes. At Q4 it generates ~3960217-4719710.7 tok/s and fits up to 1956k context; beyond that it spills to system RAM offload (depends on your system RAM) and slows to ~33001.8-39330.9 tok/s.
How much context fits?▶
At Q4, up to about 1956k tokens stay in fast memory; longer context spills to system RAM offload (depends on your system RAM) and throughput collapses.