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 ERNIE-GEN (large)?
Yes — here's how fast and up to what context, with real measured numbers.
ERNIE-GEN (large) on the NVIDIA H200
| Quant | Fits? | Max context | Speed | Beyond context |
|---|---|---|---|---|
| FP16 | yes | 1946k | 5152.9-6141.2 t/s | ~42.9-51.2 (slow) |
| Q8 | yes | 1951k | 10305.9-12282.4 t/s | ~85.9-102.4 (slow) |
| Q4 | yes | 1954k | 18403.4-21932.8 t/s | ~153.4-182.8 (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 ERNIE-GEN (large)?▶
Yes. At Q4 it generates ~18403.4-21932.8 tok/s and fits up to 1954k context; beyond that it spills to system RAM offload (depends on your system RAM) and slows to ~153.4-182.8 tok/s.
How much context fits?▶
At Q4, up to about 1954k tokens stay in fast memory; longer context spills to system RAM offload (depends on your system RAM) and throughput collapses.