New: connect Claude & other AIs to GenAIList over MCP — research the catalog and contribute to the shared knowledge base. Learn how →
Can the NVIDIA B200 run DEQ-Transformer (Post-LN) + Jacobian Regularisation?
Yes — here's how fast and up to what context, with real measured numbers.
DEQ-Transformer (Post-LN) + Jacobian Regularisation on the NVIDIA B200
| Quant | Fits? | Max context | Speed | Beyond context |
|---|---|---|---|---|
| FP16 | yes | 2669k | 29795.9-35510.2 t/s | ~149-177.6 (slow) |
| Q8 | yes | 2671k | 59591.8-71020.4 t/s | ~298-355.1 (slow) |
| Q4 | yes | 2672k | 106414-126822.2 t/s | ~532.1-634.1 (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 B200 run DEQ-Transformer (Post-LN) + Jacobian Regularisation?▶
Yes. At Q4 it generates ~106414-126822.2 tok/s and fits up to 2672k context; beyond that it spills to system RAM offload (depends on your system RAM) and slows to ~532.1-634.1 tok/s.
How much context fits?▶
At Q4, up to about 2672k tokens stay in fast memory; longer context spills to system RAM offload (depends on your system RAM) and throughput collapses.