New: connect Claude & other AIs to GenAIList over MCP — research the catalog and contribute to the shared knowledge base. Learn how →
Can the Intel Gaudi 3 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 Intel Gaudi 3
| Quant | Fits? | Max context | Speed | Beyond context |
|---|---|---|---|---|
| FP16 | yes | 1771k | 10949-13591.8 t/s | ~118.4-146.9 (slow) |
| Q8 | yes | 1772k | 21898-27183.7 t/s | ~236.7-293.9 (slow) |
| Q4 | yes | 1773k | 39103.5-48542.3 t/s | ~422.7-524.8 (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 Intel Gaudi 3 run DEQ-Transformer (Post-LN) + Jacobian Regularisation?▶
Yes. At Q4 it generates ~39103.5-48542.3 tok/s and fits up to 1773k context; beyond that it spills to system RAM offload (depends on your system RAM) and slows to ~422.7-524.8 tok/s.
How much context fits?▶
At Q4, up to about 1773k tokens stay in fast memory; longer context spills to system RAM offload (depends on your system RAM) and throughput collapses.