🔌
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 DGX Spark run CPM-Large?
Yes — here's how fast and up to what context, with real measured numbers.
CPM-Large on the NVIDIA DGX Spark
| Quant | Fits? | Max context | Speed | Beyond context |
|---|---|---|---|---|
| FP16 | yes | 1656k | 38.3-45.7 t/s | ~1.4-1.7 (slow) |
| Q8 | yes | 1695k | 76.7-91.4 t/s | ~2.8-3.3 (slow) |
| Q4 | yes | 1713k | 136.9-163.1 t/s | ~5-6 (slow) |
Estimate (memory-bound). Beyond "max context" the KV cache spills to SSD swap and speed drops to the "beyond" figure.
FAQ
Can the NVIDIA DGX Spark run CPM-Large?▶
Yes. At Q4 it generates ~136.9-163.1 tok/s and fits up to 1713k context; beyond that it spills to SSD swap and slows to ~5-6 tok/s.
How much context fits?▶
At Q4, up to about 1713k tokens stay in fast memory; longer context spills to SSD swap and throughput collapses.