🔌
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 BERT-Large?
Yes — here's how fast and up to what context, with real measured numbers.
BERT-Large on the NVIDIA DGX Spark
| Quant | Fits? | Max context | Speed | Beyond context |
|---|---|---|---|---|
| FP16 | yes | 1725k | 293.1-349.3 t/s | ~10.7-12.8 (slow) |
| Q8 | yes | 1730k | 586.1-698.6 t/s | ~21.5-25.6 (slow) |
| Q4 | yes | 1732k | 1046.7-1247.4 t/s | ~38.3-45.7 (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 BERT-Large?▶
Yes. At Q4 it generates ~1046.7-1247.4 tok/s and fits up to 1732k context; beyond that it spills to SSD swap and slows to ~38.3-45.7 tok/s.
How much context fits?▶
At Q4, up to about 1732k tokens stay in fast memory; longer context spills to SSD swap and throughput collapses.