🔌
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 ProteinBERT?
Yes — here's how fast and up to what context, with real measured numbers.
ProteinBERT on the NVIDIA DGX Spark
| Quant | Fits? | Max context | Speed | Beyond context |
|---|---|---|---|---|
| FP16 | yes | 1734k | 6227.8-7422.2 t/s | ~228.1-271.9 (slow) |
| Q8 | yes | 1735k | 12455.6-14844.4 t/s | ~456.3-543.8 (slow) |
| Q4 | yes | 1735k | 22242.2-26507.8 t/s | ~814.7-971 (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 ProteinBERT?▶
Yes. At Q4 it generates ~22242.2-26507.8 tok/s and fits up to 1735k context; beyond that it spills to SSD swap and slows to ~814.7-971 tok/s.
How much context fits?▶
At Q4, up to about 1735k tokens stay in fast memory; longer context spills to SSD swap and throughput collapses.