🔌
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 LLaVA-CoT?
Yes — here's how fast and up to what context, with real measured numbers.
LLaVA-CoT on the NVIDIA DGX Spark
| Quant | Fits? | Max context | Speed | Beyond context |
|---|---|---|---|---|
| FP16 | yes | 560k | 9.1-10.8 t/s | ~0.3-0.4 (slow) |
| Q8 | yes | 627k | 18.1-21.6 t/s | ~0.7-0.8 (slow) |
| Q4 | yes | 656k | 32.4-38.6 t/s | ~1.2-1.4 (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 LLaVA-CoT?▶
Yes. At Q4 it generates ~32.4-38.6 tok/s and fits up to 656k context; beyond that it spills to SSD swap and slows to ~1.2-1.4 tok/s.
How much context fits?▶
At Q4, up to about 656k tokens stay in fast memory; longer context spills to SSD swap and throughput collapses.