🔌
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 Deephermes 3 Llama 3 8B Preview?
Yes — here's how fast and up to what context, with real measured numbers.
Deephermes 3 Llama 3 8B Preview on the NVIDIA DGX Spark
| Quant | Fits? | Max context | Speed | Beyond context |
|---|---|---|---|---|
| FP16 | yes | 745k | 12.5-14.8 t/s | ~0.5-0.5 (slow) |
| Q8 | yes | 806k | 24.9-29.7 t/s | ~0.9-1.1 (slow) |
| Q4 | yes | 833k | 44.5-53 t/s | ~1.6-1.9 (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 Deephermes 3 Llama 3 8B Preview?▶
Yes. At Q4 it generates ~44.5-53 tok/s and fits up to 833k context; beyond that it spills to SSD swap and slows to ~1.6-1.9 tok/s.
How much context fits?▶
At Q4, up to about 833k tokens stay in fast memory; longer context spills to SSD swap and throughput collapses.