🔌
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 gpt-sw3-40b?
Yes — here's how fast and up to what context, with real measured numbers.
gpt-sw3-40b on the NVIDIA DGX Spark
| Quant | Fits? | Max context | Speed | Beyond context |
|---|---|---|---|---|
| FP16 | yes | 103k | 2.5-3 t/s | ~0.1-0.1 (slow) |
| Q8 | yes | 225k | 5-5.9 t/s | ~0.2-0.2 (slow) |
| Q4 | yes | 279k | 8.9-10.6 t/s | ~0.3-0.4 (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 gpt-sw3-40b?▶
Yes. At Q4 it generates ~8.9-10.6 tok/s and fits up to 279k context; beyond that it spills to SSD swap and slows to ~0.3-0.4 tok/s.
How much context fits?▶
At Q4, up to about 279k tokens stay in fast memory; longer context spills to SSD swap and throughput collapses.