// RUN — FITS

Can the NVIDIA DGX Spark run LLaMA-7B?

Yes — here's how fast and up to what context, with real measured numbers.

LLaMA-7B on the NVIDIA DGX Spark

QuantFits?Max contextSpeedBeyond context
FP16 yes 765k 14.9-17.7 t/s ~0.5-0.6 (slow)
Q8 yes 816k 29.7-35.4 t/s ~1.1-1.3 (slow)
Q4 yes 839k 53.1-63.3 t/s ~1.9-2.3 (slow)

Estimate (memory-bound). Beyond "max context" the KV cache spills to SSD swap and speed drops to the "beyond" figure.

→ All models on the NVIDIA DGX Spark → LLaMA-7B specs → Other hardware

FAQ

Can the NVIDIA DGX Spark run LLaMA-7B?▶

Yes. At Q4 it generates ~53.1-63.3 tok/s and fits up to 839k context; beyond that it spills to SSD swap and slows to ~1.9-2.3 tok/s.

How much context fits?▶

At Q4, up to about 839k tokens stay in fast memory; longer context spills to SSD swap and throughput collapses.