// RUN — FITS

Can the NVIDIA DGX Spark run Code Llama-34B?

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

Code Llama-34B on the NVIDIA DGX Spark

QuantFits?Max contextSpeedBeyond context
FP16 yes 174k 2.9-3.5 t/s ~0.1-0.1 (slow)
Q8 yes 304k 5.9-7 t/s ~0.2-0.3 (slow)
Q4 yes 361k 10.5-12.5 t/s ~0.4-0.5 (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 → Code Llama-34B specs → Other hardware

FAQ

Can the NVIDIA DGX Spark run Code Llama-34B?▶

Yes. At Q4 it generates ~10.5-12.5 tok/s and fits up to 361k context; beyond that it spills to SSD swap and slows to ~0.4-0.5 tok/s.

How much context fits?▶

At Q4, up to about 361k tokens stay in fast memory; longer context spills to SSD swap and throughput collapses.