llamaperf

DGX Spark vs M4 Pro 48GB

For running local LLMs · 30 reports across 11 models

Side A
DGX Spark
Vendor
nvidia
VRAM
128GB
Memory
Unified
Side B
M4 Pro 48GB
Vendor
apple
VRAM
48GB
Memory
Unified

Highest reported generation speed by model family

These maxima come from different community setups. Model size, quantization, GPU count, offloading and concurrency may differ. They do not establish which GPU is faster for the same workload. Check the comparison conditions →

ModelDGX SparkM4 Pro 48GB
DeepSeek V4 Flashup to 285B90.0n=10
Qwen3.8up to 2400B181.0n=420.3n=1
Ling-3.0up to 124B40.9n=4
Museup to 30B37.0n=118.0n=2
GLM-5.2up to 744B24.0n=2
DeepSeek V3n=1
GLM-5.1n=1
GLM-5.3up to 321B1.8n=1
Kimi K320.0n=1
LingBot-World-V2up to 1Bn=1
Qwen3.6up to 35Bn=1

More comparisons