llamaperf

NVIDIA DGX Spark vs M3 Ultra 256GB for local LLMs

57 reports on the NVIDIA DGX Spark and 6 on the M3 Ultra 256GB

Which is faster for local LLMs?

On Qwen3.8 125B · 6B active at 4-bit, the NVIDIA DGX Spark runs at 29.8 tokens per second (2 runs) and the M3 Ultra 256GB at 25.7 (one run). Those are community reports whose engines and settings differ, so the gap is a rough sign, not a controlled measure of the two cards. Memory bandwidth, which caps how fast a card can write, is 3.0x higher on the M3 Ultra 256GB (273 GB/s on the NVIDIA DGX Spark, 819 on the M3 Ultra 256GB).

NVIDIA DGX Spark
Memory
128GB unified
Memory bandwidth
273 GB/s
FP16 compute
125 TFLOPS
Reports
57
M3 Ultra 256GB
Memory
256GB unified
Memory bandwidth
819 GB/s
FP16 compute
28.4 TFLOPS
Reports
6

Measured on both

Median tokens per second from plain runs (one device, one request, no speculative decoding, the whole model in the device's memory) of the same model size at the same quant level. Engines and context lengths can still differ between the runs. How to read these

ModelNVIDIA DGX SparkM3 Ultra 256GB
Qwen3.8 125B · 6B active4-bit29.82 runs25.7one run

Estimated by the calculator

The quant the calculator recommends for each card and its estimated speed at a 32,768-token context with 32 GB of system RAM, from memory bandwidth and the model's shape. Where a model was also measured above, the measurement wins.

ModelNVIDIA DGX SparkM3 Ultra 256GB
Qwen3.8 125B · 6B active14.4 t/sQ6_K27.3 t/sQ8_0
Qwen3.8 27B4.0 t/sFP168.0 t/sFP16
DeepSeek V4 Flash 284B · 13B active38.3 t/sQ2_K36.0 t/sQ4_K_M
Qwen3.6 35B · 3B active14.5 t/sFP1628.7 t/sFP16
Qwen3.6 27B4.0 t/sFP168.0 t/sFP16
Gemma 4 26B · 4B active10.8 t/sFP1621.5 t/sFP16
DeepSeek V4.1 Flash 552B · 16B activedoesn't fit at these settings48.0 t/sQ2_K

Change the context, RAM or model in the calculator: NVIDIA DGX Spark · M3 Ultra 256GB

Every report on each card, with its source: NVIDIA DGX Spark · M3 Ultra 256GB

More comparisons