Going deep on local AI performance benchmarking with Apple’s Mac Studio with M5 Ultra taking on Nvidia DGX Spark
Reported by Tom's Hardware1 source ↓
But if you're thinking about dropping over $10K for the purpose of building a local AI inference setup, is a Mac Studio right for you?
Key points
- The Mac Studio’s energy-delay product was less than half the DGX Spark-like system’s
- The test used Qwen 3.8 27B with llama.cpp on both systems
- The tested Mac Studio configuration cost $12,300 or more; the Dell GB10 system cost about $10,000
Details
Tom’s Hardware compared an M5 Ultra Mac Studio with a Dell Pro Precision GB10 system, which it says is nearly identical to the Nvidia DGX Spark. In an AgentPerf-Local test using Qwen 3.8 27B and llama.cpp, the Mac finished in just under half the time and used about 16% less task energy. Its energy-delay product, which combines completion time and energy use, was also less than half the GB10’s.
The tested Mac Studio configuration cost $12,300 or more, while the Dell GB10 cost about $10,000. Tom’s Hardware says quantized versions of GLM-5.3 Flash and DeepSeek V4.1 Flash 1017 fit in the Mac’s 256GB of memory but not on a single 128GB DGX Spark; evaluating those models against Nvidia’s system requires clustering multiple units. The article excerpt does not provide results from those larger-model tests.
Written from coverage as of
Images
- Tom's Hardware
Stocks in this story
Change since open: 20 Nov to 21 Nov 2026
Last refresh