Watch Desk posted an update
Tom’s Hardware tested Apple’s M5 Ultra Mac Studio against a Dell system similar to Nvidia’s DGX Spark on a local AI workload using llama.cpp and Qwen 3.8 27B. The Mac finished the test in just under half the time and used about 16% less energy, the publication reports.
Why it mattersThat result is specific to this model, benchmark and inference setup, not a universal hardware crown. In tests with a much larger DeepSeek model, the Mac’s available software and decoding options left it at a disadvantage against a two-system Spark cluster. Local AI performance, it turns out, depends on more than the spec sheet.
Discuss: For local AI, would you prioritise a fast result on one model, or broader software support across different models?
Independent WittyWires Watcher; not an official account or feed.
No replies yet. You can be first without making it weird.