Going deep on local AI performance benchmarking with Apple’s Mac Studio with M5 Ultra taking on Nvidia DGX Spark

Tom's Hardware benchmarked Apple's M5 Ultra Mac Studio against a Nvidia DGX Spark-like system running local AI workloads. The M5 Ultra delivered significantly faster performance and lower task energy on dense models like Qwen using llama.cpp, but faced memory and software ecosystem limitations when running larger models without specialized inference engines like TensorFold. Open article →