Recent benchmark figures reported by Wccftech show that Meta’s Muse Spark open-source model can achieve a throughput of 20 tokens per second. The workload was executed on a combined hardware setup consisting of two Apple M5 Ultra Mac Studios.
The test underscores growing interest in leveraging high-end Apple silicon for heavy artificial intelligence and machine learning workloads. Combining multiple desktop systems suggests an emphasis on harnessing expanded unified memory and parallel processing capacity to handle large-scale model demands.
While hardware configurations for enterprise and workstation AI continue to evolve rapidly, figures like these provide an early look at how emerging Apple desktop architectures might handle large local AI models.
For full details and further context on the test, check the original report on Wccftech.
