Harmonious timing: TheVerge reports that the fastest supercomputer in the world, o-developed by Riken and Fujitsu, is now powered by ARM processors.
Using Fujitsu’s 48-core A64FX system-on-chip it turned in a Top500 HPL result of 415.5 petaflops, 2.8x as fast as IBM’s Summit, the nearest competitor. It also attained top spots in other rankings that test computers on different workloads, including Graph 500, HPL-AI, and HPCG. No previous supercomputer has ever led all four rankings at once.
Apple obliquely told yesterday at WWDC about the power and performance the 1st-gen ARM Macs. Looking at the power/performance Keynote slide, you can see today’s notebook @ 30w at 3 grid boxes:
On another forum someone noted that if you’re taking the chart metrics as accurate it shows performance/watt more of than 2x, which may indeed be reasonable for A13/A14 processors, and an A14Z designed to run in a MacBook Pro/iMac using TSMC’s Chip-on-Wafer-on-Substrate could dramatically increase performance between the unified memory architecture and different cores (GPU, Neural processor, etc) especially as Apple could also implement customized macOS low-level runtime support.
The processor customization we’ve seen for iOS devices - multiple cores at different speeds that preserves battery while leading performance per watt - could be similarly customized for Macs, or with desktop chips that forgo battery saving in favor of Moar Power. From A-series portable devices to this Fujitsu supercomputer we see the tremendous flexibility and power of the ARM line.