The ARM microprocessor architecture used to be an exclusively low-power affair, but today there are full-fledged computers running on ARM chips. In fact, Apple is going to move its laptops to ARM in the coming years. Japanese technology firms Fujitsu and Riken have shown what ARM can really do with their new Fugaku supercomputer. With more than 150,000 CPUs, Fugaku has now become the most powerful supercomputer in the world by a wide margin.
Riken and Fujitsu started developing the system in 2014, working closely with ARM to design the A64FX processor. Each of these ships has 48 CPU cores based on the ARM architecture version 8.2A, making it the first such chip in the world. Fugaku consists of 396 racks as seen above — each full-sized rack contains 384 processing nodes. Each of those CPU cores is clocked at 2.2GHz, sharing 32GB of RAM per chip.
So, the Fugaku system runs on a monster of a chip, and the supercomputer has 158,976 of them. That works out to more than 7 million CPU cores. It pulls a similarly shocking 28.3MW of power. That’s considerably more than the now-runner-up fastest supercomputer, Summit, at 10.1MW. Summit is an IBM-based system running 4,356 nodes, each equipped with two 22-core Power9 CPUs, and six NVIDIA Tesla V100 GPUs. The next system on the list is Sierra from Lawrence Livermore National Laboratory, which runs the same 22-core PowerPC chips in smaller numbers.
The A64FX processor layout.
Fugaku has successfully taken the top spot in the TOP500 supercomputer tracker from Summit with a score of 415.53 petaflops in the LINPACK benchmark. Summit manages a “mere” 148.6 petaflops, making Fugaku almost three times faster. However, Fugaku has no GPUs. The system has also taken the top spot in measures of AI application processing (HPL-AI), Conjugate Gradients (HPCG), and data-intensive workloads (Graph500). No system has led in all four categories at the same time before. It’s also the first Japanese system to top the rankings in almost a decade.
The new Japanese supercomputer is not in full operational mode yet. It will start processing workloads for researchers in 2021. So far, it has been used on an experimental basis for COVID-19 research, diagnostics, and simulations of future virus spread. There’s no telling what this system will manage once it’s fully operational.
Now read:
- Microsoft Built One of the Most Powerful Supercomputers in the World to Develop Human-Like AI
- Intel’s Ponte Vecchio GPU Tips Up as an Add-in Card
- The Fastest Supercomputer on Earth Is Being Deployed Against Coronavirus
No comments:
Post a Comment