Fugaku is the new King

The biggest GOOD news in IT industry this year must be the announcement of Fugaku supercomputer in earlier June, taking the top spot on the Top500 list, a ranking of the world’s fastest supercomputers. It also swept the other rankings of supercomputer performance, taking first place on the HPCG, a ranking of supercomputers running real-world applications, HPL-AI, which ranks supercomputers based on their performance capabilities for tasks typically used in artificial intelligence applications, and Graph 500, which ranks systems based on data-intensive loads. This is the first time in history that the same supercomputer has become No.1 on Top500, HPCG, and Graph500 simultaneously.

Why am I so excited about Fugaku and use this news as my opening post for a new HPC series?

The concept of Fugaku is proposed 10 years ago and the project started around 2014. Its concept is not brand new but significant design and packaging improvements are applied over its predecessors. These changes and improvements reflected some of most heated debate/discussions in HPC industry and can serve very well as testimony in these topics.

A few key differentiation factors of interest are:

  • ARM v8 SVE instruction set vs Fujitsu’s traditional SPARC solution vs Intel x86
  • 48 low Freq/low power cores per CPU vs less core with higher Freq/power per CPU
  • HBM vs DDR4/GDDR6
  • 6D Mesh/Torus interconnect topology vs Blue Gene’s 5D Torus vs Summit’s Fat-Tree
  • 3D stacked memory vs integrated memory on chip

Why am I even interested about HPC/Supercomputer, as a networking (in macro way) guy?

  1. Supercomputer/HPC is technology driver for computing industry, especially in packaging and interconnect.
  2. Converged compute, storage and interconnect co-design is deep in its gene for supercomputer/HPC
  3. Boundary between HPC and non-HPC applications has blurred in recent years, especially with AI and IoT applications, making HPC solutions very relevant to general purpose compute infrastructure like DC and edge.

In the following posts in this series, I will dig deeper into above areas and reflect the knowledge into the more generic NCS discussions, stay tuned friends!