Will 1000 ExaFlop Supercomputers Come from Brute Force Scaling or New Technology?

Next year the world will have its first ExaFlop Supercomputer. This supercomputer will likely be built in China. The USA will follow about 12 months after with the A21 ExaFlop supercomputer in 2021.

Ten researchers from China’s National University of Defense Technology (NUDT) recently published a paper that describes pushing current technologies to 1000 times higher speeds by 2035.

The paper is “Moving from exascale to zettascale computing: challenges and techniques”.

They define a zettascale supercomputer as a system with peak zettaflops (1,000 exaflops or 1,000,000 petaflops) for 64-bit floating point operations.

The current fastest supercomputer is the Summit supercomputer. It has reached 200 petaflops with 64-bit floating point operations. However, it has reached over 1000 petaflops using reduced mathematical precision.

The NUDT researchers described other metrics for a zettascale system.

Power consumption of 100 MW. A power efficiency of 10 teraflops/watt
Peak performance per node of 10 petaflops
Communication bandwidth between nodes of 1.6 terabits/second
I/O bandwidth of 10 to 100 petabytes/second
Storage capacity of one zettabyte
Floor space of 1000 square meters

They expect CMOS to hit the wall in 2025. Optical computing, quantum computing, and biological computing will have growing roles to accelerate traditional computing technology. Supercomputer centers will introduce optical computing, quantum computing, and biological computing accelerators or coprocessors.

There will be increasing use of 3D integrated circuits. This has already begun for computer memory.

Non-volatile memory and interconnects can be integrated into these 3D devices. This will improve performance and efficiency as the push for higher speeds will require super-high density. The speed of light will become an increasing factor in forcing higher density systems.

Opto-electrical devices based on photonic crystals and carbon nanotubes will emerge during the next decade. This will make for more scalable and more balanced supercomputers. Interconnect speeds will reach 400 gigabits/second and chip throughput will reach hundreds of terabits/second.

Superconducting Supercomputer Alternative

In 2018, China invested $145-million into a five-year catch-up effort to fabricate their own superconducting computers by 2022.

Superconducting supercomputers may be capable of 1 PFLOP/s for about 25 kW and 100 PFLOP/s for about 200 kW, including the cryogenic cooler. In the USA, IARPA began funding superconducting computers as a long-term solution to the power-cooling problem of supercomputers. Cryogenics make superconducting chips equipped with Josephson junctions 1,000 times more expensive to cool than CMOS, but they are 100,000 times more energy efficient in operation, yielding a 100-fold increase in overall energy efficiency. Superconducting computers could enable clock speeds that are 150 times faster.

SOURCES: Top500, Frontiers of Information Technology & Electronic Engineering

Written by Brian Wang. Nextbigfuture.com