Research & Academic
Universities traditionally are at the frontline of supercomputing. They have always had a need for compute power for research and education.
According to the TOP500 list (as of November 2019) Research (101 systems representing 20.2%) and Academic (57 systems representing 11.4%) provide a total of 31.6% of all systems on the list. From the current list the top 10 are all form research or academic.
The first three EFlops supercomputer (as far as there is any information available) will also all be built for laboratories. So as it goes for supercomputing the most powerful machines are built for the academic and research arena.
However many of the industrial companies do not run or do not publish their LINPACK benchmark results so the list should not be seen as absolute.
The University of Osaka, besides other fields, does research in laser nuclear fusion, laser processing and high energy density science (HED). This is largely done by using laser light, electron, ion and neutron beams. In this process it is very important to run computational simulations of radiation hydrodynamics.
Newly developed numerical models typically have to be fully developed and tuned before they can be ported to the supercomputer. This is where the biggest advantage of the SX-Aurora TSUBASA tower model comes into play. The compact size of the machine allows it to be placed near the scientist's desk inside the laboratory. The code then can be tuned with less effort and transferred over to the supercomputer more easily.
The performance of the radiation hydrodynamics code is depends on the memory bandwidth, where the SX-Aurora TSUBASA provides market-leading advantage.
This allows a real-time simulation for data assimilation with respect to both the laser experimental data and the simulation data. This saves the university a lot of work and time they have to invest in code tuning and porting.
The Waseda Univeristy – Green Computing System Research and Development Center was founded to develop ultra-low power, high performance computers.
The university sees the SX-Aurora TSUBASA as a very promising technology for this given purpose. The objective is to use the SX-Aurora TSUBASA vector engine to speed up various applications and simultaneously reduce power consumption. This can be achieved by focusing on the research of Automatic Parallelizing Compilers.
As a result of a joint research the Waseda University was able to speed up a NAS Parallel Benchmark with CG program automatic parallelization and automatic vectorization on one single SX-Aurora TSUBASA core by a factor of 11 (from ~115sec to ~10sec). This was achieved by using the universities' own OSCAR compiler on a NEC SX-Aurora TSUBASA A101-1 tower model.