Supercomputer:”Fugaku”, 3 features: 537 petaflops, exascale, Arm base
Supercomputer “Fugaku”:
TOP500 for dense matrix calculation,
HPCG for sparse matrix iterative calculation,
Graph500 for graph processing,
HPL-AI with compound calculation accuracy,
No. 1 in the world in both benchmarks.
100 times more effective performance than ’Kei’:
The goal of “Fugaku” is not the numerical performance target.
How effective the application can be.
It is an application-oriented supercomputer.
As a result, we defined that if the effective performance is 100 times that of ’Kei’, it will be an Exaquere system.
Priority issues of “Fugaku”
“Drug discovery”,
“Life science”,
“Complex disaster prediction”,
“Weather / Global Environment”,
“Energy utilization”,
“Clean energy”,
“Devices / high-performance materials”,
“Manufacturing”,
There are nine fields of “basic science”.
Furthermore, as a budding issue added to this,
“Frontier of Basic Science”,
“Socio-economic phenomenon”,
“Planet formation inside and outside the solar system”,
“Neural circuit / artificial intelligence”,
We will tackle four issues.
Features of “Fugaku”:
It is said that there are some features that “Fugaku” should be proud of.
one is:
“The general-purpose manycore alone achieves high performance of 537 petaflops.”
Besides “Fugaku”, there is one supercomputer that uses a general-purpose CPU.
However, the supercomputer has only 38 petaflops performance.
“Fugaku” is the only general-purpose CPU that achieves such high performance.
As a result, you can use the most popular programming techniques such as OpenMP and MPI.
Realizes an environment with high memory performance and suitable for application development.
It enhances user convenience and portability of performance, and expands the range of usage.
The second is:
“Fugaku” is the point that “the exascale system was completed ahead of the rest of the world.”
Exascale system development projects in the United States and China are all behind.
“Aurora” from DoE, USA:
It was said to be completed in 2018.
It is expected to be delayed until 2022.
China’s Three Great Exascale Plans:
The information is not disclosed to the outside.
On the other hand, “Fugaku” is already in operation and has begun to produce results.
The third is:
“It has opened up the HPC era with Arm-based CPUs.”
“Fugaku” is the first full-scale HPC system using Arm.
Improving floating-point arithmetic performance is necessary for scientific calculations.
SVE (Scalable Vector Extension) was developed for “Fugaku” and adopted for ARM.
Both low power consumption and high performance:
It solved a big problem such as achieving both low power consumption and high performance.
In “Fugaku”
Eco mode that cuts some power that is not necessary for calculation,
Equipped with a boost mode that raises the frequency to accelerate the calculation.
“It is possible to switch depending on the application even at the user level.”
Original OS development:
The proprietary OS realizes massively parallel processing with extremely high efficiency.
The feature of the original OS is
It has a multi-kernel and
It doesn’t just run Linux.
RIKEN’s McKernel can be run on Linux.
As a result, it is possible to “efficiently link and operate the resources of 160,000 nodes”.
(Mynavi News) –Yahoo! News
https://news.yahoo.co.jp/articles/2972fdd91c1399107ac29aeb3e37ff8f056bda8c