コンテンツ

  1. HOME
  2. Services
  3. Computing Services
  4. Supercomputer System
  5. other
  6. New supercomputer system (scheduled to be replaced in FY2022)

New supercomputer system (scheduled to be replaced in FY2022)

Please note that the information on this page include the undetermined contents. We will update the information as needed.

Update History
- December 23, 2021: Article published
- June 8, 2022: Information on the new system updated
- August 10, 2022: Schedule of the operation of the first configuration and the call for application of the new system updated. Schedule table added. Additional information on trial operation and cloud system added.

Contents

Schedule

Currently, we are scheduled to replace a current supercomputer system according to the schedule shown in the following table. Initially, we were scheduled to start service of the new supercomputer system in October 2022, however, the schedule has been delayed due to the global shortage of semiconductors. We apologize for any inconvenience this may cause.

End of the Current System Service
(Current System A,B,C)
Thu. July 28, 2022 at 5:00 p.m.
Suspension of the Service associated with System Replacement From Thu. July 28, 2022 at 5:00 p.m. to the Start of the New system Operation
File access period
(File cannot be written)
From 9:00 a.m. on Mon. August 1, 2022 to 3:00 p.m. on Fri. August 26, 2022(Scheduled)
Start of the New system (First Configuration)
(New system B, C, G and Storage.)
Trial operation scheduled during December 2022, full-scale operation scheduled from January
Start of the New system (Last Configuration)
(New System A and Flash Storage)
Scheduled between around June and August, 2023

system_update_schedule_202208.png

Trial Operation and Cloud System

  • The trial operation period and the cloud system will be available for users who have applied to use the service before July with no additional usage fees.
  • The number of nodes will be increased step by step through the trial operation to start the new system at the earliest.
  • The cloud system will be configured to be used in the same way as the new system B/C.
    • The new system and the cloud system will use the same job scheduler.
    • File reference with NFS will be available for the new storage.
    • We are working to provide approximately 50 nodes.
    • We will announce the details later.

【Important Notice and Request】

  • As a result of the procurement of the new system, the system will be delivered in two stages: the first configuration and the last configuration. It is difficult to provide a definite date for the system installation at this time, since the manufacturing schedule and the delivery date of the equipment have been delayed due to the global shortage of semiconductors. We are working hard to start the operation of the first configuration around December 2022, however we hope for your kind understanding that it will be difficult in this situation. We expect to install the last configuration, system A in June 2023 or later.
  • Since the situation will result in lower computing performance than the current system until the new system A is installed, we would like to ask you to consider the use of supercomputers operated by other centers if necessary.
  • The data migration to the new system is scheduled to start in July 2022 or later, in parallel with the operation. In order to migrate all of the data, over 4 billion files will need to be transferred, and it will likely be difficult to migrate all of the data. In order to reduce the burden of the migration process and shorten the migration period, we ask for your cooperation in reducing the number of files by organizing unnecessary files and combining multiple files into a tar file.

Call for application of the new system

Call for application of the new system is scheduled as follows.

Call for application of the new system Scheduled around October, 2022

Please refer to Appendix 1 of "Regulations for Usage Fees of Academic Center for Computing and Media Studies (ACCMS), Kyoto University" below for the usage fees of the new system.

  • "Regulations for Usage Fees of Academic Center for Computing and Media Studies (ACCMS), Kyoto University"(Enforced on 2022/10/01)

System Configurations

The new system is expected to consist of three types of systems based on the concept of a current system (System A, B and C), a new system equipped with an accelerator (System G), and a storage system.The CPU in systems A, B, and C are all Intel Xeon of the same generation, however there are differences in memory speed and capacity.System A is a configuration that emphasizes memory speed, System B is a configuration with a good balance between capacity and performance, and System C is a configuration that emphasizes memory capacity.

system_simple_2022.png

Camphor 3 (New System A, Successor to the Current System A) Scheduled to be released between around June and August, FY2023

Specifications Machine DELL PowerEdge C6620
Node 1,120
Performance 5.82 PFlops
Memory 140 TByte
Node Performance 5.19 TFlops
Processor 2
Memory 128 GByte
Memory bandwidth 3.2 TByte/sec
Injection Bandwidth 50 GByte/sec
Processor Processor Intel Xeon
Architecture x86-64
Performance 2.6 TFlops

Laurel 3 (New System B, Successor to the Current System B) Scheduled to be released around December, FY2022

Specifications Machine DELL PowerEdge C6620
Node 370
Performance 2.19 PFlops
Memory 185 TByte
Node Performance 5.91 TFlops
Processor 2
Memory 512 GByte
Memory bandwidth 614 GByte/sec
Injection Bandwidth 25 GByte/sec
Processor Processor Name Intel Xeon
Architecture x86-64
Performance 2.96 TFlops

Cinnamon 3(New System C, Successor to the Current System C) Scheduled to be released around December, FY2022

Specifications Machine DELL PowerEdge C6620
Node 16
Performance 94.6 TFlops
Memory 32 TByte
Node Performance 5.91 TFlops
Processor 2
Memory 2 TByte
Memory bandwidth 563 GByte/sec
Injection Bandwidth 25 GByte/sec
Processor Processor Name Intel Xeon
Architecture x86-64
Performance 2.96 TFlops

Gardenia (New System G, a new system equipped with an accelerator) Scheduled to be released around December, FY2022

(Accelerator)
Specifications Machine DELL PowerEdge XE8545
Node 16
Performance(Processor) 42.59 TFlops (double precision)
Memory (Processor) 8.19 TB
Performance(Accelerator) 18.04 PFlops (half-precision)
Memory (Performance) 5.12 TB
Node
Performance(Processor)
2.66 TFlops (double precision)
Memory(Processor) 512 GByte
Memory Bandwidth (Processor) 409 GByte/sec
Performance(Accelerator) 1,128 TFlops or more (half-precision)
Memory(Accelerator) 320 GByte
Memory Bandwidth (Accelerator) 8.15 TByte/sec
Injection Bandwidth 50 GB/sec
Processor Processor AMD EPYC
Architecture x86-64
Performance(double precision) 2.66 TFlops
Accelerator Accelerator NVIDIA A100 80GB SXM
Performance(half-precision) 312 TFlops
Memory 80 GB
Memory Bandwidth 2,039 GB/sec (HBM2e)

Storage System Scheduled to be released in FY2022.

    Physical capacity 40.32 PB
    Effective capacity 31.99 PB
    Data transfer performance 280 GB/sec

    Flash Storage System Scheduled to be released in FY2023.

      Physical capacity 4.06 PB
      Effective capacity 3.12 PB
      Data transfer performance 768 GB/sec

        Software

        New software stack is shown in the following table and basically offers equivalent compiler and libraries to the previous ones. However, the ISV applications is yet to be determined, as they are scheduled to be reviewed.

          OS Red Hat Enterprise Linux 8
          Job Scheduler Slurm
          Compiler Intel oneAPI (Fortran, C/C++, OpenMP)
          NVIDIA HPC SDK (Fortra, C/C++, OpenMP)
          MPI Open MPI
          Library Intel oneAPI MKL (BLAS,LAPACK,ScaLAPACK)
          NVIDIA HPC SDK (BLAS, LAPACK, ScaLAPACK)
          Machine Learning Framework TensorFlow, PyTorch, MXNet
          ISV applications Scheduled to procure the software separately. Software to be provided will be reviewed.

          software_stack_simple_2022.png

           

          Copyright © Institute for Information Management and Communication, Kyoto University, all rights reserved.