Supercomputer service
Introduction
User ID is required for using Supercomputer system. For the details of the application of User ID, please see this page. Each user must apply an individual User ID, and it is prohibited to use other user’s User ID.
Overview of provided service
Basic service (User ID is required.)
- – Use of Supercomputer storage (home area 100GB)
- – Use of the application server (limited by User’s attributes*): for details, please see this page.
*JHPCN/HPCI users need to purchase the basic service additionally to use the application server.
Additional service (Additional application is required.)
Shared course:
- Each user purchases operation time (token), and a job is executed by consuming the operation time (token). Computational resources are shared by among users. A job that needs a large number of nodes can be run.Note: The operation time that a job consumes is calculated as the product of the elapsed time and the number of used CPU sockets or GPU cards.
- – Example 1: using 1 CPU socket and elapsed time 1 hour (3600 seconds) -> 3600 seconds (=1 x 3600)
- – Example 2: using 8 GPU cards and elapsed time 10 minutes (600 seconds) -> 4800 seconds (=8 x 600)
Exclusive course:
- Each user purchases CPU socket(s) or GPU card(s) (yearly amount per resource), and the user (and members of the user’s group) can use the resources(s) exclusively at a fixed rate. User(s) can run jobs without waiting other users’ jobs.
- – Resource unit of one exclusive CPU socket corresponds to 1/2 node (32 CPU-cores and 256 GiB memory) with 4TB supercomputer storage.
- – Resource unit of one exclusive GPU card corresponds to 1/4 node (16 CPU-cores and 128 GiB memory) with 4TB supercomputer storage.
Supercomputer storage:
- Each user can apply an additional amount of Supercomputer storage (home and work areas).
*The above additional services can be shared by among members of a group, which consists of Users appointed by the applicant of the additional service.
Job class
CPU nodes
| Name of resource group (job queue) |
Limit of usage of computational resource | Limit of jobs per user | ||||
| CPU Sockets per job | Elapsed limit per job | Maximum memory per CPU socket |
Simultaneous job submission | Simultaneous job execution | ||
| Shared course |
ec | 4 | 1 hour | 240GB | 4 | 1 |
| sc | 32 | 24 hours | 240 GB | 128 | 40 | |
| lc | 256 | 240 hours | 240 GB | 128 | 40 | |
| Exclusive course |
######c *1 | number of purchases CPU sockets | 336 hours | 230 GB | 500 | unlimited *2 |
*1: “######” corresponds to the group’s ID.
*2: Jobs can be simultaneously run until the number of total CPU sockets used by the jobs exceeds the number of purchases CPU sockets.
(i.e. if 8 CPU sockets are purchased, 8 jobs, each job used 1 CPU socket, can be simultaneously run.)
GPU nodes
| Name of resource group (job queue) |
Limit of usage of computational resource | Limit of jobs per user | ||||
| GPU cards per job | Elapsed limit per job | Maximum memory per GPU card |
Simultaneous job submission | Simultaneous job execution | ||
| Shared course |
eg | 1 (MIG) | 1 hour | 16 GB *2 | 4 | 1 |
| sg | 4 | 240 hours | 120 GB | 32 | 5 | |
| lg | 32 | 72 hours | 120 GB | 32 | 5 | |
| Exclusive course |
######g *1 | number of purchases GPU cards | 336 hours | 120 GB | 500 | unlimited *3 |
*1: “######” corresponds to the group’s ID.
*2: This is the amount of memory per 1MIG.
*3: Jobs can be simultaneously run until the number of total GPU cards used by the jobs exceeds the number of purchases GPU cards.
(i.e. if 8 GPU cards are purchased, 8 jobs, each job used 1 GPU card, can be simultaneously run.)
Please note that the number of simultaneous job execution may change from time to time based on system utilization.