Skip to content

Resource Groups

Last Updated: October 9, 2025


Genkai's compute nodes are configured with multiple resource groups for each node group, and jobs must be executed with resource groups in mind when using the compute nodes.

There are two main methods to translate, edit, combine, and execute programs: interactive processing and batch processing.

Interactive processing is a method where the user interactively performs tasks such as translating source programs and executing executable programs by entering commands. Interactive processing is suitable for translations and executions of small-scale programs.

Batch processing is a method where the user creates a “job script file” that describes a series of program executions and has the computer execute the job in bulk. Batch processing is suitable for executing programs that require a large amount of memory, have long execution times, or have high parallelism.

In addition to the resource group limit, a limit is set for the number of cores in use at the same time. When a job that exceeds the limit is submitted, it will be placed in a waiting state regardless of the resource group congestion status.


Node Group A

Resource Group

Number of CPU Cores

Number of Nodes

Available Memory

Maximum Execution Time

a-batch

1~120

3.7GiB * Number of Cores

168 hours

1~128

453GiB * Number of Nodes

Up to 1 node: 168 hours

2~3 nodes: 96 hours

4~15 nodes: 48 hours

16~31 nodes: 24 hours

32~63 nodes: 12 hours

64~128 nodes: 6 hours

a-inter

1~120

3.7GiB * Number of Cores

6 hours

1

453GiB

6 hours

  • The available memory and maximum execution time are determined by the specified number of CPU cores or nodes.
  • If fewer than 120 cores are specified, execution performance may degrade due to the influence of other jobs running on the same node.

Node Group B

Resource Group

Number of GPUs

Number of Nodes

Assigned Number of CPU Cores

Available Memory

Maximum Execution Time

b-batch

1~4

30 cores * Number of GPUs

226.7GiB * Number of GPUs

168 hours

1~4

120 cores * Number of Nodes

907GiB * Number of Nodes

Up to 1 node: 168 hours

2~3 nodes: 96 hours

4 nodes: 48 hours

b-inter

1~4

30 cores * Number of GPUs

226.7GiB * Number of GPUs

6 hours

1

120 cores

907GiB

6 hours

b-batch-mig

1~28 (sub)

4 * Number of sub GPUs

32.3GiB * Number of sub GPUs

168 hours

b-inter-mig

1~28 (sub)

4 * Number of sub GPUs

32.3GiB * Number of sub GPUs

6 hours

  • The number of CPU cores, available memory, and maximum execution time are determined by the specified number of GPUs.
  • In b-batch-mig and b-inter-mig, NVIDIA’s MIG function is used to divide one physical GPU into seven sub GPUs.
  • If fewer than 4 GPUs are specified in b-batch and b-inter, or fewer than 28 GPUs are specified in b-batch-mig and b-inter-mig, execution performance may degrade due to the influence of other jobs running on the same node.
    If multiple GPUs are specified in b-batch-mig/b-inter-mig, the GPU may be allocated across multiple physical GPUs. (There is no way to enforce allocation within the same physical GPU.)

Node Group C

Resource Group

Number of GPUs

Number of Nodes

Assigned Number of CPU Cores

Available Memory

Maximum Execution Time

c-batch

1~8

14 cores * Number of GPUs

930.6GiB * Number of GPUs

168 hours

1

120 cores

7.1TiB

168 hours

c-inter

1~8

14 cores * Number of GPUs

930.6GiB * Number of GPUs

6 hours

1

120 cores

7.1TiB

6 hours

  • The number of CPU cores, available memory, and maximum execution time are determined by the specified number of GPUs.
  • If fewer than 8 GPUs are specified, execution performance may degrade due to the influence of other jobs running on the same node.

Concurrent Core Usage Limits

For resource groups a-batch and b-batch, a limit is set for the number of concurrent cores used per user and per group, respectively.
If a job is submitted in excess of the limit, it will not be executed until the number of concurrent cores is lower than the limit.


a-batch

Resource Group

Limits of the number of cores

per User

15,600 (=130x120)

per Group

31,200 (=260x120)


b-batch

Resource Group

Limits of the number of cores

per User

1,000

per Group

2,000