GPU Nodes under construction: Difference between revisions
| Line 1: | Line 1: | ||
==ARROW GPU Nodes== | ==ARROW GPU Nodes== | ||
Arrow currently has two sets of GPU nodes.The first set consists of three NVIDIA RTX A4000 GPUs. The second set consists | Arrow currently has two sets of GPU nodes.The first set consists of three NVIDIA RTX A4000 GPUs. The second set consists | ||
of four NVIDIA RTX A6000 GPUs. These GPU nodes must be requeted using the PBS scheduler. <p> First you must use your client computer running a terminal program to log onto one of the logon nodes. In this case I used logon1.arrow.tracc.anl.gov. Then I used login1 to | of four NVIDIA RTX A6000 GPUs. These GPU nodes must be requeted using the PBS scheduler. <p> First you must use your client computer running a terminal program to log onto one of the logon nodes. In this case I used logon1.arrow.tracc.anl.gov. Then I used login1 to join the queue for a4000 using the qsub command. <p>Following is a sample program | ||
that shows how to log onto the a4000 specifying two of the 3 GPUs, 16 of the CPUs, 150G of memory, and a maximum running time of three hours. The request is | that shows how to log onto the a4000 specifying two of the 3 GPUs, 16 of the CPUs, 150G of memory, and a maximum running time of three hours. The request is | ||
queued using the qsub command with the PBS scheduler. Any of a4000 components not requested are then available for another user (or another instance of this user) to request and use.<p> | queued using the qsub command with the PBS scheduler. Any of a4000 components not requested are then available for another user (or another instance of this user) to request and use.<p> | ||
Latest revision as of 23:01, February 3, 2025
ARROW GPU Nodes
Arrow currently has two sets of GPU nodes.The first set consists of three NVIDIA RTX A4000 GPUs. The second set consists
of four NVIDIA RTX A6000 GPUs. These GPU nodes must be requeted using the PBS scheduler.
First you must use your client computer running a terminal program to log onto one of the logon nodes. In this case I used logon1.arrow.tracc.anl.gov. Then I used login1 to join the queue for a4000 using the qsub command.
Following is a sample program that shows how to log onto the a4000 specifying two of the 3 GPUs, 16 of the CPUs, 150G of memory, and a maximum running time of three hours. The request is queued using the qsub command with the PBS scheduler. Any of a4000 components not requested are then available for another user (or another instance of this user) to request and use.
ssh [email protected] [email protected]'s password:PaSsWD