Versions Compared

Key

  • This line was added.
  • This line was removed.
  • Formatting was changed.

What is TACC?

Texas Advanced Computing Center provides:

...

Stuck with TACC specific errors or need a specific tool installed on TACC? Submit a ticket to TACC consulting.  TACC's Cluster Systems

What is a cluster?

Cluster systems are made up of multiple computers, connected together to act as one. Each computer is called a node in the cluster and can and can have multiple processors (called cores).  Users log in to the cluster through a limited number of login/head nodes and submit jobs to the many compute nodes.  These systems are inherently parallel and can be greatly beneficial when your jobs are also parallelized.

HPC System ArchitectureImage Added

TACC's Cluster Systems

LONESTAR6:                                               

...

  • 560 nodes (computers) with 128 cores per node
  • 71, 680 cores (processors)
  • 84 GPU nodes with 128 cores per node (for more computationally intensive tasks)
  • 256 GB RAM
  • Max run time:  48 hours 
  • USE: For running large, parallel computation jobs.

STAMPEDE2STAMPEDE3:

  • 4,200 KNL nodes 1060 SKX nodes (computers) with 68 48 cores (processors) per node); 1736 SKX , 560 SPR nodes (computers) with 112 cores (processors) per node, 224 ICX nodes (computers) 285,600 cores (processors); 83,328 cores (processors)with 80 cores per node
  • GPU nodes: 20 PVC nodes with 96 cores per node, 24 H100 GPU nodes with 96 cores per node 
  • Max run time: 48 hours (without approval), 120 hours (with approval)
  • USE: For running large, parallel computation jobs


Frontera:


TACC's

...

Storage Systems


CORRAL:
  • Replicated storage
  • 6 Petabytes of storage
  • Accessible on lonestar and stampede systems
  • $250 per terabyte (First 5 terabytes free for UT users)
  • USE: Backup data, analysis results.

RANCH:

  • Tape storage
  • Archival storage- not replicated or backed up.
  • 60 Petabytes of storage
  • Immediate access can be difficult.
  • USE: Long term archival of data. One of two copies.


Now on to how to use the lonestar6 cluster...