NERSCPowering Scientific Discovery Since 1974

Dirac GPU Cluster

Current Status: Up

Dirac is a testbed experimental system and is not considered a NERSC production system. Therefore, policies regarding user access and staff support differ substantially from other NERSC systems. As Dirac is a sub-cluster of Carver, whenever Carver is down for maintenance Dirac will be down. Every effort will be made to give users advance notice in these situations.  Users who desire access to this resource must have research objectives that align with Dirac's research objectives.

Updates and Status

A discussion mailing list has been created so users can ask questions and help each other, gpgpu-discuss@nersc.gov. All users with accounts on Dirac will be added to this list. Please report system problems to the NERSC consultants at consult@nersc.gov. Read More »

Getting started

Request an account on Dirac and learn how to login and run your first program on a GPGPU cluster. Read More »

Node and GPU Configuration

Dirac is a 50 node GPU cluster with NVIDIA Fermi chips. Read More »

Programming

Information on compiling CUDA codes and hybrid MPI+CUDA applications. Read More »

Running jobs

Learn how to submit a job to the queues and request special GPU resources. Batch classes and policies are explained. Read More »

Performance and optimization

Benchmarks and performance of applications running on Dirac Read More »

File Storage

Dirac has 3 kinds of file systems available to users: home directories, scratch directories and project directories, all provided by the NERSC Global File system. Read More »