High Performance Computing (Savio)

Services

Research IT provides the Savio computing cluster for use by campus researchers and their collaborators. Savio is a Linux cluster with high-speed, low-latency networking and a high-speed parallel filesystem for high performance computing (HPC).

Key features of Savio include:

  • High-speed, low-latency networking.

  • High-speed parallel filesystems.

  • Pools of a variety of types of compute nodes: standard, high-memory, GPU, and nodes allocated per compute core.

  • Parallel processing across cores, across nodes, and across multiple GPUs.

  • Dedicated permanent personal and group disk space (backed up) and extensive temporary scratch space (not backed up) for active use.

  • Research IT staff provide sysadmin and consulting support and training for users of the system.

Campus researchers can use Savio via several mechanisms:

  • Faculty Computing Allowances (FCAs) provide free access for principal investigators and their group or their collaborators.

    • Each Berkeley faculty member is eligible for 300,000 service units (scalable core-hours) per year.

    • The allowance can be written into start-up and retention letters and used as institutional cost-share on grant proposals.

  • The Condo cluster and Condo storage services provide a framework for principal investigators to buy compute nodes and large amounts of disk storage for dedicated use by their groups and collaborators.

    • The supporting infrastructure and admin support is provided by campus at no charge (and this can be considered institutional cost-share on grant proposals).

    • Condo owners can "burst" beyond their hardware to access additional resources.

    • Funds for condo buy-in can be committed by deans or chairs during faculty recruitment and retention.

  • Instructional Computing Allowances (ICAs) provide free access for students in classes.

More details can found on our extensive Savio documentation site.