Berkeley Research Computing runs the Savio high performance cluster in collaboration with staff from LBL. Department members can make use of the cluster in a few ways:

The Statistics Condo

We have two regular and two GPU nodes (with a total of 8 GPUs, four per node) that serve as a beachhead onto the larger system. The nodes are freely available to all members of the department and there is no yearly core-hour time limit. Please see the Savio webpage for instructions on getting an account and submitting jobs, and email if you have questions. When filling out the account request form, please specify the 'co_stat' account.

Let us know if you are interested in expanding the size of the departmental condo.

Submitting Jobs

When submitting jobs to the condo nodes, specify the 'co_stat' account and either the 'savio2' or 'savio2_gpu' partition. Note that the 'savio2' partition is scheduled on a per-node basis, while the 'savio2_gpu' partition is scheduled based on units of {1 GPU + 2 CPUs}.


  • sbatch -A co_stat -p savio2
  • sbatch -A co_stat -p savio2_gpu -o log.txt --gres=gpu:1 --cpus-per-task 2


If you need more cycles, you can "burst" onto unused campus nodes as Low Priority jobs. The limitations are that when the system is busy, these jobs will be queued and:

  • yield to (be passed by) other jobs with higher priorities,
  • get preempted (interrupted) by pending higher priority jobs.

    Preempted jobs can choose whether the job should be simply killed, or be automatically requeued after it's killed, at submission time. Please note that, since preemption could happen at any time, it would be very beneficial if your job is capable of checkpointing/restarting by itself, when you choose to requeue the job. Otherwise, you may need to verify data integrity manually before you want to run the job again.

Faculty Compute Allowance (FCA)

Each faculty member receives 300,000 "service units" (approximately 300,000 core cpu hours) per year to run jobs at the regular priority. If you run a job for 10 hours on a 32-core machine, it will cost approximately 320 core hours (cost varies depending on the exact type of machine being used). If your job only utilizes 4 cores on the same 32-core node, it will still cost 320 core hours since your job has exclusive access to the node, unless run in either the savio2_htc or savio2_gpu pools, which are scheduled on a per-core basis.

Faculty members can request an FCA here. Once the FCA is created, researchers working with the faculty member (including graduate and undergraduate students, postdocs, and colleagues outside of Berkeley) can request accounts that can make use of the FCA.


Once you go beyond the FCA you can purchase more time through recharge at $0.01/hour, e.g. $2,000 for an additional 200,000 core hours.


Your Savio account has access to a 10 GB home directory, 200 GB of shared project space (shared amongst all department members), and a large scratch area. You can purchase more if needed, including large amounts of disk space through the condo storage program, which costs about $6200 per 42 TB, good for five years.