Using Hive

Hive is a newer, bigger HPC@UCD cluster. The lab uses it alongside Farm. It works almost exactly like Farm (Ubuntu 22.04, Slurm, the same modules and conda), so read Using Farm first. This page covers only what's different. Official docs: https://docs.hpc.ucdavis.edu/hive/.

Last checked: September 2026.

Why use Hive

• Same lab storage. /quobyte/jrigrp is mounted on both clusters, so there's no need to copy data between them. Your /home is separate on each cluster (20 GB on Hive).
• More resources. Nodes have 64–224 CPUs and 256 GB–2 TB RAM, and there are many more GPUs (A100, H100, L40S, A6000, RTX 6000 Blackwell, and others).
• Different limits. When the lab's 608-CPU share on Farm high is full, Hive is often the better place for non-restartable jobs.

Account and login

Request an account in HiPPO (log in with CAS):

1. Pick the cluster: choose Hive.
2. Pick a sponsor: lab members search for Jeff (Ross-Ibarra). People with no sponsor can pick "HPC@UCD Sponsored Public Access" (publicgrp), which is free for UC Davis staff, faculty, and grad students.
3. Pick access: sshkey for the command line (you can reuse your Farm public key; Hive also accepts your campus passphrase), and/or OpenOnDemand for the web interface.
4. Wait for the email: your sponsor approves, and the account is ready about an hour after it's created.

Add the Genome Center group. Once you're on Hive, log in to HiPPO and request an additional group affiliation, choosing genome-center-grp. You need it to use the lab's main Hive allocation (below).

Host hive
    HostName hive.hpc.ucdavis.edu
    User USERNAME
    ServerAliveInterval 60

OnDemand (RStudio, Jupyter, VS Code) is at https://ondemand.hive.hpc.ucdavis.edu. Use this one for lab data. Hive's OnDemand can see /quobyte/jrigrp, and Farm's can't.

Accounts and partitions

The account names differ from Farm. There's no jrigrp account on Hive. The lab uses:

Account	Partition	Group limit	Per-user / per-job limit	Use it for
genome-center-grp	high	1,303 CPUs, 13,978 GB (shared by all Genome Center groups)	64 CPUs, 1,024 GB per user	Main choice for non-preemptible CPU jobs
genome-center-grp	gpu-a100	128 CPUs, 2,000 GB, 8 GPUs	—	GPU jobs that must not be killed
publicgrp	high	128 CPUs, 2,000 GB, 5 GPUs (shared by all free-tier users)	8 CPUs, 128 GB, 1 GPU per job	Small non-preemptible jobs
publicgrp	low	none	none	Big arrays, restartable jobs, overflow, any idle GPU

(There's also a classgrp account on some accounts; it's for courses, so don't use it for research.)

As on Farm:

• high jobs are never preempted. The max time is 30 days (Farm allows 150).

• low jobs can be killed and requeued at any time. The max time is 7 days. See Farm Tips and Tricks for making jobs restartable.

• Your default account is publicgrp, so always set -A:

    #SBATCH -A genome-center-grp
    #SBATCH -p high
  

• The default memory is 4 GB per CPU (2 GB on Farm). Set --mem explicitly anyway.

• GPU example: sbatch -A genome-center-grp -p gpu-a100 --gres=gpu:1 -c 8 --mem=64G job.sh, or use -A publicgrp -p low --gres=gpu:1 for any idle GPU.

To see your current limits:

/opt/hpccf/bin/slurm-show-resources.py --full

All the monitoring commands in Farm Tips and Tricks (squeue, sacct, sprio, the myeff function) work the same on Hive. Just swap in the right account name, e.g. squeue -A genome-center-grp -p high.

Help

Email hpc-help@ucdavis.edu for Hive (use farm-hpc@ucdavis.edu for Farm) and include the same details as in Using Farm.