Science IT & The Berkelium Kubernetes Service

Science IT & The Berkelium Kubernetes Service

Empowering Scientific Computing at Berkeley Lab

The Science IT Department at Lawrence Berkeley National Laboratory (LBNL) partners with domain scientists across all research areas—including Physical Sciences, Energy Sciences, Computing Sciences, Earth & Environmental Sciences, and Biosciences—to deliver modern, reliable, and scalable cyberinfrastructure.

Our mission is to lower the barrier to computational science, providing consulting, infrastructure management, and specialized compute platforms tailored to experimental, observational, and simulation workloads.

What is Berkelium?

Berkelium is LBNL’s centralized, enterprise-grade multi-tenant Kubernetes cluster, engineered specifically for scientific computing workflows. Named after element 97 synthesized at Berkeley Lab, Berkelium bridges the gap between traditional HPC (Slurm batch scheduling) and cloud-native computing.

Key Highlights of Berkelium:

  1. Interactive & Batch Workloads: Seamlessly host continuous services (APIs, web portals, databases, live experiment dashboards) alongside batch compute jobs.
  2. GPU Acceleration: Access shared and dedicated NVIDIA A100, H100, and L40S GPUs with hardware-accelerated drivers and CUDA support.
  3. Institutional Single Sign-On: Secure multi-tenancy integrated directly with LBNL OneID / LDAP credentials and role-based access control (RBAC).
  4. Direct Parallel Storage Connectivity: High-throughput access to LBNL research filesystems, Ceph object storage, and persistent block volumes.
  5. Self-Service Portal: Automated namespace generation, resource quota monitoring, and instant kubeconfig downloads.

Science IT Services & Support Matrix

Science IT provides comprehensive services for the laboratory research community:

  • Containerization & Workflow Modernization: Assistance converting legacy codes, Singularity/Apptainer images, and Docker containers to production Kubernetes deployments.
  • AI/ML Infrastructure: Accelerated Ray clusters, PyTorch distributed training, Kubeflow pipelines, and Hugging Face model hosting.
  • Data Pipelines: Stream ingestion from beamlines, detector arrays, and sequencing machines.
  • Office Hours & Consultations: Weekly one-on-one technical office hours with Science IT engineers.

For support or custom architecture reviews, reach out to our team at scienceit@lbl.gov or join #berkelium-users on LBNL Slack.