Stanford
Carina Computing Platform

Carina 2.0 FAQ

Carina Logo

Key Dates

January 13, 2026: Carina 2.0 Launch

  • Carina 2.0 becomes the primary system.
  • All new projects and PI requests will be provisioned on Carina 2.0.
  • A snapshot of all project data from Carina 1.0 will be migrated to Carina 2.0.
  • All current Carina users will have an account on Carina 2.0.

February 18, 2026: Open OnDemand Workshop

  • A hands-on workshop to introduce the new Open OnDemand interface and its features.

December 31, 2026: Carina 1.0 Decommissioned

  • The Carina 1.0 system will permanently shut down. Any data remaining on Carina 1.0 will be inaccessible after this date.

Data Migration

Q: How will my data be moved to Carina 2.0?

Carina 1.0 and 2.0 are completely separate systems, so data must be actively migrated. The process differs for your personal $HOME directory and your shared project folders.

Shared Project Directories (Migration Managed by the Carina Team):

  • The Carina team will perform a comprehensive copy of all shared project directories from Carina 1.0 to Carina 2.0 just before the January 13th launch. If you continue to work on Carina 1.0 after this date, please contact us when you are ready to fully transition, and we will help you sync your most recent project data.

Individual $HOME Directories (Migration Managed by You):

  • You are responsible for migrating your personal $HOME directory. You can easily do this using standard command-line tools such as rsync or scp. A step-by-step guide will be published with the launch of Carina 2.0.

Q: What can I do to prepare my data for a smooth migration?

This is an excellent opportunity to perform some digital housekeeping. To ensure a faster and more efficient migration, we recommend you:

  • Review your $HOME and shared project directories.
  • Delete or archive old files, expired datasets, and unneeded folders.
  • Start 2026 with a clean and organized workspace on a more powerful system.

New Features & What to Expect

Q: Do I have to learn a completely new system?

No. Your core workflows will remain the same. The biggest change is the web dashboard. Carina 2.0 replaces JupyterLab with Open OnDemand as its primary web portal. 

Key benefits of this upgrade include:

  • You can still launch JupyterLab sessions directly from the Open OnDemand dashboard.
  • You can now manage and submit Slurm jobs directly from your web browser.
  • It provides a more robust and flexible environment for accessing the cluster's resources.
  • Open OnDemand is used on other Stanford clusters, including Sherlock and Farmshare.

 

We will hold a workshop in February to help everyone get comfortable with Open OnDemand.

Q: Will Carina 2.0 be more stable than 1.0?

Absolutely. Carina 2.0 is built on next-generation hardware and software, incorporating all the lessons we learned from operating Carina 1.0. We've focused heavily on improved stability, monitoring, and overall reliability. Here's how we're addressing the most common complaints:

"Why can't I get a GPU? They're always in a 'drain' state!"

Ever tried to grab a GPU only to find it's unavailable? This happens when a previous job didn't exit cleanly and the system took the GPU offline. We recently implemented a clean-up script that runs after every GPU task. It's showing great results on Carina 1.0, and we're building it into Carina 2.0 from the start.

"My jobs are so slow, especially when I have tons of files!"

Does your work slow to a crawl when dealing with thousands of small files? The current storage system, Ceph, isn't great at handling that workload. For Carina 2.0, we're swapping it out for a much faster PowerScale (Isilon) system. If you've used Sherlock, you'll recognize it—it handles huge numbers of files with ease and should make your jobs run much more smoothly.

"The login nodes are so laggy!"

Sometimes the login nodes are so sluggish thatit's impossible to work. This happens when someone runs a heavy computing task on a login node (usually by accident), consuming all the resources. Carina 2.0 will prevent this by enforcing sensible limits and reminding users to use compute nodes for computing tasks.

"Why does Carina seem so unreliable and go down so often?"

We know Carina has had unacceptable stability issues and downtime. The main culprit has been the Ceph storage system—it's both slow and fragile. By replacing it with the reliable PowerScale system and protecting the login nodes from overload, we expect Carina 2.0 to be much more stable and predictable.

"Moving my data to and from Carina takes forever!"

Currently, data transfers share bandwidth with everyone's login activity. Carina 2.0 will have a dedicated "express lane" for data: a brand new Data Transfer Node (DTN). We'll also offer Globus on Carina 2.0 as soon as it's approved for use with high-risk data at Stanford.

Upgrades for PIs

It's been difficult for PIs to keep track of who has access to their projects. We're fixing this by adopting the Sherlock model: PIs will receive quarterly emails with a list of all current group members who have access to Carina. 

We are roadmapping other requested features, including usage statistics.

System Capabilities & Specs

Q: Does Carina 2.0 have GPUs?

Oh yes. Carina 2.0 has exciting GPU resources available. The cluster includes 40 NVIDIA L40S GPUs, which are among the most powerful modern GPUs.

Q: What are the technical specifications of the cluster?

Carina 2.0 is comprised of 5 Dell PowerEdge XE7745 compute nodes, each equipped with:

  • 8 NVIDIA L40S GPUs
  • 1.5 TB of RAM
  • ~28 TB of local NVMe storage

The system is housed in the energy-efficient Stanford Research Computing Facility (SRCF) on the SLAC campus, leveraging its robust power and cooling infrastructure.

Q: Can I store and process NIH Controlled-Access Data?

Yes. Carina 2.0 is certified for high-risk data, including NIH Controlled-Access Data. For detailed information, please see FAQ: Where to use NIH-controlled access data at Stanford. 

Access, Cost, and Support

Q: Who is eligible to use Carina 2.0?

Carina 2.0 is available to any Stanford researcher with Principal Investigator (PI) privileges, as well as members of their research teams. The platform is exclusively for research involving high-risk data.

Q: How much does it cost?

Use of Carina 2.0 is free for eligible researchers. The hardware, services, and baseline storage allocation are fully subsidized.

  • Default Storage: Each project receives 1 TB of storage at no cost.
  • Additional Storage: If you require more space, you can purchase it in 10 TB increments for $97.50/month. Extra Storage Order Form