The Anvil login and GPU (Nvidia A100 GPUs) nodes, along with AnvilGPT, Anvil Notebook, and storage will stay online.

How Does This Affect You? Anvil CPU and AI nodes, along with Anvil Composable, will be unavailable during all outage windows. Jobs on CPU and AI nodes will not run, and new jobs to CPU and AI nodes will not start while they are offline. Jobs on CPU and AI nodes that cannot be completed before an outage begins will not start. They will remain in the queue until these nodes returns to service.

Any Questions? If you have questions about these outages or need assistance, please submit a ticket through the ACCESS Help Desk.

What’s happening: Our engineers identified a problem with the Gautschi scratch filesystem that requires an emergency upgrade and repair. The procedure must be done with the filesystem offline and is expected to take 8-10 hours or more. To avoid any risk of further damage, we have stopped new jobs from starting on Gautschi as of Monday afternoon, November 2.

Timeline

Now: No new jobs will start.

Tuesday, November 3, 8:00 AM: Remaining non-interactive batch jobs will be requeued, and interactive jobs (including OnDemand sessions) will be ended. Scratch will be unmounted and work will begin.

Estimated completion: Late Tuesday, though it may run longer.

As the renovation and capacity improvement project in the MATH datacenter moves into its final stages, a planned maintenance outage is required to continue bringing the newly installed power and cooling infrastructure online. This work is part of a broader renovation that will result in a 32% increase in floor space, a 60% increase in usable power, and a two-fold increase in cooling capacity — all crucial facility ingredients needed to continue providing world-class high-performance computing resources to Purdue faculty and the nation.

**Outage Window: **Saturday, September 19, 2026, 12:00 AM – Sunday, September 20, 2026, 6:00 PM

What will remain online

The following resources are not affected by this power work and will remain available throughout the maintenance window:

• Data Depot and Data Depot Object
• ScienceDMZ
• Scholar
• Fortress
• HubZero
• Hammer
• Weber

What will go offline

The following resources, including their associated Data Depot storage and Data Depot Object allocations, will be unavailable during the outage window and will return to service gradually once the power work is complete:

• GenAI Studio
• Geddes
• Anvil
• Gautschi
• Gilbreth
• Bell
• Negishi
• Rowdy
• Rossmann

How does this affect you?

Jobs on the affected clusters will not run, and new jobs will not start during this window. Jobs that are expected to not complete before the maintenance begins will not start until after the maintenance window ends. Resources listed above as remaining online will continue to operate normally throughout.

Questions

If you have questions about this outage or need help from our support staff, please contact rcac-help@purdue.edu.

Best, RCAC

The scheduled maintenance on Anvil has been successfully completed as of 3:37 pm EDT. All services are now available. The queues have been enabled and jobs have resumed scheduling. If you encounter any issues following this maintenance, please contact Anvil team through ACCESS Help Desk.

**Updates (September 18, 2026): **

This is a reminder that Anvil is scheduled for a power outage and maintenance starting tomorrow, Saturday, September 19, at 12:00 a.m. through Monday, September 21, at 8:00 p.m. EDT.

Updates (September 10, 2026):

This is a reminder that Anvil is scheduled for a power outage and maintenance on Saturday, September 19, at 12:00 a.m. through Monday, September 21, at 8:00 p.m. EDT.

Updates (August 13, 2026): The power outage originally scheduled for Saturday, August 15, at 12:00 a.m. through Monday, August 17, at 5:00 p.m. EDT has been cancelled, therefore Anvil remains available during this time.

Anvil is still scheduled for a power outage and maintenance on Saturday, September 19, at 12:00 a.m. through Monday, September 21, at 8:00 p.m. EDT.

Original post (August 7, 2026): The Anvil clusters and services will be unavailable during two scheduled maintenance events in August and September to support datacenter improvement projects that require power outages.

The first power outage is scheduled for Saturday, August 15, at 12:00 a.m. through Monday, August 17, at 5:00 p.m. EDT.

The second power outage and Anvil maintenance are scheduled for Saturday, September 19, at 12:00 a.m. through Monday, September 21, at 8:00 p.m. EDT.

The Anvil quarterly maintenance previously announced for August 25 has been cancelled, and Anvil willl remain available on August 25.

How Does This Affect You?

Anvil CPU, GPU, and AI clusters, along with AnvilGPT, Anvil Notebook, Anvil Composable, and storage, will be unavailable during all outage windows. Jobs will not run, and new jobs will not start while Anvil is offline. Jobs that cannot be completed before an outage begins will not start. They will remain in the queue until Anvil returns to service.

Any Questions?

If you have questions about these outages or need assistance, please submit a ticket through the ACCESS Help Desk.

Original Post: The Anvil cluster will be unavailable on Tuesday, August 25, 2026, from 8:00 AM to 5:00 PM EDT while scheduled maintenance is performed. During this maintenance window, we will update the operating system and kernel, storage and file systems, system drivers, and software packages.

Any Slurm job requesting a walltime that extends beyond Tuesday, August 25, at 8:00 AM EDT will not start and will remain in the queue until the maintenance is complete.

Anvil is expected to return to full production by Tuesday, August 25, 2026, at 6:00 PM EDT.

Please submit a ticket through ACCESS Help Desk if you have any questions.

We are investigating the issue. The cause and an estimated time to resolution are not yet available. We will provide an update as soon as possible or if service returns to normal.

There is no ETA for restoration at this time. We will update as soon as we have more information or if the power is restored.

We will provide another update by 9:00 AM tomorrow.

We will provide an update at 3 PM.

Users may see failures connecting via Windows/Samba and jobs are affected because all clusters are currently paused while recovery work proceeds. Our engineers are working with the storage platform to restore service; we will share further updates as we learn more.