New cluster Pelle ready for your test use open

Dear UPPMAX user,

If you’re a local Uppsala University user awaiting access to our new cluster Pelle, this message is for you. If not, I just want to wish you a nice and pleasant summer.

The contract to procure Pelle was signed by the vice-chancellor in March 2024. After long delays with the vendor, mainly due to network problems uncovered in our initial qualification tests, the delivery of the system was finally approved in February 2025.

Since then, we have been preparing the new system for production use, in parallel with keeping our existing systems up and running. For the new 20PB storage system Gorilla, we still have open support cases regarding hardware reliability. For now, Gorilla is therefore not available.

The cluster itself, however, is. This is still a testing phase. After successful runs with some pilot users, we’ve now allowed login and job submission access for all users who are members of local UPPMAX projects with allocations. As the storage is still based on the Rackham storage Crex, you need to use your existing project and home directory storage on the new system.

In the cluster, you have access to 4 H100 GPUs, 40 L40s GPUs, two high-mem nodes and over 100 new CPU nodes, each with 48 cores and 768GB of RAM. This means that tasks that used to require “fat” nodes can now run on any node. Each core is also substantially faster than what you’re used to. Later on, we’ll add some compute hardware from our existing clusters, such as our workhorse T4 GPUs for light testing and surge CPU capacity.

This is a new software setup, in addition to the new hardware. The Slurm configuration and modules are not completely finalized. If you have compiled any code yourself, you probably want to recompile that to make use of the more recent CPUs. Sometimes, modules have new names and a new OS version means that other things can also change. Naturally, you will also need to update your project name in any job scripts, and probably consider what core counts and time limits you specify based on the new memory and CPU characteristics.

We provide a short guide Feel free to check it out using ssh to pelle.uppmax.uu.se or ThinLinc. Any feedback is welcome to support@uppmax.uu.se. Still, do note that this is a test phase and that our staffing during the next few weeks is limited due to vacations. For now, Rackham is still our production system. If something you want to do doesn’t work on Pelle, the immediate response is “try it on Rackham instead”.

For sensitive data, we will use a section of this hardware for our new setup Maja, with surge overflow from both onto older CPU nodes. The intent is that the software environment there will be quite similar, just like the software setup on Bianca is similar to the one on Rackham. Expect updates on the progress for Maja during the fall, together with our migration strategy for the new storage projects on Gorilla.

When we’re talking about testing stuff, you might also want to check out lab.uppmax.uu.se for those times when you just want to spin up a notebook in Julia, R, or Python.

Have a nice summer, and Happy Computations!