HPC News at the University of Bonn

Second update April 25, 2024:

All filesystems are now working normally again

Update April 25, 2024:

The workspaces and other filesystems are still not working correctly. We will let you know when the situation changes.

Update April 24, 2024:

Marvin is now available again. Since the shutdown happened suddenly, we highly recommend that you check all your jobs, as some of them might have been terminated prematurely.

You can use all queues again, however all queues have only half their nodes online since we still cannot run with full cooling capacity. You do not have to do anything different than normally, it just might result in longer waiting times.

We will let you know if the situation changes.

Update: Marvin has to stay down until at least Tuesday noon (April 23, 2024). We will notify you of all further developments

Dear HPC users,

Marvin was shut down over the weekend due to a cooling failure. We will notify you when it is available again.

We apologize for the inconvenience.

Dear HPC users,

due to a firmware upgrade we need to schedule a Marvin downtime for Tuesday, March 26, 2024. It will start at 9:00 and most likely take around four hours.

No job that is currently running should be affected. Any jobs that are in the waiting queue on that day will simply start running after the maintenance window.

We will notify you when the maintenance work has been completed.

Dear HPC users,

we are happy to announce that you can now register for using Marvin. Below are the most important things you need to know:

Testing phase for experienced users will be re-launched on February 13th

"Marvin", Uni Bonn's supercomputer, installed in 2023, is nearing its final release day. The cluster, consisting of 192 MPP nodes with a total of 18400 compute cores plus an additional 300 NVIDIA GPUs, had been presented to the public in a festive inauguration ceremony on October 20th, marking the start of an intense testing phase allowing highly experienced "power-users" to already use the cluster for numerical simulations while at the same time reporting any potential problems or hickups to the maintenance team at the Computing Centre. This first testing phase started overwhelmingly positively, showing that the performance of the cluster is even higher than had been previously promised by the vendor Megware.

We regret that we have to postpone the opening of Marvin to everyone. It will be delayed to at least the end of February. The closed testing phase will most likely resume before that date.

This delay is due to necessary additional work on the cooling infrastructure and monitoring hardware.

As always, we will keep you up to date via at least our two main communications channels: our website https://www.hpc.uni-bonn.de and the hpca-forum@listen.uni-bonn.de mailing list.

Best regards

Your HPC Team

Dear HPC users,

the HPC/A Lab presents our new training course "Advanced Machine Learning". It continues where the Foundations of Machine Learning Course stopped. We will cover second-order optimization, derivative evaluation, generative learning, reinforcement learning and more.

This one-week course will take place in the week of March 11-15, 2024. For more information on the course topics, prerequisites and to register, visit the course page on eCampus:

https://ecampus.uni-bonn.de/goto_ecampus_crs_3255322.html

We are looking forward to seeing you during the course.

Your Lecturers, Elena and Moritz

To free up disk space on Bender, we will soon delete orphaned user data. Initially this will only be data from users who do not exist at the university anymore.

If you know of data from a former colleague that you might require, please contact us.

If you yourself are not using Bender anymore, please consider deleting your old data or notifying us about what else can be deleted.

We will not delete any data before February 15.