[Berzelius-users] Inviting project proposals for allocation on Berzelius-Hopper

[Filip Polbratt]

Dear Berzelius Users,

Following the latest expansion of the Berzelius service, 
Berzelius-Hopper is now available to users. Berzelius-Hopper consists of 
16 NVIDIA® DGX-H200 compute nodes[1] (128 GPUs in total) and 8 CPU 
nodes. It is primarily intended for GPU-intensive workloads, 
particularly those that benefit from H200 GPUs and large GPU memory. You 
can find the complete system configuration[2] on the NSC website.

Access to Berzelius-Ampere (the previous Berzelius system) does not 
automatically include access to Berzelius-Hopper. We therefore encourage 
existing Berzelius-Ampere users, as well as new prospective users, to 
apply for access to Berzelius-Hopper. Proposals should be submitted via 
SUPR, following the available guidelines[3].

If a PI already have an allocation on Berzelius-Ampere and would like to 
access Berzelius-Hopper, we encourage the PI to submit a continuation 
proposal requesting both instead of a new project proposal for access to 
Berzelius-Hopper. It can be done by cloning an existing proposal, 
instead of waiting until the project end date.

Please note that Berzelius-Hopper is a smaller resource compared to 
Berzelius-Ampere with shared storage, and we therefore cannot guarantee 
approval of a large number of proposals.

If you have any questions, please feel free to contact us.

Best regards,
Filip Polbratt
NSC

[1] https://www.nvidia.com/en-us/data-center/dgx-h200/

[2] https://www.nsc.liu.se/systems/berzelius/

[3] https://www.nsc.liu.se/support/systems/berzelius-resource-allocation/