HPC2020 FAQs

Based on my Hopper usage, how much capacity should I purchase in the HPC2020 cluster?
Please review the Hopper Monthly Usage Report to assist in purchasing decisions for the new cluster. Specifically, the YTD Hopper Usage v. Purchased Capacity for 2020 is a good place to start. For example, if your lab group owns 3% of the cluster, but uses 6%, this suggests that you need twice the capacity in the new cluster ( a usage factor of 2 ). Here’s a general formula:
Hopper nodes x Cores/node x Hopper Usage Factor / ( Cores/node of node purchase in HPC2020 )

For example:
3 Hopper nodes x 20 cores/node x 2 / 48 cores/node in new cluster
= 2.5 nodes ( round up to 3 )

Will there be access to the new cluster for researchers who are not investors?
The new cluster will operate much the same as the Hopper cluster in which only investors and their sponsored accounts have access.

When is the deadline for submitting purchase orders for the new cluster?
Price quotes from Dell are valid thru July 31, 2020. Orders should be submitted by then. We expect to place the initial order the first week of August 2020.

I’m new to HPC, how should I buy-in to HPC2020?
We recommend a minimum of two nodes or six cores per lab group member whichever is higher. At least two nodes will allow MPI-enabled jobs while remaining in the investor’s purchased capacity. Six cores per member will allow sufficient capacity for the group as a whole within the purchased capacity.

Will there be other opportunities to purchase capacity outside the initial purchase?
We plan to accept orders up to 18 months after the go-live date, but pricing will most likely be higher. As in the Hopper cluster, the initial investors will get the most for their money.

Do I need a GPU node?
GPU nodes are best suited for specialized workloads that involve similar parallel operations. In addition, code must be written in a language specifically for GPU nodes like CUDA or OpenCL. Unless your program/workload fits this criteria, you probably are best served with Intel nodes.

Do I need an AMD node?
The AMD node is going to have a larger number of cores (128) with a lower clock speed compared to that of the Intel nodes which will have 48 cores. Because it really depends on the workload, it is hard to say which machine will perform better for you. Generally, we recommend purchasing the standard Intel node as a very capable general use system.

As an investor, how are others in my lab granted access?
Much like on Hopper, members of lab group will submit an account request via an online form on the HPC Portal. They will select a sponsor and that sponsor will receive an email asking them to approve or deny the request. If approved, the account will be automatically created and the lab member notified. These lab members will have dedicated access to the sponsor’s purchased capacity via a reservation.

Am I allowed to install software in the new cluster?
Much like on Hopper, software can be installed by any user in their home directory or it can be installed by the HPC sysadmins for global use by using the software request form on the HPC Portal.

Can sensitive or regulated data ( e.g., CUI, Classified, HIPAA, PHI, PII, ITAR/EAR ) be stored on the new cluster?
No, not without first formally notifying the HPC sysadmins via email to confirm that the specific security requirements of your data can be met.