Dear Researchers, Colleagues, and Students,

In 2017, a group of 51 UMBC researchers was successful in competing for an MRI (Major Research Infrastructure) grant from the NSF. Having this many faculty -- by my estimate one eighth of all UMBC research/teaching faculty! -- is a powerful signal of the importance of HPCF to the campus and well-appreciated by the NSF program directors. This was in fact the third successful MRI for HPCF, after 2008 and 2012, which in total provided over $1M in funding from the NSF to this campus initiative.

The support of $789,076 financed the extension and update of HPCF with the 2018 equipment, comprising 44 CPU nodes with two 18-core Intel Skylake CPUs and 384 GB of memory each, one node with four NVIDIA Tesla V100 GPUs connected by NVLink, and 8 Big Data nodes with 48 TB disks across 12 hard drives each. The total includes $236,723 cost-sharing funded by UMBC, and we acknowledge this excellent cooperation, in particular with DoIT.

The 3-year grant has formally ended, and an Outcomes Report provides a two-page highlighted summary of the achievements during the funding period. A PDF file of this report from 2020 is now posted on the HPCF webpage under the Research tab, adding to the corresponding report from 2015. I thank the colleagues Daniel Lobo, Curtis Menyuk, Jerimy Polf, Jianwu Wang, Meilin Yu, and Zhibo Zhang for providing the visuals and project information in the Outcomes Report. It represents a wonderful cross-section of the activity using taki.

Matthias Gobbert, Department of Mathematics and Statistics, gobbert@umbc.edu

(Chair of the HPCF Governance Committee and PI of the MRI grants)

Dear Researchers, Colleagues, and Students,

User support in collaboration with system administrators have developed visual representation of monthly usage levels of taki's CPU and GPU clusters. You can find this data at hpcf.umbc.edu under the Research tab, where it will be updated every month.

The raw data are in units of core-hours for the CPU cluster and GPU-hours for the GPU cluster, separated by architecture denoted by the year of acquisition. Stacked bar charts add up to the total available hours, while bar charts and line plots show the percentage of use on each architecture. Analysis by most active PI groups are shown in various styles including line graphs.

The visuals show that decommissioning the 2009 CPU nodes in October did not change the usage levels. They also show that usage levels saw a very precipitous drop after spring break 2020 and that research productivity has recovered fully since then. This goes along with continued increase in use of the 2018 hardware in the CPU cluster. The analysis of most active PI groups indicates shifts over time and brings out that most groups use either the CPU or the GPU cluster.

As regular reminder, please send me references to publications! We can also post other newsworthy stories here, so feel free to contact me.

Matthias Gobbert, Department of Mathematics and Statistics, gobbert@umbc.edu

(Chair of the HPCF Governance Committee and PI of the MRI grants)