top of page

Tier-1 Compute, past, present, and future

Decommissioning of BrENIAC: A Look Back at its Impact



The Tier-1 platform of the Vlaams Supercomputer Center (VSC) has three components, Compute, Cloud, and Data. The Compute platform already has a history spanning more than a decade, starting with the inauguration of the first generation, called muk, in 2012, hosted by Universiteit Gent. The second generation, BrENIAC, was inaugurated in 2016 and hosted by KU Leuven. At present, the third generation, Hortense, is in production again hosted in the data center of Universiteit Gent, while the preparation of the procurement process for its successor has already started. That system will be hosted by the Vrije Universiteit Brussel.


Recently, the second-generation system BrENIAC was decommissioned, and that is a good opportunity to reflect on the past, and present of the Tier-1 Compute platform and how its use changed over time.


When BrENIAC entered production in 2016 it was ranked 197th on the June Top-500 list of supercomputers. The system had 16,128 cores and clocked in at 548 TFlops/s (Rmax), 619 TFlops/s (Rpeak), using 255 kW of power. The heat produced by the system was recaptured to heat nearby buildings of the university. The system was delivered by NEC and the process of installation and testing went very smoothly. The original 580 nodes had 2 Intel Broadwell CPUs with 28 cores each, a quarter of them 256 GB RAM while the other nodes had 128 GB RAM. The interconnect was Mellanox Infiniband EDR.


For comparison, the system that ranked number one in the Top-500 of June 2016 was the Chinese Sunway TaihuLight with a whopping 10,649,600 cores, 93,015 TFlops/s (Rmax), 125,436 TFlops/s (Rpeak), using 15,371 kW of power, so almost 170 times faster than Breniac.


Only very large systems last long in the Top-500, and in November 2016 BrENIAC ended at place 278, while it ranked 371st in the ranking of June 2017. In November of that year, BrENIAC was no longer one of the 500 fastest supercomputers in the world.

Midlife, the system was expanded by another 408 compute nodes with 2 Intel Skylake CPUs, 28 cores each, and 192 GB RAM. This addition was greatly welcomed by its user base since demand for computing time was rising.


BrENIAC continued to serve as the VSC’s Tier-1 Compute platform even after its successor, Hortense, was taken into production. The last nodes were switched off in December 2022. BrENIAC has faithfully served the VSC user community for 6 years. However, the story doesn’t end there. Some of the nodes or network components got a new lease on life at Universiteit Gent in the Tier-1 Cloud platform and at Universiteit Antwerpen, at VUB and at KU Leuven.


Considering the evolution of the number of CPU hours, it is very clear that BrENIAC has been well-used, and that it has met an increasing demand for computing power (see figure 1). This trend continues to this day after Hortense, the third-generation Tier-1 compute system, replaced BrENIAC.



A Graph showing the Evolution of CPU Hours of BrENIAC
Evolution of CPU Hours

The Tier-1 Compute platform serves researchers from various scientific domains, as illustrated in the figure below. The “traditional” HPC domains such as physics, chemistry, and engineering still dominate the usage, but it is interesting to note that also projects in the humanities make it onto our most powerful regional computing infrastructure. The scientific domains were also represented on the system itself, a renowned graphic artist designed a decal that has been on the rack doors since the inauguration event.


A Graph showing the Total Allocated Core-Hours from BrENIAC of various scientific domains2017-2023 of
Total allocated computing hours per research domain

To the credit of both NEC which delivered and supported the system and the VSC team at KU Leuven that managed the infrastructure, BrENIAC has been running smoothly with few glitches and only relatively short downtimes.

Currently, Hortense hosted at Universiteit Gent is doing an excellent job serving as the VSC’s current Tier-1 compute platform, adequately meeting the demand for more computational resources.

Comments


bottom of page