Just ten months ago, the High-Performance Computing Center of the University of Stuttgart (HLRS) inaugurated its new “Hawk” supercomputer, which made its debut on the Top500 list last month as the 16th most powerful publicly ranked supercomputer in the world. Now, HLRS is announcing an upgrade to the already powerful system: an array of new nodes powered by Nvidia GPUs.

At launch, Hawk – an HPE system – comprised 5,632 CPU nodes equipped with dual AMD Epyc Rome 7740 CPUs, a total 1.44 petabytes of memory, Mellanox InfiniBand HDR200 networking and 25 petabytes of DDN disk storage. Through the upgrade, this hardware will be supplemented by 24 HPE Apollo 6500 Gen10 Plus systems equipped with 192 Nvidia A100 GPUs — around two petaflops of new peak double-precision computational power.

HLRS says that the upgrade will provide “120 [additional] petaflops of AI performance” in response to user interest in AI applications, and Hawk’s users will also be able to access Nvidia’s NGC catalog, which provides GPU-optimized software for AI and HPC applications. 

“At HLRS our mission has always been to provide systems that address the most important needs of our key user community, which is largely focused on computational engineering,” said Michael Resch, director of HLRS. “For many years this has meant basing our flagship systems on CPUs to support codes used in computationally intensive simulation.”

Hawk, however, arrived right on the heels of HLRS’ 2019 installation of an Nvidia GPU-based Cray CS-Storm system – which HLRS reports has been very popular and is currently nearing capacity.

“It turned out to be an extremely popular system because there’s a growing community of people who understand AI has a benefit for them,” Resch said. “By the middle of this year it was clear we had to expand to cover our growing AI requirements.” Now, he says, “adding this second key type of processor to Hawk’s architecture will improve our ability to support scientists in academia and industry who are working at the forefront of computational research.”

“Once NVIDIA GPUs are integrated into Hawk, hybrid workflows combining HPC and AI will become much more efficient,” added Dennis Hoppe, who leads artificial intelligence operations at HLRS. “Losses of time that occur because of data transfer and the need to run different parts of the workflows in separate stages will practically disappear. Users will be able to stay on the computing cores they are using, run an AI algorithm, and integrate the results immediately.”

At launch, HLRS emphasized Hawk’s potential for sustainability applications, such as wind turbine and power plant optimization and aerodynamics work. Now, HLRS is highlighting new projects already under development at the center that combine deep learning and HPC: better simulations of material performance for sheet metals (such as those used in vehicle manufacturing); better understanding of how deflagration turns into detonation (as in the August explosion in Beirut); and better fluid dynamics for a variety of fields.

Of course, Hawk now finds itself in a very different world than it did ten months ago, and its work will not be all sheet metals and power plants – HLRS is planning to use its new AI capabilities to assist a project attempting to estimate when ICU beds will fill up in European hospitals as a result of the COVID-19 pandemic. 

“It’s a critical question with so many people dying — we’ve seen scenarios in places like Italy, New York and Wuhan where ICUs filled up in the first weeks of pandemic,” Resch said. “So, we will conduct simulations and predictions of the outlook for the pandemic over the next weeks and months, and GPUs will be extremely helpful for that.”

The upgrade, with systems again provided by HPE, deepens HLRS’ working relationship with the company. At Hawk’s inauguration, Heiko Meyer, HPE’s chief sales officer, signaled a “a long-term development partnership with HLRS in which we optimize applications, test future technologies, and bring them to a mature, market-ready state.”

Header image: the Hawk supercomputer.