What is High Performance Computing?

High Performance Computing (HPC), also known as supercomputing, is the use of supercomputers and parallel processing techniques to solve complex computational problems. HPC systems achieve computing performance in the petaflops range - equivalent to quadrillions of computational operations per second. This exceptional performance enables the solving of scientific and technical problems that would be unsolvable with conventional computers.

The definition of High Performance Computing encompasses three core components: First, the hardware infrastructure consisting of thousands of processors, accelerators, and specialized storage systems. Second, the software architecture that enables parallel processing and optimized algorithms. Third, the network infrastructure that ensures high-speed communication between all components.

As of November 2024, the El Capitan system at Lawrence Livermore National Laboratory leads the TOP500 list with a performance of 1.742 exaflops. This development marks the beginning of the exascale era, in which supercomputers can perform more than one quintillion computational operations per second. MEGWARE is represented with eight systems in the current TOP500 list, including the Helma system at the University of Erlangen ranking 51 worldwide.

The fundamental difference between HPC and conventional computing lies in parallel processing. While a desktop computer typically processes tasks sequentially and computes in isolation, HPC systems divide complex problems into thousands of smaller subtasks that are processed simultaneously via multiple servers. This approach enables, for example, weather forecasts to be calculated in hours instead of weeks, or molecular interactions in drug development in real-time.

Advantages and Features of HPC Systems

• Extremely high computing power enabled by simultaneous processing of thousands of tasks (parallel computing), e.g., through GPUs
• Cluster architecture: Many individual computers (nodes) work together in a network
• High speed and efficiency in processing large amounts of data
• Scalability: Computing power can be flexibly expanded

Typical Application Areas of High Performance Computing

• Science & Research: Climate simulation, astrophysics, particle physics
• Medicine & Biotech: Gene analysis, drug development
• Industry & Technology: Fluid mechanics, structural and material simulation
• Business & Finance: Risk analysis, Big Data, AI models
• Artificial Intelligence: Training of deep learning models

Why is HPC Important?

High Performance Computing enables breakthroughs in science, technology, and industry. Without HPC, many modern innovations - from precise weather forecasts to new vaccines - would be hardly feasible. HPC plays a particularly crucial role in Big Data and AI.

The functionality of High Performance Computing is based on the principle of massive parallel processing. An HPC cluster consists of hundreds to thousands of compute nodes connected via a high-speed network. Each node has multiple processors (CPUs) and increasingly also specialized accelerators such as GPUs or FPGAs.

The architecture of a modern HPC system follows a hierarchical structure. At the base are the compute nodes, equipped with current processors such as AMD EPYC 5th generation with up to 192 cores or Intel Xeon Scalable processors. These are supplemented by GPU accelerators such as NVIDIA B200 or AMD Instinct MI400, which are particularly optimized for AI and scientific calculations. The connection is made via high-performance interconnects such as InfiniBand, Omnipath, or UltraEthernet, which enable data transfer rates in the range of thousands of gigabits per second.

A critical aspect of modern HPC systems is thermal management. The enormous computing power generates considerable waste heat that must be efficiently dissipated. MEGWARE's EUREKA technology uses Direct Water Cooling (DLC) for this purpose, cooling all components of a server with temperatures of up to 50°C. This innovative solution not only enables year-round free cooling without energy-intensive cooling machines but also the use of waste heat for building heating or district heating networks.

The software side of an HPC system is just as complex as the hardware. Parallel programming models such as MPI (Message Passing Interface) and OpenMP enable applications to be distributed across thousands of processor cores. Modern HPC systems also use container technologies such as Apptainer for portable and reproducible software environments. Job schedulers like SLURM orchestrate resource distribution and ensure that all components are optimally utilized.

The storage architecture in HPC systems also follows a hierarchical approach. Parallel file systems such as Lustre or BeeGFS enable thousands of compute nodes to access data simultaneously. Modern systems also integrate burst buffers with NVMe SSDs, which serve as intermediate storage for particularly I/O-intensive applications. The Helma system at the University of Erlangen, for example, has a 5-petabyte all-flash storage system.

Well-known HPC Systems

• El Capitan (USA): Currently one of the fastest supercomputers in the world
• JUPITER (Germany): Europe's first exascale HPC system at Forschungszentrum Jülich
• Fugaku (Japan): For medical research and climate simulations

Since founding in 1990, MEGWARE has developed into one of Europe's leading HPC specialists. With numerous HPC projects and currently various systems in the TOP500 list, the company demonstrates continuous innovation strength. The Helma system at the University of Erlangen, at position 51 in the TOP500, and the Capella system at TU Dresden, position 6 in the Green500 list, show the combination of highest performance and energy efficiency.

As a company that develops and produces entirely in Germany, MEGWARE combines all competencies under one roof in Chemnitz. More than 50 highly qualified employees cover the entire spectrum from consulting, development to service. In-house production enables customer-specific configurations and guarantees the highest quality standards according to DIN EN ISO 9001:2008.

Partnerships with leading technology providers such as Intel, AMD, and NVIDIA ensure access to the latest technologies. MEGWARE was, for example, with the CooLMUC-3 at LRZ, the world's first company to initiate an HPC system operated 100% with direct warm water cooling. This pioneering achievement was praised by Dr. Herbert Huber from LRZ as a "pilot system for future system cooling concepts”.

MEGWARE's strength lies in the holistic support of HPC projects. From initial needs analysis, system architecture to commissioning and ongoing support, the company offers everything from a single source. Proprietary technologies such as ClustWare® for cluster management and EUREKA for flexible hot water-cooled server solutions complement the portfolio.

The European perspective is a central competitive advantage. While many providers are US-centric, MEGWARE understands the specific requirements of European customers - from GDPR compliance to EU energy efficiency directives, to local support requirements. Membership in the European Technology Platform for High Performance Computing (ETP4HPC) underscores the commitment to the European HPC landscape.

What is the difference between HPC and Cloud Computing? While cloud computing is based on virtualization and shared resources, HPC uses dedicated hardware with high-speed connections for maximum performance. HPC systems are optimized for intensive, parallel workloads, while cloud systems offer flexibility and scalability for diverse applications. Modern approaches combine both worlds in the hybrid HPC concept.

How much does an HPC system cost? Costs vary greatly depending on requirements. The Total Cost of Ownership (TCO) is crucial, which includes acquisition, but also energy, cooling, and maintenance. MEGWARE's energy-efficient solutions significantly reduce operating costs.

Which programming languages are used for HPC? The dominant languages in HPC are C, C++, and Fortran, supplemented by parallel programming models such as MPI and OpenMP. Python with scientific libraries and CUDA for GPU programming are also increasingly gaining importance. The choice depends on the specific application and available software infrastructure.

How is performance of HPC systems measured? The primary metric is FLOPS (Floating Point Operations Per Second), measured by the LINPACK benchmark for the TOP500 list. Other important metrics are energy efficiency (FLOPS/Watt for Green500), memory bandwidth, network latency, and application-specific benchmarks. Real-world performance often differs from theoretical peak values.

What role does Artificial Intelligence play in HPC? AI and HPC are increasingly merging. HPC systems train large AI models, while AI methods optimize HPC applications. Examples include AI-supported weather forecasts, molecular simulations with machine learning, and intelligent resource allocation. Modern HPC architectures integrate specialized AI accelerators for optimal performance.

How sustainable is High Performance Computing? Modern HPC systems focus on energy efficiency with innovative cooling concepts, optimized hardware, and intelligent load distribution. The MEGWARE platform EUREKA reduces energy consumption for cooling. The use of waste heat and the deployment of renewable energies make HPC increasingly sustainable. The Green500 list honors the world's most energy-efficient supercomputers.