High Per­form­ance Computing (HPC)

High per­form­ance computing is used to handle large volumes of data and complex computing tasks in parallel. Typical areas of ap­plic­a­tion include economics, science, sim­u­la­tions, and business in­tel­li­gence. But which HPC methods are there and how do they work?

High Per­form­ance Computing, or HPC for short, is not so much a clearly defined tech­no­logy, but rather a set of pro­ced­ures that use or make available the per­form­ance and memory capacity of ordinary computers. There are no fixed criteria for HPC, as HPC changes with the times and adapts to new computing tech­no­lo­gies. In general, it can be said that HPC solutions are used for complex computing op­er­a­tions with very large amounts of data or for the analysis, cal­cu­la­tion, and sim­u­la­tion of systems and models.

On the one hand, HPC processes can be used on in­di­vidu­al, very powerful computers. More often, however, HPC is found in the form of HPC nodes in su­per­com­puters, known as HPC clusters. Su­per­com­puters are capable of parallel, high-per­form­ance computing with multiple ag­greg­ated resources. Early HPC su­per­com­puters were developed by current Intel partner Cray. Today, su­per­com­puters are much more powerful, since complex hardware and software ar­chi­tec­tures are linked via nodes and per­form­ance cap­ab­il­it­ies are combined.

When data volumes overwhelm the per­form­ance of con­ven­tion­al computers, HPC en­vir­on­ments are called for. As a form of dis­trib­uted computing, HPC uses the ag­greg­ated per­form­ance of coupled computers within a system or the ag­greg­ated per­form­ance of hardware and software en­vir­on­ments and servers. Modern HPC clusters and ar­chi­tec­tures for high-per­form­ance computing are composed of CPUs, work and data memories, ac­cel­er­at­ors, and HPC fabrics. Ap­plic­a­tions, meas­ure­ments, cal­cu­la­tions, and sim­u­la­tions on a large scale can be dis­trib­uted to parallel processes thanks to HPC. The task is dis­trib­uted via special computing software.

Two main ap­proaches are found in High Per­form­ance Computing ap­plic­a­tions:

1. Scale up: HPC tech­no­lo­gies use a complex ar­chi­tec­ture of hardware and software to dis­trib­ute tasks across available resources. The dis­tri­bu­tion to parallel computing processes takes place within a system or server. When scaling up, the per­form­ance potential is high, but is partial to the system’s lim­it­a­tions.
2. Scale out: In scale-out ar­chi­tec­tures, in­di­vidu­al computers, server systems, and storage ca­pa­cit­ies are connected to form nodes and HPC clusters using clus­ter­ing.

In theory, a system’s in­di­vidu­al coupled computers can suffice for scale-up HPC re­quire­ments. In practice, however, the scale-up approach hardly proves efficient for very large ap­plic­a­tions. Only the com­bin­a­tion of computing units and server systems ac­cu­mu­lates required ca­pa­cit­ies and scales the required per­form­ance as needed. The com­pil­a­tion, dis­tri­bu­tion, or sep­ar­a­tion of HPC clusters is usually done via a single server system with merged computing units or via a HPC provider’s automated cloud computing.

In contrast to local or supra-regional stan­dalone systems that run HPC ap­plic­a­tions via a server, HPC via cloud computing offers sig­ni­fic­antly more capacity and scalab­il­ity. HPC providers provide an IT en­vir­on­ment con­sist­ing of servers and computer systems that can be booked on demand. Access is flexible and fast. In addition, the cloud services offered by HPC providers are almost un­lim­itedly scalable and guarantee a reliable cloud in­fra­struc­ture for HPC processes. The on-premises model with in­di­vidu­al systems, con­sist­ing of one or more servers and complex IT in­fra­struc­ture, offers more in­de­pend­ence but is dependent on higher in­vest­ments and upgrades.

Just like the fluid defin­i­tion of HPC, the ap­plic­a­tion of HPC can be found almost every­where where complex computing processes take place. HPC can be used locally on-premises, via the cloud, or even as a hybrid model. In­dus­tries that rely on or regularly use HPC include:

• Genomics: For DNA se­quen­cing, lineage studies, and drug analysis
• Medicine: Drug research, vaccine pro­duc­tion, therapy research
• In­dus­tri­al sector: Sim­u­la­tions and models, e.g. ar­ti­fi­cial in­tel­li­gence, machine learning, autonom­ous driving, or process op­tim­isa­tion
• Aerospace: Sim­u­la­tions on aero­dy­nam­ics
• Finance: In the context of financial tech­no­logy to perform risk analysis, fraud detection, business analysis, or financial modelling
• En­ter­tain­ment: Special effects, animation, transfer of files
• Met­eor­o­logy and cli­ma­to­logy: Weather fore­cast­ing, climate models, disaster forecasts, and warnings
• Particle physics: Cal­cu­la­tions and sim­u­la­tions of quantum mechanics/physics
• Quantum chemistry: Quantum chemical cal­cu­la­tions

HPC has long been more than a reliable tool for solving complex tasks and problems in the sciences. Today, companies and in­sti­tu­tions from a wide variety of fields also rely on powerful HPC tech­no­logy.

The ad­vant­ages of HPC include:

• Cost savings: HPC from the cloud allows larger and complex workloads to be processed even by smaller companies. Booking HPC services via HPC providers ensures trans­par­ent cost control.
• Greater per­form­ance, faster: Complex and time-consuming tasks can be completed faster with more computing capacity thanks to HPC ar­chi­tec­tures con­sist­ing of CPUs, server systems, and tech­no­lo­gies such as Remote Direct Memory Access.
• Process op­tim­isa­tion: Models and sim­u­la­tions can be used to make physical tests and trial phases more efficient, prevent failures and defects, for example in the in­dus­tri­al sector or in financial tech­no­logy, and optimise process flows through in­tel­li­gent auto­ma­tion.
• Knowledge gain: In research, HPC enables the eval­u­ation of enormous amounts of data and promotes in­nov­a­tion, fore­cast­ing, and knowledge.