How to Grow your HPC Capability

Hardware and software are not ‘one size fits all’. But there’s a simple way to deploy a tested and verified HPC solution that’s optimised for real-world performance.

CIOs have long had a choice between customised and off-the-peg software. Most organisations work using a mixture of the two, keen to tap into the accessibility and cost-value of off-the-shelf software but aware that bespoke code can deliver advantages that a more value-orientated solution cannot.

However, there are some areas where customised software dominates. In the HPC field, for example, the specific needs of researchers will often entail the use of highly customised software: it’s the nature of the beast.

HPC workloads are often designed for specific research projects and, as such, are highly customised. They need to be so. To be effective, they need to operate speedily, as time is of great essence in many HPC applications. Users are looking for time-critical information.

One of the key elements in this process is parallelism, the processing of many different tasks simultaneously. This involves the use of a technology called Message Passing Interface (MPI), a means by which data can be exchanged between multiple computers. That means that applications need to take particular care in deploying MPI, ensuring that data processing is as seamless as possible.

While there are many advantages of going with a customised approach to such applications, Intel has introduced a series of performance-optimised solutions that can aid companies venturing down the HPC route. These Select Solutions currently provide pre-validated and tested hardware/software stacks in key areas such as professional visualisation, genomics analytics, plus simulation and modelling.

Discover how Intel® technologies can unlock the potential of your business data ›

“Working with Intel on Select Solutions, we’re able to go to the customer and be confident in our conversation with them,” says Angelo Apa, Technical Sales and Business Development Director for the Lenovo Data Center Group in the UK. “We can provide the solution that they’re looking for, in the environment that they’re looking for it, with [the] confidence that it’s going to work in exactly the way that the customer is looking for it to do so.”

Putting together an HPC cluster from scratch, complete with modelling and analytical software, can be very time-consuming.

Intel® Select Solutions for Simulation and Modelling, for example, provide a means by which research scientists can construct complex simulation/engineering models using the latest HPC techniques. This ultimately helps to accelerate workflow, especially in areas such as computer-aided engineering (CAE), computational fluid dynamics (CFD) and weather forecasting.

A minimum ‘Base’ configuration in these scenarios might include two Intel® Xeon® Gold 6126 processors (12 cores/24 threads) running at 2.60GHz, 96GB of memory and a low-latency drive like the Intel® SSD DC S3520 Series, which bridges the gap between warm and hot tier storage. A ‘Plus’ Select Solutions configuration, one optimised for even higher performance, might include a pair of Intel® Xeon® Gold 6148 processors (20 cores/40 threads) or an Intel® Xeon® Platinum processor, with up to 28 cores/56 threads.

Computational fluid dynamics (CFD) is instrumental in aviation design, helping to improve the efficiency of today’s passenger aircraft.

Another growth area as far as computer modelling is concerned is in the field of genomics research. Scientific organisations are dealing with vast volumes of genomic sequencing data, many producing more than 20TB a day. In such environments, it’s essential to have an HPC cluster that can process high volumes of data accurately. Intel® Select Solutions for Genomics Analytics offers a fast-to-deploy solution for analysing complex genome pipelines, ensuring that researchers can operate at scale.

In all such cases, putting together a powerful HPC cluster from scratch, complete with modelling and analytical software, can be very time-consuming. It means recruiting and managing highly trained and specialist staff, for one thing – such people are not easy to find. And the cost of finding and retaining them could be prohibitive for smaller organisations, while the larger research bodies may find the cost of developing and maintaining such software does not provide an effective return on investment.

And then there’s the nature of the HPC clusters themselves. Putting these together effectively needs to ensure that multiple components are working in harmony with each other. There’s very little margin for error in such deployments: information is wanted quickly and accurately, no mean task when dealing with such large datasets.

HPC can be used for prototyping, visualising complex data in areas such as medicine and architecture.

Intel and its OEM partners also provide customised Select Solutions for Professional Visualisation - turning complex data into photorealistic 3D representations. The combination of high performance computing power and Intel’s software-defined visualisation libraries allows scientists to visualise geological data, engineers to interact with new products and digital artists to render future buildings.

Advantech, for example, partners with Intel to verify its HPC-8212 and ASMB-925 server board, as well as its SKY-524 server. “As data size increases significantly,” the company explains, “the limitations of graphics-processing hardware makes visualising data at large scale a significant challenge. While interactive visual computing of large 3D data is becoming critical in many industries, reducing system costs also becomes a significant factor when making decisions.”

What Intel offers with Select Solutions is confidence. It’s the confidence that even the most complex HPC applications can be handled seamlessly, that complex modelling can achieved with little pain, and that researchers and engineers can ensure that MPI libraries are accessed effectively.

While there will always be space for customised solutions, in such high-cost environments such as scientific modelling and professional visualisation, Intel® Select Solutions can provide a scalable, power-efficient and cost-conscious solution for rendering 3D data. This is especially important at a time GPU-based rendering often lacks a compelling balance between performance, visual fidelity, and cost.

Unleash the power of your business data with leading-edge Intel® technologies ›

Crucially, the Select Solutions approach can be used to speed up the deployment of HPC resources and, by providing effective analytics in a timely manner, they can ensure that customers can maximise ROI in what can be a highly complex area.