A Primer on High Performance Computing (HPC) Systems
Although HPC systems traditionally come with a high CapEx, clouds now offer a cheaper initial on-ramp – and their ability to handle demanding AI/AL workloads makes them an attractive option.
January 8, 2021
![abstract blue tech hpc systems abstract blue tech hpc systems](https://eu-images.contentstack.com/v3/assets/blt8eb3cdfc1fce5194/bltb657245842e0e2c4/6621157d947fddf8129c62a4/abstract-blue-tech.jpg?width=1280&auto=webp&quality=95&format=jpg&disable=upscale)
Tech has no shortage of buzzy new technologies – and cutting through the hype to see what will actually impact the enterprise can be challenging. We're here to help. Starting in 2021, our contributors will give a rundown on an emerging tech and whether it'll pay off to pay attention to it. Here, we look at high performance computing (HPC) systems.
To see the other trends highlighted in our IT Trends To Watch series, read our Emerging IT Trends To Watch report.
What Is HPC?
Due to the huge uptake of artificial intelligence and machine learning, one of the trends to watch in 2021 will be HPC systems.
Defining all of the elements of HPC is complicated, but to keep it short and simple, it's a computer system designed to handle more data more quickly, using standard out-of-the-box servers and storage devices.
"Your standard HPC platform is going to be some number of servers with a high performance network interconnect [and] with dozens to hundreds of servers hooked together," said John Leidel, chief scientist and founder at Tactical Computing Laboratories. "There's a significant amount of power and infrastructure that goes into just doing that in terms of your network infrastructure, and it's very hard to amortize that in your application performance."
HPC solutions are expensive, both in terms of purchasing the equipment and in terms of operating costs. The servers used in these systems are not off-the-shelf servers that might be used to serve websites, but typically use multiple CPUs with many cores, as well as a high number of GPUs and other silicon used as accelerator chips – all tied together using infrastructure designs that try to amplify resource allocation.
Just for an example, Leidel talked about a new design he's working on now.
"One of the problems we have in HPC is we build these giant networks to hook all these machines together and the networks use a tremendous amount of power," he said. "What we're doing is we're actually attaching RISC-V CPUs directly to the network devices in the network, so instead of just sending data back and forth you can send the data and do some compute on it while it's in flight – so instead receiving data at the other end, you actually receive an answer."
How Long Has It Been Around?
Vendors have long been offering increased performance computers for specific workloads – since at least the beginning of the microprocessor age.
Traditionally, some of the biggest users of HPC have been medical research, where HPC systems are now being harnessed in the fight against COVID-19, the oil industry and the financial sector.
Pankaj Goyal, VP of product management at HPE, told us that in financial institutions, "Monte Carlo simulations are frequently used to predict markets, and those use HPC techniques."
He added that the oil industry has been a big user of HPC for exploration. "Basically, they get imagery data from their wells and they use different simulation techniques to predict how much oil, at what quality level and at what depth."
Why Are People Paying Attention to It Now?
Interest in HPC systems first began to grow outside traditional markets around 2006 as Hadoop, big data and data mining came to the forefront. As the use of AI and ML for data analysis rose across enterprises, so did interest in more powerful computing.
Another driving force is that public clouds are now offering HPC and HPC-like services, which give companies the opportunity to try HPC workloads before investing in their own on-premises HPC systems.
"I think because those technologies are available in a pay as you go kind of engagement, that lowers the barrier for people getting started in HPC," Chris Porter, project manager for converged HPC and AI for IBM Cognitive Systems, said.
"I can try to run a sample of my SAP environment in the cloud on some high performance hardware, using a high performance parallel file system, perhaps using a high performance network, to see if that really has benefits for me," he said. "I can do that for three months, six months, or whatever it would take for me to gather the data to answer my question."
Who Benefits From It?
Basically, whether or not an enterprise needs HPC systems is determined by the software it plans to deploy and at what scale. Again, the current HPC revolution is being driven primarily by data intensive AI/ML workloads, whether in an on-premises data center, in the cloud or at an edge location.
"I would say that many HPC techniques and AI techniques are merging together," HPE's Goyal said. "For example, we are seeing demand with our retailers who want to use AI in their stores to improve the customer experience. The use case might be video and text in their own store, or it might be predictive modeling of customer behavior based on their history. Those techniques are typical AI techniques, like typical AI training techniques, but they use a lot of underlying infrastructure which are common to HPC."
Where Can You Get It?
All of the major computer vendors like HPE, Dell and Lenovo, market HPC equipment and most public clouds offer HPC capabilities.
Although out-of-the-box HPC solutions are available, most enterprises will be better served to work with a vendor to build a system that's customized to meet their needs.
About the Author
You May Also Like