June 12, 2019 - Critical Facilities
By Jake Ring
2019 is the year high-performance computing (HPC) becomes cost-effective for small and mid-sized organizations. If this statement was uttered just five years ago the very notion would have sounded preposterous. Today, we are embarking upon HPC services that can be attained by almost any organization, but the path to achieving it has been long and winding.
Each generation of computers has seen a category classified as “high performance.” Often referred to as supercomputers, they were huge and very costly to operate, which made their use only practical for government and some industrial research projects. It wasn't until the late 1980s, when the x86 server breakthrough came onto the scene, that we began to see HPC clusters more commonly emerge. Soon after, we had PCs and servers connected to local area networks (LANs) and strung across wide area networks (WANs) encircling the globe.
Almost every technological innovation has begun with a cost barrier for adoption, and HPC is no exception to the rule. However, competition and new innovations such as higher performance, lower cost chipsets have a way of lowering the adoption bar so many other organizations can clear the hurdle and take advantage of technology innovations.
According to Grand View Research, the global high-performance computing market size was valued at $34.62 billion in 2017, and it is expected to expand at a compound annual growth rate of 7.2 percent from 2018 to 2025. Most of this growth is attributed to diversification, advances in virtualization, and an increase in hybrid HPC solutions.
In addition, much HPC technology has been commoditized so it’s now within most budgets. In fact, Grand View Research has also placed the North American market at more than 30 percent of the HPC share in 2017. This strong adoption is propelled by organizations such as the OpenHPC Community, which has created new standards on how HPC software can effectively be developed. According to the OpenHPC Community website, their packages are pre-built with HPC integration in mind and include reusable building blocks. Obviously, anything that simplifies HPC use will dramatically impact the adoption and stimulate mainstream use.
Until now, the cost was a barrier too high for most small to mid-size organizations to clear in order to take advantage of HPC. Additionally, the adoption of HPC was hampered by the designs of existing data center infrastructure that could not easily accommodate the higher power distribution and cooling requirements. However, in 2019, data centers and colocation facilities now have access to new power and rack-density innovations that are finally lowering the price points for almost any company to quickly take advantage of cutting-edge computing resources.
In the same spirit that the OpenHPC Community is easing adoption rates, so too are the creators of a new generation of modular data center enclosures that use innovative modular cooling systems. These new systems provide hot and cold aisle isolation and adiabatic cooling to support HPC loads from 25kW to 40kW, while also delivering a high efficiency power use effectiveness (PUE) of 1.15 or better. The power innovations allow data center colocation providers to actually lower their price points to within the range of small to medium-size businesses that have been traditionally sitting on the sidelines because operating expenses were out of their range. The lower price points also help create a better return on HPC investments. IDC research has previously noted that investments in HPC already can be associated with substantial returns such as:
The promise of HPC is to unlock potential new applications or use existing ones more efficiently. The computational simulations are not merely relevant to artificial intelligence, machine learning, and research labs, they are also directly applicable for oil and gas companies, as well as financial services and media and entertainment use — if major adoption hurdles are cleared. Ever since the advent of the first electronic general purpose computer, the ENIAC, there have been 5 major hurdles to clear for mainstream technology adoption:
With these hurdles cleared, market adoption comes quickly. Market Research Future predicts that the global high-performance computing market is expected to reach a value of $50 billion by 2023, growing with an 8 percent compound annual growth rate.
When it comes to seeking new data center/colocation space or expanding existing on-prem services for HPC needs, two additional bullets need to be added to the list of technology adoption hurdles, namely power density and efficiency. In 2019, even these hurdles are being lowered so that the masses can take advantage of HPC.
As a case in point, consider the U.S. Department of Energy's (DoE) National Energy Technology Laboratory in Morgantown, W.Va. An innovative design was created to support 1 MW of power for the supercomputer used for simulations on the impact of various fossil fuels. The innovation resides in the form of a modular enclosure built directly onto a concrete pad — occupying only 40 feet by 60 feet of available space. The need for a chiller plant was removed by using adiabatic cooling in the enclosed cold aisles and creating a single, centralized enclosed hot aisle for better heat removal. Now, during the cooler months, the cold outside air is brought in through the HEPA filters for free cooling, and during the warmer months, the air is brought into the adiabatic cooling mixing chamber to achieve a range of 74 degrees to 83 degrees F. Using this design, DOE achieved an acceptable power density to support its HPC equipment racks, over 25kW per cabinet, while achieving greater efficiency measured to an average PUE of only 1.06 annually. These same innovative design principles are now being incorporated in civilian data center construction and are used with HPC workloads at very low price points.
HPC innovations — once only in reach of government facilities or large enterprise organizations — are now attainable by mainstream businesses. Small and midsize organizations do not have to spend their budgets building on-premises HPC clusters and can now outsource their need to colocation data centers built to handle the workloads. Today’s providers help companies of all sizes meet their HPC needs with purpose-built solutions, tested and validated by designs such as the DOE-NETL's West Virginia facility, to clear all seven technology adoption hurdles. Leveraging these types of outsourced services, organizations will minimize their operational costs while achieving enhanced efficiency through cost-effective and fully-managed HPC computing services.
Jake Ring is president and chief executive officer of GIGA Data Centers. Prior to GIGA, he founded DC BLOX as CEO in 2014 to provide comprehensive data center services for the SMB market including colocation and managed services.
