About the authors:
Adrian Wander is CEO and Director of A Wander Consulting, an IT consultancy specializing in HPC procurement and tenders from both the vendor and customer side. Adrian has over 30 years’ experience of working in the high-performance computing and IT sectors and is a well-known and respected figure within the community with an enviable network of contacts.
Rob Elder is Vice President of Sales, Bulk Data Centers, one of the Nordics’ leading data center providers delivering ultra-flexible, scalable, highly connected and sustainable solutions. Rob brings over 15 years experience in the data center sector where he has worked in Senior Management and as a Board Director with a wide range of experience from strategic planning, sales and marketing, technical projects and operational management.
High performance computing (HPC) uses parallel processing to run advanced, specialized applications typically designed to perform modelling and analysis on massive data sets. More specifically, we use the term to refer to systems that function above a teraflop or 1012 floating-point operations per second. HPC differs slightly from supercomputing, where a given system performs at or near the highest achievable rate for computers, more than a petaflop or 1015 floating-point operations per second. By contrast, a typical PC with an i7 processor functions around 350 gigaflops or .35 teraflops.
Typically, HPC systems are used by scientific researchers, engineers and academic institutions, as well as for government and defence applications. These systems often use custom-made components alongside commodity components. Recent studies, including Market Estimates & Trend Analysis to 2027 by Grandview Research, put the USD $39 billion HPC market at an annual growth rate of six to eight percent. At that rate, the global HPC market could reach USD $60 billion by 2030.
To add a bit of context, at $39 billion, HPC represents about 23 percent of the $170 billion global data center market. The International Energy Agency (IEA) estimates data centers account for 1 percent of global electricity demand at nearly 205 TWh. The fastest computer in the world, Supercomputer Fugaku at Riken in Japan is consuming over 28MW of power, roughly equivalent to 68,000 EU homes.
HPC is a major driver in scientific discovery and technological innovation. Major users include the world’s largest and most respected research laboratories including the Science and Technology Facilities Council here in the UK. Industry is increasingly adopting HPC systems to drive business value. Aerospace is doing some amazing work in computational fluid dynamics. Oil and gas is another big user - working out where to drill is expensive if you get it wrong. The automotive industry is improving crash testing and aerodynamics to improve fuel efficiency. There are major increases in life sciences using artificial intelligence (AI), including a major surge in 2020 to focus on COVID-19 work. This last is a stark example of how critical it can be to be able to rapidly scale up, control costs and ensure availability. Finance, weather and climate science, and defence and intelligence applications are other major users.
In Adrian’s view, the value created through HPC in R&D, fundamental science, safety, sustainability and prosperity is more than sufficient to drive continued growth. Based on decades of HPC consulting, including advising on tenders from both sides of the table, a big challenge is that most contracts still don’t account for total cost savings properly.
High Value versus High Return
Still, the return on investment (ROI) from HPC remains a topic of some debate among experts. A study by Hyperion Research finds that high performance computing generates $44 in profit for every dollar of investment in HPC systems. Hyperion says that while ROI is high across the entire HPC ecosystem, certain sectors get even more bang for their buck.
At the same time, Intersect360’s Solve Report takes a wider view, looking at stock prices versus the Dow Jones Industrial Average (DJIA) for companies with increasing, flat and decreasing HPC investment budgets, finding little difference. They find little to no improved financial performance among those companies increasing spending. Their view is that HPC is critical to those who use it. For many of these companies, HPC is part of the cost of doing business, not a driver of growth or success.
This debate begs the question: Am I getting the most out of my HPC investment? Put another way, if I’m going to invest in HPC, how can I maximize the value of that investment? Rob Elder points to the Nordics as a viable and valuable option.
For example, Bulk’s Nordic data centers allow customers to operate at the lowest total cost of ownership at the same time as being directly powered by 100 percent renewable energy. Norway offers robust connectivity to and through Europe and North America. Bulk connects the Nordics with the world’s major markets via low-latency and high-capacity fiber networks. This unique blend of performance, reliability, affordability and sustainability is practically custom made to suit HPC needs, especially where solving climate and sustainability challenges are part of the equation.
Emissions and Cost: Doing More for Less
In Adrian’s years of consulting in the field, the HPC growth hypothesis begins to look something like this. More and more disciplines and industries are getting into HPC, with more and more applications addressing public good issues like health, safety and climate issues. As these applications grow and expand, they need more and more power to meet higher and higher rack densities. Alignment with their own mission, in many cases, means new HPC investments should also be done sustainably.
Put another way, significant sectors of the growing HPC market should look to move these workloads to locations with lower cost, cleaner energy sources. HPC tenders too often ignore the wider data center infrastructure. They should be asking themselves, have I got enough power? Have I got enough cooling? Far too often, the answer there is No.
To address this, Adrian sees great opportunity and alignment in the Nordics. There are typically zero days above 24 degrees Celsius in the Nordics. This means that free air cooling, rather than chillers, can be utilised throughout the year reducing the PUE of the data center and hence the total cost of ownership. The cooler climate also reduces the need for water which is an important factor given issues on water scarcity in parts of Europe.
For many HPC users, they are solving big problems that revolve around soft issues, including sustainability. Building or contracting HPC data center space in a “dirty” grid market runs counter to the good work they are doing. There are big and bold challenges coming from industry on climate and sustainability. Earlier this year, Unilever set the audacious goal of doubling their business with their Sustainable Living Plan which was updated recently to include reducing its environmental impact by half by 2030.
A more common approach for organizations is seeking opportunities to offset carbon emissions with green investments elsewhere. The problem there is that for every watt of green energy you buy you are still consuming a dirty watt and producing carbon.
On-Prem, Cloud or Colocation?
For HPC, the on-premise segment held the largest revenue share in 2019 at nearly 60 percent. This makes sense because government and defense applications hold the majority of market share. Governments want to protect national security and citizen data, and enterprises want to guard proprietary commercially sensitive data. Still, we are seeing significant growth in other sectors including transportation, manufacturing and financial services.
Cloud HPC is expected to expand at the highest CAGR of 7.6% in the coming years. But for many, cloud HPC can be price prohibitive and is bound by network bandwidth performance. If the workload is peaky, cloud can make sense for those applications.
Rob points out that, for research and other applications where the workload is more steady, colocation will work well for the foreseeable future. It will take some time for cloud pricing to come down - and flexibility and performance to come up - to meet a significant share of emerging HPC needs. There is more control over data in colocation environments versus the cloud, and much more flexibility over technology choice, performance and how to configure and manage systems.
But for enterprise HPC needs, low-cost, highly sustainable colocation at Bulk’s data centers in the Nordics offers many advantages. And, Rob points out, Bulk has developed an extensive set of relationships and skill sets that dramatically lower the bar and shorten the learning curve for organizations doing business in the Nordics for the first time. This makes navigating the legal and regulatory requirements for setting up data center operations easier than ever before.
So, where data does not need to be on premise and where we are taking HPC applications off-shore, colocation in the Nordics offers many significant advantages as a lower cost strategy.
To Maximize HPC ROI, Focus on TCO
What is the true total cost of ownership? Breaking down the cost structure in a colocation environment, we have equipment, rent or floor space and the associated opportunity cost, power and cooling. When you look at these factors, this makes the emerging Nordics data center market an intriguing option. It is difficult in HPC tenders to calculate the true cost savings. Adrian believes most don’t do it properly.
Fundamentally, for those where colocation is an option, the decision comes down to the question, Do I have faith in the ability of the provider? For example, do they have operating experience with high density and water cooling? And it doesn’t stop there.
Too often, those scoping jobs in data center design “assume” the electrical and mechanical infrastructure is there and already suited to purpose. There is often a disconnect between the data center infrastructure and the terms in the Service Level Agreement and performance needed. This can lead to issues in linking the technical solution to the procurement process.
When choosing a partner, is there sufficient data to support nameplate capacity, availability and redundancy? My advice is to list out the single points of failure. What are the maintenance requirements? How do they handle failovers? What does the testing program look like?
Take a Hard Look at HPC Colocation Solutions
Buying cheap is not the same as making a low-cost investment. The Nordics are growing because they offer high levels of sustainable power at low cost. The workforce is highly skilled and experienced and offer trusted partner relationships to ensure high levels of security and reliability.
This, Rob Elder says, is precisely where Bulk Data Centers shines. The Nordics offer resilient, low-cost, renewable energy to serve growing HPC needs. We help customers operate sustainably and minimize costs, yes. Bulk goes further to ensure success. We work as both a trusted advisor as well as a service management provider to offer flexibility to grow to massive scale, where needed. It’s really quite incredible what we have been able to do with the support of our own growing and trusted network of partners and colleagues.
Colocation is addressing many of the challenges faced by the rapidly growing high performance computing industry. Going to the Nordics solves the problems and adds benefits in key areas of sustainability and total cost of ownership.