Is Iceland the Future of Cheaper, Greener Supercomputing?
Complete the form below to unlock access to ALL audio articles.
High-performance computing (HPC or, to use another name, supercomputing) is a pretty darn useful technique. With the data deluge that is sweeping much of modern research, having a big bruiser of a computer nearby to crunch every zero and one within sight can make or break big research projects.
However, access to HPC is also pretty darn pricey - the Cray Titan supercomputer at Oak Ridge National Laboratory, a 17.59 petaflop beast, cost $60 million when it was first unveiled in 2013. It's also a fundamentally energy-guzzling business, which poses a problem for environmentally conscious researchers. Cheaper, greener HPC solutions are therefore a target for many research institutions who want the power without the price or pollution.
Verne Global, an HPC-specialized data center supplier, believe they might have found the answer in their 44-acre campus in Keflavik, Iceland. We talked to Spencer Lamb, Verne Global’s Director of Research, to find out more.
Ruairi Mackenzie (RM): How much do research organizations spend on HPC?
Spencer Lamb (SL): To answer this question, we have two sources of information. First, we have one of the most innovative research organizations in the UK as a customer (Earlham Institute) and from their deployment at Verne Global, how it’s grown, and what they’ve told us about their plans, the use is only going to increase. Secondly, we recently placed a Freedom of Information Act request to the leading publicly grant-funded research institutes in the UK to understand how much power, and the cost of that power, is currently being used. We have this information for the past three years.The data from both these sources paints a picture of the increasing use of intense, power hungry compute in their work, and we know HPC is the major portion of institute’s computing portfolios.
HPC has become the vehicle to help drive research further, as scientists rely on the high-capacity supercomputers to help tackle some of the most intricate and nuanced research in the world. Two recent examples at both ends of the investment spectrum include firstly the University of Leicester – a medium sized university-based research facility – who confirmed an investment of £1M in 2017 in new HPC hardware to aid their research activities. Secondly, at the other end of the spectrum, Edinburgh University’s Parallel Computing Centre (EPCC) hosts the UK’s national supercomputer – ARCHER, which received an investment in 2017 of £20M in six, new Tier-2 HPC centers in 2017.
RM: What areas of research are most affected by HPC costs?
SL: HPC is being used for a wide variety of research including genome analysis, climatic and environmental study, bioinformatics, astrophysics and quantum mechanics, and engineering research into new materials, to name just a few areas. HPC is also becoming more widely used to support the growth in AI, machine learning and deep neural network training. Fundamentally its becoming more mainstream than ever before and more and more Researchers and trying to access these systems for their projects. This is not only in the field of publicly funded research but across industry who are approaching Research Institutions for HPC/Supercomputing expertise.
RM: Where are these research organizations currently getting their HPC power from? Do they use in-house power centers?
SL: The vast majority of energy is provided by the UK National Grid. Power in the UK is predominantly fueled by fossil fuel (almost 50% being gas) and energy costs are among the highest in Europe at approximately £0.17 kWh. In addition, due to the low margin between demand and supply (approx. 6.2%-8.2%) the provision and price of power in the UK is a major risk factor for energy-intensive computing such as HPC. Successive UK Governments have failed over the last 20 years to introduce a cost-effective renewable energy supply plan and as a result the country’s electricity is still reliant upon traditional fossil fuel generation along with ever increasing costs.
RM: What exactly enables Nordic centers like Verne Global to quadruple the efficiency of UK centers?
SL: In contrast to the UK, Iceland provides a perfect, optimal environment for intensive computing infrastructures such as HPC. Firstly, Iceland’s power profile is massively abundant and scalable, offering a margin between demand and supply of over 90%. As a result, power is extremely low cost (approx. £0.04p kWh) and also long-term, with 12-15 year pricing contracts provided by the national power company, Landsvirkjun. This provides research organizations with savings on power of 75% compared to the UK, and unrivalled pricing predictability.
Secondly, within a data center around 30-40% of the cost of hosting infrastructure is cooling the hardware. This is especially the case with massively compute-intensive supercomputers such as HPC. In Iceland however, due to the countries cool, temperate climate, free-cooling is possible via the use of ambient air. This significantly reduces the cost of operating HPC infrastructure in Iceland as compared to the UK.
Lastly, because the power profile in Iceland is dual-sourced and 100% renewable (geothermal and hydro-electric), HPC can be operated at practically carbon-neutral levels. Considering there are institutes in the UK that use HPC for the study of climate change and sustainability, the environmental impact of their compute footprint is important, and Iceland provides an advantageous location for this.
As an example of the overall contrast between the UK and Iceland by moving their HPC workloads to data centers like Verne Global, institutions could cut the power costs associated with their HPC and intensive computing by approximately 75% in most cases. These savings can be translated into these institutes being able to power 429,796 HPC cores in Iceland vs. 118, 080 HPC cores in the UK, resulting in almost 4x more HPC research output for the same power cost.
Spencer Lamb was talking to Ruairi J Mackenzie, Science Writer for Technology Networks.