Supercomputers are used for a variety of compute-intensive purposes; from the Met Office's weather predictions to the United States government's management of its vast stockpile of nuclear weapons. However the cost of using such high powered systems has typically been a barrier to wider use.
According to US firm Cray - which builds some of the largest supercomputers in the world - providing high performance computing infrastructure via the cloud will help to broaden access in future. This means offering services to organisations that may not have the budget, or skills, to run complex hardware at large scale.
[See also: The 11 fastest supercomputers on the planet]
It is pushing its ‘supercomputing as a service’ concept as a means to achieve this, partnering with hosting firm Markley. The first product of the partnership is the Cray Urika-GX, a system that combines supercomputer architecture elements with pre-configured data analytics tools. There are now plans in place to rapidly add Cray's full range of infrastructure solutions.
“The idea here is to lower the barriers to get access to such a technology and the capabilities that that technology offers you," says Dominik Ulmer, Cray's vice president of EMEA business operations.
"By offering the Urika-GX in a cloud environment, you reduce the time to insight even further. because you don't have to go through a procurement process in our own institution anymore.
"You don't have to book it via your IT department to set up a Urika-GX in your data centre and roll it out there. You can directly go to Markley, and basically lease a system.
"Your finance director or CFO may like that as well because it's just an opex now, it's not a capex anymore. So from a practical point of view, it's also easier to get access to internal resources to use such a system."
Cray is targeting pharmaceutical and genomics companies as initial clients, An MIT and Harvard biomedical science research centre called the Broad Institute used the system to link cancer patient data with genomics and proteomics. This allowed them to detect relationships between an individual genomic setup and the mortality rates of cancer, to better understand the ideal method of treatment.
However, the company expects applications to expand across cyber security, oil and gas, finance and numerous other sectors.
"Industry is discovering supercomputing as a competitive advantage, and they're starting to explore this more and more,” says Ulmer. "The fundamental drive behind all this is that they can run bigger models in faster time to solution, which means that they can have more complex problems and bring their [products] faster to market. But they all see now, that the industry standard system designs very quickly reach the limits of capability."
Changing supercomputing market
The supercomputing market has declined dramatically in recent months, and Cray's financial results have suffered as a result. The company's revenues fell from $105.5 million (£82.3 million) in the first quarter of 2016 to $59 million (£46.3 million) in the first quarter of 2017, with net losses hitting $19.2 million (£15 million).
Ulmer attributes the drop primarily to dwindling budgets in the public sector, traditionally the primary source of its business. Supercomputing as a service can open up the company to new customers.
High performance computing (HPC) is already offered by cloud providers such as AWS and Google. Cray claims to offer the same convenience of the cloud, but with greater computing power and more control over the hardware.
"You don't compromise on performance, that's the main difference," says Ulmer.
"When you go to a standard cloud provider, you don't get that level of performance that you normally would get. Now if you look at the model that we implemented with the Urika-GX, when you lease the system over at Markley, say for a week or so, the system is not shared with anybody else. So you don't have the influence of activities of other people on that system.
"You get the full performance of it. It's not a general purpose cloud infrastructure. It's really designed as a supercomputer, and you get that full design dedicated to you."
[Read next: 11 data centre disasters]
Ulmer adds that Cray's supercomputing as a service model also gives a level of control lacking in the other HPC options available in the cloud.
"If you want to achieve performance, you have to control the hardware to a certain extent," he says. "That's the whole thing behind supercomputing. And what we will do is we'll give you the ease of access, but we will provide that the system hardware stays controlled and you get the full Cray software stack, which allows you to scale your application further and let it run faster.
"You get the know-how and the expertise of Cray. You get the control over the hardware, as much as possible. But at the same time you get the ease of use of the cloud."