Patience in a slow-growing market is hard, but supercomputing cloud startup Rescale is reaping the rewards of its persistence.
When the company launched in 2011, companies were just getting to grips with jumping to the cloud for overcapacity and in some cases moving all operations off-prem. Cloud was the buzzword with plenty of emerging use cases to bolster its legitimacy — at least in enterprises. High-performance computing was a reprieve back then, and some might argue that it has remained so until recently. For starters, it was difficult, especially with restrictive software licenses and implementation complexity. But the cloud providers themselves have made onboarding apps a snap, and there are plenty of ways to build even the most latency-sensitive, proprietary licensed HPC applications for both hybrid and full cloud runs.
Before and even after these additions with or without specialized cloud tooling from vendors like Rescale, HPC was slower than the rest of the world to adopt the cloud. Part of this was because technical computer software licensing models weren’t ready for the cloud, but in many cases it was because performance was too important for the latency hit. But those days are long gone, especially when it comes to performance. With some quick kicks from global events, HPC clouds are now a contender, especially for medium HPC deployments.
And it’s clear that some of the world’s most powerful investors see plenty of room at the top of this market, even if the biggest public clouds have worked to make all those old-fashioned HPC cloud problems naturally disappear.
HPC research firm Hyperion says the cloud for HPC applications is expected to grow 2.5x faster than the on-prem HPC server market with a new turning point in demand driven by the pandemic, which will drive administrators off-site and to various public clouds pushed. While the COVID crisis may have been a physical driver, other factors including leaps in performance capabilities (computing, network, storage), simpler cloud/hybrid deployments, and more attractive pricing models have also pushed HPC stores to the cloud.
Rescale has about 200 customers, which may not seem like much compared to the major public clouds, but remember they don’t provide the hardware, meaning they can scale with demand without investing in new data centers or colo facilities.
Although they call themselves a cloud, it is a collection of automation tools to enable easier deployments of complex, performance-sensitive applications on specific public clouds or a combination thereof. We described the platform itself back in 2015, and while the company is (and always has been) coy about how it all works, it has hooked up with AWS and now Azure (Microsoft acquired Cycle Computing, which did essentially the same thing – calling itself a cloud, but really just providing an application and automation platform to sit on top of that perceived complexity).
In that 2015 piece, based on an interview with Resacle CEO, Joris Poort, he said each of the data center/cloud providers could bring something different to the table. Some provide resources that exist solely because of their geographic importance to a particular set of technical end users, others because they are equipped with GPU accelerators, thick memory, low-latency InfiniBand networks, and other more specialized hardware elements.
What was true in 2015 remains today. Then Poort told us, “For users, the real benefit is that they can get on-demand or hourly licenses – that removes a lot of the limitations of what can be achieved. For example, at the most basic level there is an engineering software package that a user can license to want to bring – that’s easy through our portal. But at the top level, with something like Siemens PLM, we have that whole suite of simulation tools that we can buy on demand or hourly directly through our interface. There is an hourly rate, although there is a premium is because it’s hourly versus yearly, but it’s there and available immediately.
Since then, they’ve brought in quite a bit of business, or so we assume, adding up-to-date ANSYS, LS_Dyna, CD-ADAPCO, and other technical tools. The next frontier will be adding more EDA suites to capture that growing market, something we’ll get to shortly.
In Rescale’s announcement of $105 million in Series C financing, Poort points to an acceleration in demand at total HPC market spend of $60 billion by 2025, with HPC clouds showing an 80% compound annual growth rate. “Many workloads in the general science research and development category—estimated at $185 billion by 2020 and mostly still running in on-prem data centers—are also reaping the benefits of hybrid cloud computing.”
Investors are also seeing demand for specialized cloud services such as Rescale’s peak. The startup’s list of individual investors includes Jeff Bezos, along with Peter Thiel, Sam Altman and Richard Branson. Institutional investors include NVIDIA, Prometheus Ventures, Keen Venture Partners, Samsung Catalyst and Hitachi Ventures.
Wolfgang Seibold of Hitachi Ventures sees growing demand for HPC areas. “Computational science and engineering underpin the innovations our society will depend on in the next decade in energy, life sciences, manufacturing and transportation.” He adds, “The core software and algorithms used in these industries are computer-bound, linking the pace of innovation with the rate of computation. Rescale frees up innovators to create the next great ideas.”
While Rescale would have been able to slowly conquer HPC stores that are either bursting or moving completely to the cloud, the next leap for HPC cloud is much bigger. The first step is to collect some of the AI training activities, the second is to convince enterprises that what they are doing is “enough HPC” to be a great fit for their cloud. This means moving beyond the HPC application catalog to more business and analytic applications, which the company has been working towards with its Rescale Insight enterprise management product. This is a suite of more than 400 applications that serves as a managed full-stack IT control center for enterprises.
Another growth segment ripe for Rescale to pick is government. The company recently secured federal compliance through its ScaleX Government platform, a fully managed HPC service that is FedRAMP Moderate and ITAR compliant for US customers.
What Rescale must continue to prove is that it can secure large organizations. In 2019, the company added Nissan to the list of public users. Nissan used the Rescale ScaleX platform on AWS for selected engineering applications. Another engineering simulation company, Vertical Aerospace, has also used Rescale for its aircraft simulations.
“By choosing Rescale as our strategic cloud partner for digital R&D, Vertical Aerospace has dramatically increased technical efficiency and accelerated the commercialization of new products without compromising quality or safety,” said Madhu Bhabuta CIO, Vertical Aerospace. “The shorter lead time delivered by Rescale’s offering has enabled us to perform approximately four times as many aircraft envelope calculations as we work to refine our flight models within the same timeframe as before. In addition, we have been able to run other simulation programs using their comprehensive suite of optimized core architectures that have also seen a similar degree of speed.”
EDA has been one of the most talked about HPC cloud capabilities of 2020 with those professionals moving off-site and new applications (and redesigned to fit more modern devices coming to market). Rescale also sees a valuable opportunity in EDA, especially with increased demand for semiconductors on tight schedules. EDA company pSemi used Rescale for its simulations and claimed a 15-hour reduction in time to results, compared to 7 days with their on-prem systems.
Rescale can only grow as fast as the HPC cloud market, and while it’s growing fast, their main competitors are the cloud vendors they rely on for infrastructure. With AWS and Azure making it easier than ever to onboard HPC applications, see clear pricing, grab spots and other lower-cost instances, and easily acquire SaaS licenses and go straight into production, one day where rescaling is less necessary.
While these are still development days for HPC clouds, they are not early days. The retention that was necessary in 2015 or even 2018 is not as necessary as when Rescale and Cycle Computing started early. AWS, Azure and Google Cloud Platform are making cloud adoption downright lighthearted compared to even five years ago, forcing Rescale to work harder to tell a compelling TCO story.