High-performance computing is now available on the cloud as a service. That has opened new vistas for organisations to manage high-compute, critical applications
We don't support landscape mode yet. Please go back to portrait mode for the best experience
Chennai-headquartered automobile major TVS Motor Company uses high-performance computing (HPC) for running R&D simulations and testing the aerodynamics of two-wheelers, which allows it to make the vehicles stable at speed and more efficient, cool engines faster, and achieve better top speeds, quicker. It has been doing this for, like, forever, but something changed in 2020. Earlier, TVS would purchase HPC hardware upfront for five-year periods, to be installed, run and maintained by its own team. As Ranjith Radhakrishnan K.C., AVP-IT, TVS Motor Company, informs us, in this ‘capex’ model, “we had to actually pay upfront for five years, and when we revised something, we had to make sure that it lasted five years”.
In 2020, TVS switched to HPE GreenLake’s HPC-as-a-Service (HPCaaS), an ‘opex’ model perched on its premise. Ergo, pay per use, and all other attendant cloud benefits. “Since this HPCaaS model is scaleable, we don’t have to invest in what we may require after two-three years, or even more,” says Radhakrishnan. “We are able to invest only in whatever is required for this current financial year.” Another perk is the reduction in lead time for ordering new hardware, which has come down to about four weeks from eight weeks earlier. “This model enabled us to be much more nimble as we don’t have to plan two-three months in advance.” Explains Sinisa Nikolic, Director & Segment Leader, HPC & AI, Lenovo ISG, Asia-Pacific: “It is not the technology that differentiates HPC and HPCaaS, but the way an organisation uses and funds it—which means it is either a capex or an opex strategy.”
As TVS’s use case demonstrates, HPC is not your usual network of desktop computers running ERP or CRM software. It is immensely powerful. This power is needed, say, for drug discovery of Covid-19, which requires running a few trillion calculations with accurate simulation and modelling in a short period of time. “HPC conducts complex calculations at high speeds across multiple servers, simultaneously. Collectively known as clusters, these groups of servers are composed of hundreds or even thousands of servers connected through a network,” explains Sayed Peerzade, EVP & Chief Cloud Officer, Yotta Infrastructure, a managed data centre service provider. Many times, HPC networks are actually networks of supercomputers. Globally, such systems are used in multiple high-compute areas such as semiconductor design, streaming a live sporting event, oil and gas simulations, genomics,etc. India, too, has been using HPC in weather and disaster forecasting, academic research, etc., for decades.
It is not the technology that differentiates HPC and HPCaaS, but the way an organisation uses and funds it—which means it is either a capex or an opex strategy.
Sinisa Nikolic
Director & Segment Leader
HPC & AI, Lenovo ISG, Asia-Pacific
We know what you’re thinking—such systems are expensive. Installing HPC systems can cost anywhere between $25,000 and $1 billion, plus operational and maintenance costs. That’s costly and complex, both factors influencing companies’ tech departments to veer towards HPCaaS. “HPCaaS removes some of the complexities of designing and deploying an HPC data centre, as well as running optimised HPC applications,” explains Alex Norton, Research Director, Lead AI and Co-Lead Cloud Analyst at Hyperion Research. Available as a service through on-prem or in the cloud, on a monthly subscription model (eliminating upfront investment), it makes compute-intensive processing possible for enterprise and academia. In HPCaaS, all the HPC infrastructure including switching, routing, internet bandwidth, intrusion protection systems, etc., are managed and delivered by vendors as a part of the subscription. Scaling up, and incremental payment, is flexible based on an enterprise’s workload change.
According to Hyperion Research, the on-premise HPC server market was $13.5 billion in 2020 globally, projected to grow to nearly $20 billion by 2025. Include add-on storage, applications, and maintenance services, and the on-premises HPC market is expected to reach $40 billion by 2025, from $26.3 billion in 2020. On the other hand, HPCaaS worldwide is around $4.5 billion today and is expected to double to $9 billion by 2025. In India, on-premise HPC spend is around $400 million a year, projected to reach $600 million a year by 2026. As for HPCaaS in India, it is expected to grow to an estimated $102 million by 2025 from $47 million in 2020. “HPCaaS is growing in India, but slightly slower than the rest of the world,” says Norton.
HPCaaS removes some of the complexities of designing and deploying an HPC data centre, as well as running optimised HPC applications.
Alex Norton
Research Director, Lead AI and Co-Lead Cloud Analyst
Hyperion Research
Industry Adoption
Earlier, supercomputers were largely used for nuclear test simulation, mapping the human genome or resurrecting dinosaurs in movies, and reserved for government and medical researchers, academics, and innovative moviemakers. But the rise of data-intensive technologies such as artificial intelligence (AI) and machine learning (ML), which require massively parallel compute (MPC) capabilities, has resulted in a wide range of organisations examining HPC solutions. These are being used for varied cases such as computer-aided design and engineering, autonomous driving, production optimisation, predictive maintenance, drug discovery, precision medicine, fraud and anomaly detection, treasury and trading analytics, Internet of Things (IoT), smart cities, and more.
While HPCaaS follows a cloud-based delivery model, these sophisticated systems feature an advanced combination of hardware as well as software. The key hardware for this includes high-performance CPUs, fabric, memory and storage, networking components, plus accelerators for specialised workloads. This high-performance infrastructure at the core is backed by data connectivity and redundant availability. HPC platform software, libraries, optimised frameworks for big data and deep learning, and other software tools, are value-adds one gets access to while signing up for HPCaaS.
What’s the outcome? Stunningly complex work can be done, relatively affordably, on the cloud. For example, Samsung Electronics is providing its fabless customers a virtual environment—Samsung Advanced Foundry Ecosystem (SAFE) Cloud Design Platform (CDP)—to design chips in the cloud in collaboration with Rescale. By adopting CDP, Samsung’s customers can reduce the burden of building their own server infrastructure, while flexibly utilising additional computing power required for chip design and verification.
HPE Cray Liquid Cooled Supercomputer
Cool, right? Yes, but in case you are wondering, HPCaaS has extended its reach beyond such exotic applications to the relatively mundane. A leading private sector bank in India has opted for HPCaaS for algorithms such as quick loan approvals and fraud detection in loans to industry. CARE Ratings uses HPCaaS for data analytics projects to develop sentiment analysis and predictive models for its ratings business. “As neural network models recommend GPU-based processing, we use HPCaaS in the development of neural network-based deep learning models,” says Tarun Bangera, Head-Data Science & Analytics, CARE Ratings. “The high-performance cloud-based environment has helped us develop a scaleable framework for building AI/ML models, ultimately helping us in distributing workloads seamlessly. Cloud computing has enabled us to automate complex, repetitive, high-processing tasks without any human intervention.”
Som Satsangi, SVP and Managing Director of HPCaaS vendor Hewlett Packard Enterprise, India, says: “In the recent past, there has been a steady shift of HPC to the cloud, which has been the fastest-growing part of the HPC market. It becomes difficult for enterprises to manage infrastructure because of the rate and pace of change in technology and the skills needed to operate on-premises HPC. So, instead of buying technology, they are looking to use it as a cloud service.”
Technology consulting companies, too, are leveraging HPCaaS. Tata Elxsi, which provides technology consulting, product design and development services to organisations globally, is leveraging a cloud-based HPC solution as it found it far more feasible in the context of service to clients, sales closures and timely deliveries, apart from other usual benefits of a cloud model. “The system goes through many upgrades over the years as technology evolves, resulting in additional costs incurred every few years to maintain an up-to-date physical infrastructure. The on-cloud HPC on (Microsoft’s) Azure was a much better option,” says Sagar MSV, the company’s Manager-Transportation Design & Engineering.
Microsoft is making its HPC services available through the Azure portal and marketplace—Azure Cycle Cloud and Azure Batch. “These services are free to use, and customers pay for the underlying compute to use with these services. The overall cost depends on the customer use case, workload, and the time duration of running those workloads on Azure,” explains Shivir Chordia, Azure Business Group Lead, Microsoft India. In India, Microsoft says it has improved time to market with faster solve times for a large engineering services organisation, enabled a large engineering university to run molecular simulation workload, and enabled researchers in a large organisation to focus on innovation and research, taking away the overheads of running HPC clusters on-prem.
HPCaaS is proving to be a boon for enterprises and start-ups relying heavily on AI and ML. Orbo.ai, an AI and computer vision-based platform focussed on image enhancement and automation, has to perform extensive AI training over huge image and video datasets. This training requires compute-efficient systems that can run for days without any interruption. Currently, the company is training AI for hair segmentation, face landmarks, signature matching and foetus scan enhancement on Yotta’s HPC virtual machines. “We have been using HPCaaS since the past seven months, and we are deriving two direct benefits from it—uninterrupted system availability for our AI research team, and cost savings,” says Md Danish Jamil, Co-Founder and Technical Lead, Orbo.ai.
In the recent past, there has been a steady shift of HPC to the cloud, which has been the fastest-growing part of the HPC market. It becomes difficult for enterprises to manage infrastructure because of the rate and pace of change in technology and the skills needed to operate on-premises HPC. So, instead of buying technology, they are looking to use it as a cloud service.
Som Satsangi
SVP and Managing Director
Hewlett Packard Enterprise
As HPCaaS is still nascent in India, players in this space are experimenting with new ways to bring awareness and increase adoption. Bengaluru-based NxtGen Datacenter & Cloud Technologies is offering 100,000 free core hours of accessing HPCaaS for educational and technical institutes. “More than 10 prestigious institutes are running multiple complex job runs on the cluster consuming more than 50,000 to 70,000 core hours each,” says Rajesh Dangi, Chief Digital Officer, NxtGen Datacenter & Cloud Technologies, adding that most of these institutes are already in advanced discussions for long-term contracts with the company.
The Cost Benefit
Instead of investing millions, HPCaaS can be accessed for a starting price of as low as Rs 11,000 per month, which could run into a few lakhs depending on the configuration and workload. Yotta’s most basic Eco plan starts at `11,000 with unlimited data transfer and host anti-virus. And its most high-end offering—Baremetal Premium—is available for `1.74 lakh per month (see The Pricing for details). The company says these plans are capable of handling different kinds of workloads. The Eco configuration is targeted at students and academics who use legacy compute (old technology, computer system or application, related to an older or outdated computer system). And premium plans are designed for industrial and enterprise use, for example, manufacturing industries using AI for production, or financial services companies working on analysis of big data and real-time fraud detection, among others.
A big concern that HPCaaS addresses is security, as it typically follows the same checks and balances as any other application being served through the cloud. Using a consumption model, especially through external providers, is a two-way street. “Both the user and the vendor have a significant role to play in securing the cluster. Securing data and IP become key areas that need special attention, and encrypting the data while at rest and in flight play a key role in securing the same,” explains Manish Gupta, Senior Director and General Manager, Infrastructure Solutions Group, Dell Technologies, India. HPCaaS providers follow a strict security and protection protocol and, as a result, many companies who started with HPC on-premises, tend to supplement their workload with HPCaaS.
Which brings us back to TVS Motor. The company has moved 25 per cent of its HPC workload to HPCaaS, and as the existing HPC cluster reaches the end of its life, the newer clusters will be taken on HPCaaS. Given that TVS has had a history of using HPC on-premise, you would be forgiven for thinking that TVS would be more capable of leveraging HPCaaS. Does that mean others who have not used HPC before, can’t use HPCaaS? Just the opposite holds true. That’s the beauty of the cloud-based approach.
