Leveraging High Performance Computing to drive AI/ML workloads

Artificial Intelligence (AI) has the potential to transform our lives – from enabling breakthrough scientific discoveries, personalised medical diagnosis and treatment, accurate weather forecasting, advanced fraud detection, autonomous driving, and more.

The transformational power of AI is not only recognised by the private enterprises but even the governments across the globe, including India, are acknowledging its potential and applicability in empowering citizens and societies. In a recently organised virtual global AI summit, Responsible AI for Social Empowerment (RAISE 2020), Prime Minister Narendra Modi highlighted the importance of AI in today’s life and the big role this technology plays in sectors like agriculture, creating next-generation urban infrastructure, and making disaster management systems stronger. Apparently, the Government of India is planning to create a robust AI-powered public infrastructure that would benefit people in India as well as across the world.

Today, businesses across industries are reinventing their processes to take advantage of AI-driven capabilities, from chatbots to robotic vehicles to smart drones, etc. This rapid adoption of AI-driven services has brought about a corresponding demand for High-Performance Computing (HPC), as it serves as the infrastructure foundation for enterprise AI.

Also, High-Performance Computing is accelerating this transformation because it takes a lot of computational power to train deep learning models, do AI inferencing in real-time, and run analytics on massive amounts of data.

According to a report from Mordor Intelligence, the High-Performance Computing market is expected to witness a CAGR of 6.13% over the forecast period from 2020 to 2025. Factors, such as increasing investments in the Industrial Internet of Things (IIoT), Artificial Intelligence, and engineering, which demand electronic design automation (EDA), are likely to drive the market over the forecast period.

Having said this, processing, moving, and storing different types of data are not the norm for High-Performance Computing infrastructure as it brings along specific challenges with it. This is where computing technology like GPU comes into the picture to help process different data types for AI use cases. GPUs are the driving force behind AI initiatives and seen as an optimal vehicle to run AI workloads.

However, one cannot deny the fact that there is a roadblock in the mass adoption of AI technology. Due to rapid advancements in GPU technologies, these supercomputers tend to get obsolete and even difficult to maintain. At the same time, investing in an HPC infrastructure is a costly affair for any enterprise. The solution to this is to make the underlying infrastructure accessible to all on an OPEX model, wherein organisations can simply rent HPC computing on-demand, without having to make any CAPEX investments.

More and more enterprises are using High-Performance Computing today

We are aware that AI, with its range of use cases, require immensely powerful processes across compute, networking and storage, and organisations benefit most when compute is closest to the origin of data, wherever it resides. AI workloads, like Machine Learning (ML) and Deep Learning (DL) are being built atop HPC infrastructure to support this demanding compute and data-intensive nature.

This is the reason more and more enterprises are increasingly using HPC solutions to drive AI-led innovation. For instance, research organisations are using HPC technologies to do data analytics for training machine learning models, enabling them to understand and gain insights from digital data.

Besides, HPC-enabled AI could provide optimisation of supply chains, complex logistics, manufacturing, simulating and support modelling to resolve any problem. For example, science and academia are using HPC-enabled AI to provide data-intensive workloads by simulation and data analytics. Other vertical industries like BFSI, automotive, manufacturing, and healthcare require powerful computing and in-depth data analysis; and this is where they are turning to HPC and powering their AI workloads.

It is also widely believed that AI plays a key role in improving governance and its use cases cater to areas like social welfare, policymaking, and healthcare. While the superpowers like the US, the UK, and China have made big strides in this direction, the country like India is on the path to becoming an AI giant as it has a robust IT ecosystem and the capabilities to democratise any such technology.

Recognising the need of HPC to power AI workloads, the Government of India is working with Centre for Development of Advanced Computing (C-DAC) to develop the country’s largest HPC-AI supercomputer, PARAM Siddhi – AI, to facilitate AI framework. In fact, computing facilities are being set up in the country so that researchers and tech companies can utilise the infrastructure to run their algorithms and build unparallel AI products.

In this endeavour, the Indian government is also planning to create AI-specific AIRAWAT (AI, Research, Analytics, and Knowledge Assimilation) compute infrastructure. Under AIRAWAT, the Indian government plans to handle issues associated with lack of access to computing resources. Besides, the Government of India will be building AI-specific compute infrastructure that will help the computing needs of Centres of Research Excellence (COREs), International Centers Transformational AI (ICRAIs) and Innovation Hubs.

Unlocking the true value of data 

The data-driven era that we are in is transforming the industries and re-inventing the future of enterprises. It is extremely critical and transformational for organisations across industries to leverage high volumes of data being generated by emerging technologies. And there is no doubt that data helps in deriving insights that lead to the success of any organisation.

This is where technologies like AI and Analytics come into the picture that helps in unlocking the true potential of data. Moreover, data is the vital national resource that will act as a raw material for AI-led development and nations like India has an advantage due to its enormous digital capital.

These technologies have the capabilities to address the challenges that have arisen due to the influx of high volumes of data which requires the power, scalable compute, networking, and storage provided by HPC. As a result, organisations are turning to High-Performance Compute solutions to enable high-performance data analytics, enabling researchers and organisations to gain faster market insights, increase efficiency, and recognise higher ROI for data‑driven investments.

Hence, HPC workloads are becoming more data-centric and adding AI technologies to advance the capabilities of traditional HPC modelling and simulation. In the next few years, we will witness HPC technologies, such as HPC-enabled machine learning training, to graduate from the experimentation phase to creating production-ready models.

Furthermore, with the advent of IoT and AI, smart living has become increasingly popular. Smart buildings, schools, factories, and hospitals are generating vast amounts of actionable data. To make this data useful and sensible in quick time, the AI engine powered by HPC is a must.

High-Performance Computing & AI – the perfect match

Both HPC and AI are intertwined and the convergence of the two will drive business and innovation for enterprises across verticals. High-Performance Computing supercharges AI to not only make it smarter and faster but also helps in providing accurate results. This clearly shows that there is no complexity in the way AI and HPC collaborate. For example, AI applications and its workloads, Machine Learning and Deep Learning are helping enterprises to train systems leveraging data to gain insights. At the same time, HPC clusters help connect the dots between this data at a much faster speed. It is very much evident here that modern AI cannot exist without access to HPC.

Moreover, HPC can better support AI model training than traditional systems. Organisations can use AI to line up and process workloads, thus, maximising the resources of HPC systems. Additionally, both HPC and AI also require a high-performance infrastructure, including large volumes of storage, compute power, high-speed interconnects, and accelerators. The organisations looking to leverage both the technologies clearly understand these requirements needed to develop powerful tools.

As AI models continue to grow in their size and so does the need for computing power, emerging workloads will rely more heavily on the technologies underlying HPC. By now, it is evident that the key to combining AI and HPC is scalability; and by incorporating HPC into your AI/ML environment, organisations will get the capability to boost and scale emerging workloads.