Empowering AI with Microsoft Azure's Purpose-Built Infrastructure

Table of Contents

1. Introduction
2. Azure AI and AI Infrastructure
3. The Need for AI-First Infrastructure
4. Microsoft Azure's Purpose-Built AI Supercomputers
5. Azure Machine Learning for AI-First Infrastructure
6. Scale-Up-and-Out HPC Supercomputers
7. AI-First Infrastructure and Toolchain for Any Scale
8. Use-Case Examples in Manufacturing
9. Use-Case Examples in Retail
10. Use-Case Examples in Healthcare
11. Conclusion

Azure AI and AI Infrastructure

Artificial intelligence (AI) has reached a critical juncture, experiencing quantum leaps in model complexity. This complexity has resulted in a corresponding requirement for IT infrastructure. The need for AI-first infrastructure has been established, one that can not only scale up to take AdVantage of accelerators within a server but also scale out to embrace many servers distributed across a network. Microsoft Azure is the only public cloud that offers purpose-built AI supercomputers. For example, massively scalable IT infrastructures comprised of NVIDIA InfiniBand interconnected NVIDIA A100 Tensor Core GPUs.

The Need for AI-First Infrastructure

The progress of AI has been astounding from just a few years ago, with solutions pushing the envelope by augmenting human understanding, preferences, intent, and even spoken languages. AI is improving our access to knowledge, helping us provide more efficient solutions that fuel transformation beyond imagination. With the rapid growth and transformation, AI demand for compute power has grown by almost four orders of magnitude, outpacing Moore's law. With AI powering a wide array of important applications, including natural language processing, robot-powered process automation, and machine learning and deep learning, AI companies are finding new ways to get more out of each piece of silicon using tools like mixed-precision modes, enabling them to do more with less.

Microsoft Azure's Purpose-Built AI Supercomputers

Microsoft Azure is the only public cloud that offers purpose-built AI supercomputers. These supercomputers are comprised of NVIDIA InfiniBand interconnected NVIDIA A100 Tensor Core GPUs. Azure Machine Learning facilitates uptake of this AI-first infrastructure from the earliest stages of development through enterprise-grade production deployments that require full-blown MLOps. Scale-up-and-out HPC supercomputers are both performance and power efficient. HPC technologies have significantly advanced science and engineering from innovations in hardware to software, parallelism and communication have been repeatedly leveraged to advance the technology that comprises HPC infrastructure.

Azure Machine Learning for AI-First Infrastructure

Azure Machine Learning is a powerful tool for building and operationalizing deep learning models that execute on NVIDIA GPUs. Model training makes use of the N series virtual machines on Azure, whereas inferencing is often shifted to the edge. These data-driven solutions require sophisticated deep learning models, models that are much more sophisticated than those offered by machine learning alone. In turn, this sophistication demands an enabling AI-first infrastructure and toolchain.

Scale-Up-and-Out HPC Supercomputers

Scale-up HPC takes advantage of shared memory to introduce parallelism via Threads. When combined with the vector-processing capabilities available from GPUs, these SIMT devices have proven extremely effective at processing arrays of data, in fact, multi-dimensional arrays of data. With the added capability of a high-bandwidth, low-latency interconnect Fabric, scale-out HPC takes advantage of distributed memory for parallel computing by interleaving computation and communication across a network of computer nodes. Scale-up-and-out HPC combines the attributes of vertical and horizontal system scaling to address the most-demanding workloads from science and engineering.

AI-First Infrastructure and Toolchain for Any Scale

An AI-First Infrastructure and Toolchain for Any Scale aims to address the challenge of scaling up real-world applications without slowing down performance via fast, low-latency connections such as InfiniBand interconnected GPUs from NVIDIA. AI-first infrastructures and toolchains are proving to be of value in most industries. Here, we provide use-case examples from manufacturing and retail as well as a compelling customer story from healthcare.

Use-Case Examples in Manufacturing

AI-first infrastructures and toolchains are having significant impact in manufacturing. Based upon industry trends, we've identified two use-case examples, namely predictive maintenance and product quality. By using NVIDIA GPUs on Azure for predictive maintenance and product quality, customers can see a plethora of benefits in scaling, speed, quality, and accuracy. Of course, these benefits are the outcomes derived from technology-based solutions that address the nuances of each Scenario. Each of these examples requires AI-first infrastructure and toolchain as the scenarios under consideration are demanding ones, for example, to significantly reduce false positives and negatives in predictive maintenance to account for subtle nuances in ensuring product quality.

Use-Case Examples in Retail

AI-first infrastructures and toolchains are having a significant impact in retail. Here, we've used deep demand forecasting and the connected store as use-case examples. Primary benefits include improvements in performance, training times, TCO, ROI, and shrinkage, all key areas for companies in the retail industries. Like in manufacturing, these benefits are the outcomes derived from technology-based solutions that address the nuances of each scenario. Each of these examples requires an AI-first infrastructure and toolchain as the scenarios under consideration are demanding ones, for example, to significantly improve the timeliness and accuracy of demand forecasting for retailers to deliver a Frictionless experience to customers in a retail storefront scenario.

Use-Case Examples in Healthcare

AI-first infrastructures and toolchains are also being used in the healthcare industry. At Sensyne Health, we work with some of the largest pharmaceutical companies, delivering research insights based on questions that they pose on real-world data. The aspect of COVID for us was a massive amount of uncertainty. The one thing that we did know at the time was that diagnosis and diagnosis at scale would be a major part of what was going to be needed. Magnify is a solution which allows You to turn a phone remotely into a diagnostic tool. We worked out that we could incorporate our image capability using very powerful neural networks to actually very accurately determine what that reading actually was after you've taken it, and the reason that's important is that the human eye and the brain that interprets the image is very varied. Too many false negatives puts an added burden on the healthcare system. It means that more people will be misdiagnosed and therefore take incorrect action. Magnify puts clinical grade diagnostics in the home of any end user, not just a healthcare professional, but a mum, a dad, an aunt, an uncle, it doesn't matter.

Conclusion

AI-first infrastructures and toolchains are proving to be of value in most industries. Microsoft Azure in partnership with NVIDIA delivers purpose-built HPC and AI hardware in the Cloud to meet even the most demanding real-world application workloads at scale, while meeting price-performance and time-to-solution requirements. And with next-generation machine learning tools included, you can incorporate intelligence within your workloads to drive smarter simulations and empower intelligent decision making.