From its founding in 1993, NVIDIA has revolutionized the world of computer graphics, having introduced the first-ever GPU technology in 1999, sparking the growth of the PC gaming market, redefining modern computer graphics, and revolutionizing parallel computing.
The company is no stranger to innovation, and over the years, has continued to reimagine what’s possible in computing. Seven years ago, a new opportunity came to the fore with Alex Krizhevsky’s 2012 submission to Standford’s ImageNet challenge, and the creation of AlexNet, the first deep neural network powered by GPUs. The experiment showed a new way to approach AI that far surpassed past successes relying on CPU-based solutions. It also showed that the strengths in GPUs to parallelize problems, such as 3D renderings, were perfectly matched for solving similar parallelizable problems in deep neural networks.
“That was the year that the idea of deep learning, as opposed to more traditional kinds of machine learning, really took off,” says Dave Salvator, Senior Product Marketing Manager, Accelerated Compute, NVIDIA.” The pace of innovation and growth since then has been simply amazing, and it all started with our compute unified device architecture (CUDA) software.”
NVIDIA’s CUDA software had been around for over a decade, but now the company had a new opportunity to help shape the future of AI computing. NVIDIA realized that while AI as a discipline has existed for decades, its GPU technology could help usher a new age of AI—one where its transformative potential is no longer hype but reality, with the power to reshape entire industries.
As NVIDIA began to invest in deepening the AI capabilities of its products, it recognized that AI represented a new approach to software development and that they would need to provide not only hardware solutions but software, process, and infrastructure guidance including support from the broader AI ecosystem.
Alex Tsado, NVIDIA Go-to-Market Lead for the Microsoft account, explains, “We made a commitment to innovate at both a hardware and software level and take a platform approach to our technology. To us, building an accelerator is just table stakes—unless you have a great software stack, that accelerator is not going to accomplish much by itself.”
Today, the company has more software than hardware engineers on board—a testament to its commitment to offer clients a holistic, platform solution for AI.
NVIDIA’s AI vision found perfect alignment with Microsoft to build a software platform to empower developers to accelerate and build AI solutions, and make AI more accessible for all. The two companies already had a long-established partnership, collaborating across multiple levels, from engineering, to research, to sales, and AI offered a path to deepen the bonds between them.
“Today’s market continues to demand more intelligent solutions, which in turn require fundamental upgrades to compute,” said Ian Buck, Vice President, Accelerated Compute, NVIDIA. To make their joint offerings easier to use, NVIDIA is working with Microsoft on several integrations across managed services including AzureML, ONNX, and Azure IoT Edge, bringing together technologies and libraries with the goal of providing an easier way for developers to get started with AI. NVIDIA also enables AI on many Microsoft products like Bing and Cortana and continues to collaborate with Microsoft Research on new applications for AI.
Recently, NVIDIA helped the Microsoft Bing group upgrade the search engine’s capabilities to allow users to search for data faster, and to do so in a multitude of ways—from verbally asking a question, to searching objects within an image, or translating text into speech (for example, as you might do when searching on your phone).
The answers to all those queries are handled through an AI-powered service that includes multiple neural networks and were originally built using CPU-only servers. That infrastructure, however, could not deliver its computations within the latency budget needed by the Bing team. Microsoft switched to NVIDIA GPUs and immediately saw significant speed-ups that reduced latency by five times to provide users a more natural search experience.
“We actually worked hand-in-hand with NVIDIA by leveraging the latest in hardware features, like TensorRT optimizer. As Bing gets more complex, more natural, more intuitive, I definitely see more models [that] will benefit from GPU optimization in terms of inferencing," said Junhua Wan, Distinguished Engineer, Microsoft.
NVIDIA is also using its GPU technologies to improve people’s lives. They worked with Intelligent Retinal Imaging Systems (IRIS) to create a platform to detect diabetic retinopathy early in patients—helping to prevent one of the leading causes of blindness in the US.
Developed by retinal surgeon Dr. Sunil Gupta in 2011, IRIS captures high-resolution images of patients’ eyes that a clinician then uses to identify if a patient is at risk. IRIS wanted to scale that solution by making it available as a cloud service, built on Microsoft Azure so that any patient in the US could get scanned by their doctor. The IRIS solution also relies on NVIDIA GPUs to run its machine learning component, which allows deep learning algorithms to identify and categorize any pathology within an image.
With AI just starting to become a reality, the teams at NVIDIA are enthusiastic about continuing to work with Microsoft on cutting-edge research that pushes boundaries on what’s possible with AI. From enabling more conversational AI solutions, to further integrating hardware and software solutions to make AI more accessible and easier to use, and to new opportunities that look beyond what’s achievable today, the Microsoft and NVIDIA partnership is helping more businesses realize their vision with AI.
