Graphics Cards

NVIDIA’s Long Game: The Evolution from GPU Maker to AI Computing Company, Part 2

4 Oct, 2017 By: Cyrena Respini-Irwin

Tony Paikeday explains where NVIDIA is taking graphics processing units (GPUs), and where artificial intelligence (AI) will take CAD users.



Editor’s note: Click here to read NVIDIA’s Long Game Takes GPU from Graphics Workhorse to AI Powerhouse, Part 1

NVIDIA is on a mission to democratize how AI is used, said Tony Paikeday, NVIDIA's director of product marketing for Artificial Intelligence and Deep Learning. “[The development of DGX supercomputers] is one of the ways we created that democratization effect that’s part of this NVIDIA transformation. We wanted to be able to take supercomputing power — literally the power of 800 CPUs — and put it in a form factor that a customer could put in their data center, to train their mission-critical AI.”

To serve a similar purpose — democratizing how AI is used — but in a smaller package than the DGX supercomputers, the company recently began rolling out offerings based on the Volta GPU architecture that now powers DGX. (Although the Tesla V100 GPU is designed for data center use, Volta-based GPUs suited to designers’ workstations will follow.) NVIDIA says that Volta is designed to “deliver the performance of an AI supercomputer in a GPU” and “bring AI to every industry.”

Bringing these plans to fruition has not been cheap. NVIDIA is making a “significant” R&D investment on an ongoing basis, with a budget to the tune of $2.5 billion, according to Paikeday. The development of Volta alone required about $1 billion. “That level of investment is unprecedented,” he commented. As far as NVIDIA’s concerned, however, it’s necessary, however, since the company is investing deeply in not only development and manufacture of hardware, but also in the software side.

“It’s been interesting to note in the last couple of years, the trajectory of AI and what it’s done for us that kind of justifies the investment,” said Paikeday. “Like, deep neural networks have been growing; neural network development has been more and more computationally expensive as these networks get smarter, and by consequence bigger … [it’s the] GPU that’s allowed these models to get so big, so extensive, so computationally expensive, because we’re now providing the compute capacity to execute these massively parallelized algorithms that power the biggest neural networks that there are today, with speed that really wasn’t possible until developers way back when started training their models on GPUs instead of CPU servers.”

Human–Machine Design Partnerships

Paikeday sees an impact of GPU-powered computing on designers’ work that’s similar to its impact on deep neural networks. “Because of the economics introduced by systems like DGX, or like the power of the GPU that we’ve unleashed on AI, designers now can have machine learning–based platforms where they just feed the essential criteria under which a product has to be designed with certain constraints that have to be observed, like dimensions or weight or materials or cost or mechanical performance, and they can now iteratively create options that they can evaluate and select from,” he said. “Generative design is one of those exciting new areas where you’re starting to see machine learning get solutionized in a way that can help CAD professionals.”

The ideal is to make better use of designers by taking tedious tasks off of human shoulders: “We see now a really nice confluence of humans and AI working together, if you will, in the design process. People can now optimize and develop larger, more complex models, as mundane and smaller design problems get addressed by AI.”

In addition, he noted, AI can improve product reliability. If it is fed with historical data about performance, an AI can use analytics to predict when and where a component or system might fail.

Not everyone is thrilled by the prospect, however; some fear that introducing AI into the equation will reduce the need for humans, cutting jobs. In response to those concerns, Paikeday pointed out that although AI can enable faster iteration of design options, understanding the context — especially where cultural or artistic references are involved — and selecting the right design for each unique circumstance or application is ultimately a human task. “The element of intuition is innate to humans; it eludes AI and can be part of the value that we bring, where AI can’t go.”

Rather than replacing people, the help AI provides can instead improve their performance, and also reduce the stressfulness of their work, he believes. When humans are able to focus more intently on high-value design goals, that allows for more creativity in the design process, Paikeday said. “Maybe [they can focus on] the outliers that sit at the extremities of their design window, while AI helps offload the more mainstream work or problem types. So you’ll hopefully have your most valued innovators solving your highest-value problems, that they never had time to look at before.”

That’s a sharp contrast to the current situation for many professionals. For product designers, for example, the design/test/fail/redesign cycle is such a “Herculean effort” that it can create a “fear of failure” mentality and inhibit creativity, Paikeday believes. When there is no penalty for failure, however, designers can feel free to be more experimental and innovative in exploring design options, and ultimately better serve their end customers.

The AI-Assisted Future

Going forward, said Paikeday, organizations will need to think about how they “infuse” their businesses with AI to improve efficiency and speed of workflow. The specifics, of course, will vary from one industry and market to the next, but the ability to evaluate more permutations or “what-if” scenarios, without investing more time or effort, should benefit all types of designers.

“Hopefully, as a consequence of this we’ll see more designs and better designs coming to fruition faster, at a lower cost than was possible before,” he predicted. “We’re going to see these waves of innovation as people start to embrace the tools.”


Add comment

Note: Comments are moderated and will appear live after approval by the site moderator.

AutoCAD Tips!

Lynn Allen

In her easy-to-follow, friendly style, long-time Cadalyst contributing editor and Autodesk Technical Evangelist Lynn Allen guides you through a new feature or time-saving trick in every episode of her popular AutoCAD video tips. Subscribe to the free Cadalyst Video Picks newsletter and we'll notify you every time a new video tip is published. All exclusively from Cadalyst!

Follow Lynn on TwitterFollow Lynn on Twitter


Poll
Which device do you typically use to read Cadalyst.com content?
A desktop computer / tower workstation
A tablet
A smartphone
A laptop or mobile workstation
I regularly use both a desktop computer and a smartphone for this purpose
I regularly use another combination of devices for this purpose
I prefer to print out articles from the website and read them on paper
Submit Vote



Download Cadalyst Magazine Special Edition