Nvidia Founder/CEO Jensen Huang, in a fast-moving fireside chat at Siggraph 2024, discussed generative AI, particularly how we got to this point and where we are headed with the technology. He also discussed one of the main issues with GenAI, power consumption. He also touched on several Nvidia Siggraph announcements related to GenAI, including a new suite of NIM microservices, its partnership with Shutterstock in its generative 3D service, and more.
Nvidia Founder/CEO Jensen Huang kicked things off at Siggraph 2024 with a fireside chat conducted by Wired’s Lauren Goode amid a packed house. Huang spoke in his often lighthearted manner while highlighting seriously complex technology—he is a person who knows how to maintain his audience’s attention while talking about some pretty complex stuff.
Huang expressed his excitement at being at Siggraph—and was sporting a new leather jacket, a present from his wife for the occasion, he said. As Huang noted, Siggraph used to be about computer graphics (and interactive techniques); now it’s about computer graphics, generative AI, and simulation—topics of significant importance to him and Nvidia. He then recounted some of the most important milestones in the computer industry and Nvidia’s related journey, leading up to where it stands today.
The road to accelerated computing and GenAI
Indeed, there are breakthroughs and discoveries, and then there are those that are total game-changers. CUDA falls into the latter category, opening parallel processing capabilities of GPUs to science and research. “We had been working on accelerated computing for a long time, promoting and evangelizing CUDA, getting CUDA everywhere and putting it on every one of our GPUs so that this computing model was compatible with any application that was written for it, irrespective of our processor generation,” said Huang, calling it “a great decision.”
In 2012, the company observed a big breakthrough in computer vision with the AlexNet neural network, and at the core was something deeply profound, deep learning, which was a new way of writing software. Instead of engineers providing input, imagining what the output was going to be, and writing algorithms, now a computer that was given input and example outputs could the program. As a result, that technique could be used to solve a plethora of problems that previously wasn’t solvable, said Huang. And, Nvidia changed everything within its company to pursue that, he said. From the processor, to the systems, to the software stack, all the algorithms, and basic research, it all pivoted toward work on deep learning.
And AI tech has progressed since then at lightning speed. In 2018, Nvidia reinvented computer graphics by introducing RTX, making it possible to use a parallel processor to accelerate ray tracing. Then, DLSS came along to give that a big boost and to make it real time, interactive, using AI to infer other pixels. Now, Huang said, the industry can render fully ray-traced, fully path-traced simulations at 4K, at 300 FPS, made possible by generative AI.
“This is really a revolutionary time that we’re in. Just about every industry is going to be affected by this, whether it’s scientific computing trying to do a better job predicting the weather with a lot less energy, to augmenting and collaborating with creators to generate images, or generating virtual scenes for industrial visualization. And, very importantly, robotic self-driving cars are all going to be transformed by generative AI,” said Huang. “So here we are in this brand-new way of doing things.”
In just a short period of time, generative AI has progressed significantly and will continue to do so, Huang said. He then advised the artists in the world to jump onto this tool and give it a try. It is not going to disappear, he told the audience. In fact, he believes that in all likelihood, every one of our jobs will change in some way because of it.
“Everybody will have an AI that is an assistant. And so every single job within a company will have AIs that provide assistance. Our software programmers have AIs that help them program and help debug software. We have AIs that help our chip designers design chips,” he said, noting that without them, AI Blackwell wouldn’t have been possible.
Perhaps surprising to many there based on Nvidia’s success in the GPU (hardware) market, is Huang pointing out that it has always been a software company. The reason for that, he said, is because accelerated computing is not general-purpose computing. You have to understand the algorithms in order to do a good job at accelerating applications 20, 40, 50, 100 times.
All that is well and good, but there are energy consumption concerns swirling around generative AI, which is incredibly energy-intensive. Huang acknowledged that the computational load is growing, but with the company’s Blackwell, an application is greatly accelerated while using the same amount of energy, he professes. Accelerated computing helps you save energy, as much as 20 times, 50 times, while doing the same amount of processing, he said.
“So, the first thing that we have to do as a society is accelerate every application we can,” Huang said. “The density of our GPUs and density of accelerated computing is higher, energy density is higher, but the amount of energy used is dramatically lower.”
The next thing pertains to the goal of generative AI, which is not training, but inferencing. GenAI is going to increase productivity, Huang pointed out, and will help make things more energy-efficient. With traditional retrieval-based computing, everything is pre-recorded and stored off somewhere in a data center. Generative AI reduces the amount of energy necessary to retrieve something from a data center and send it over a network. Huang said data centers account for about 40% of the total computing that’s done. The other 60% of energy is consumed on the Internet, moving bits and bytes around. GenAI will reduce that, he noted, because rather than retrieving info from the Internet, it can be generated right on the spot.
One final thought on that topic: “AI doesn’t care where it goes to school,” Huang said. Today’s data centers are built near the power grid, where society is, because that’s where we need it. In the future, data centers will be built in different parts of the world, where there’s excess energy. “AI’s doesn’t care where it’s trending. Don’t move the energy to the data center, use the energy where the data center is,” he added.
More GenAI insights from Huang are revealed in his discussion with Meta Founder/CEO Mark Zuckerberg, summarized here.