NVidia’s New AI SuperChip Is Set To Herald In A New Age

0
146
Nvidia

Nvidia has expanded its lead in artificial intelligence by releasing a new “superchip,” a quantum computing service, and a set of tools to help create general-purpose humanoid robots, which is every sci-fi fan’s dream come true. We’ll look at what the business is doing and what it might mean.

What Does Nvidia Do?

The “Blackwell” line of AI chips was the main thing the company announced at its annual development conference on Monday. These chips power the very expensive data centers that train cutting-edge AI models like the newest versions of GPT, Claude, and Gemini.

One, the Blackwell B200, is a pretty simple improvement over the company’s previous H100 AI chip. Nvidia said that it would take about 8,000 H100 chips and 15 megawatts of power to train a huge AI model the size of GPT-4. That’s enough power to run about 30,000 average British homes. With the new chips from the company, it would only take 2,000 B200s and 4MW of power to do the same task. That could cause the AI business to use less electricity, or it could soon cause the same amount of electricity to power much bigger AI models.

What’s So Great About A Chip?

The company announced the GB200 “superchip” as the second part of the Blackwell line at the same time as the B200. According to Nvidia, this system has “30x the performance” for server farms that run chatbots like Claude or ChatGPT instead of training them. It does this by putting two B200 chips on a single board along with the company’s Grace CPU.

The company also said that this method would cut energy use by up to 25 times. By cutting down on the time the chips spend talking to each other, putting everything on the same board makes it more efficient. This means that the chips can spend more time working the numbers that make chatbots sing – or at least talk. The company Nvidia, which is worth more than $2tn (£1.6tn), would be happy to give.

Take the GB200 NVL72 from the company as an example. It’s a single server rack with 72 B200 chips set up and nearly two miles of cables connecting them. Is that not enough? Check out the DGX Superpod instead. It takes eight of those racks and puts them together to make a shipping container-sized AI data center. The event didn’t say how much it costs, but it’s safe to say that if you have to ask, you can’t afford it. Even the most recent chips cost around $100,000 each.

NVidia Superchip

What Does It Mean For Robots?

Nvidia’s Project GR00T is a new base model for controlling humanoid robots. It was named after Marvel’s arboriform alien, though it’s not directly related to it. You can build special use cases on top of a foundation model, like GPT-4 for text or StableDiffusion for image generation. While they cost the most to make, they are the driving force behind all future progress because they can be “fine-tuned” to fit specific needs in the future.

By watching humans move and understand natural language, Nvidia’s foundation model for robots will help them “quickly learn coordination, dexterity, and other skills in order to navigate, adapt, and interact with the real world.”

GR00T works with Jetson Thor, a system-on-a-chip made to be the brains of a robot. It pairs with another piece of Nvidia technology and makes another Marvel connection. The ultimate goal is to have a machine that can work on its own and be told to do general tasks using normal human speech, even if it hasn’t been taught for those tasks specifically.

A quantum?

Quantum cloud computing is one of the few hot areas where Nvidia doesn’t have a leg to stand on. The technology is still very new, but Microsoft and Amazon have already used it in their products, and now Nvidia is joining the fun.

There will not be a connection between Nvidia’s cloud and a quantum computer, though. Instead, the company is providing a service that uses its AI chips to mimic a quantum computer. This way, experts can test their ideas without having to pay for access to the rare and pricey real one. Over time, though, Nvidia said the platform would let users connect to quantum computers owned by other companies.

Every day people will probably name OpenAI as the company at the center of the current AI craze. It has amazing public tools, grows very quickly, and has a CEO who travels the world. But if you ask someone who knows a lot about tech, they’ll say Nvidia.

Since its start in 1993, Nvidia has made chips that have been at the heart of every new tech development. The company made a risky early bet on machine learning and is now the biggest maker of the chips that are powering the AI boom. Its stock price is skyrocketing and its market cap is over $2 trillion (yes, that’s trillion with a “t”). The company is holding an event this week called “the Woodstock of AI” to show off much-anticipated new technology.

Jensen Huang Just Cemented Himself As One Of The Greatest Leaders

Nvidia has been run by the same person since the beginning, co-founder Jensen Huang. This is different from many other tech giants. Huang is an immigrant from Taiwan who had a hard start in life. He is now worth about $80 billion because he has led his company to the cutting edge of new technology.

How has he stayed so successful for so long in a field that changes so quickly and is so competitive? He recently spoke twice at Stanford, where he went to school, to talk about his leadership style and Nvidia’s plans for the future of AI. He said some great things at the Stanford Institute for Economic Policy Research and on the Business School’s View From the Top series. Both are well worth watching in full. However, just five words from the two-hour talks summed up Huang’s ideas and teach a lesson in leadership that almost anyone can use.

Shantam Jain of View From the Top asks Huang to talk about how he thinks about designing his business and how others should think about designing theirs. Huang starts his answer with five simple words: “No task is beneath me.” This seems very humble coming from a CEO who is worth a billion dollars at first glance. Yes, it is. Huang then makes fun of his early low-level jobs and how they changed the way he thinks. It was my job to clean the toilets. I scrubbed a lot of loos. “I’ve cleaned more toilets than all of you put together, and some of them you just can’t forget,” he says, making people laugh.

Being humble is a skill that many people, including psychologists, Jeff Bezos, and Wharton professor Adam Grant, have said is important for future leaders to have. These different thought leaders say that being humble helps us be open to new knowledge, connect with others, and make better choices.

Huang is being humble just because he wants to be humble. Aside from fighting pride and laziness, he says that his “no task is beneath me” motto is about more than that. It’s also about giving the people who work with him power.

About Being Humble And Having Power

Huang goes into more detail about this in his speech to the Stanford Institute for Economic Policy Research. He has, surprisingly for a CEO, more than 50 direct reports, whom he controls without being involved. These reports don’t get written reviews or regular one-on-ones with their boss. They’re mostly left alone unless they need help with something. Huang says, “Then I’ll drop everything for them.” Why? Huang says: “In that way, our company was designed for agility, for information to flow as quickly as possible, for people to be empowered.”

Huang will drop everything to help his team with anything, and it’s not just because he’s a nice person. He also does it because he thinks his main job is to teach his workers how to do their jobs so they can be free to act quickly and on their own.

“Our job at the company should depend on how well we can solve hard problems, motivate others to do great things, and help, inspire, and empower others.” There is a management team to help all the other people who work for the company for these reasons, Huang says in View From the Top.