A US-based organization has developed a model that may operate humanoid robots to take positions of leadership in AI along with other fields. The chip manufacturer Nvidia has announced a new “super chip,” a quantum computation service, and a new set of tools to assist in the development of general-purpose humanoid robotics—the ultimate science fiction goal. With these announcements, Nvidia has expanded its leadership position in artificial intelligence. Here, we examine the corporation’s actions and their potential implications.
How does Nvidia operate?
The “Blackwell” range of Intelligence chips, which are employed for running the astronomically expensive data centers that train cutting-edge artificial intelligence models like the most current versions of GPT, Claude, and Gemini, are the primary focus of the organization’s annual modern technology summit on Monday.
One is a relatively simple improvement from the company’s previous H100 AI processor, the Blackwell B200. According to Nvidia, training a gigantic AI model with the capacity of GPT-4 would currently require over 8,000 H100 chips and 15 megawatts of power, which is equivalent to powering roughly 30,000 average British houses.
The identical practice exercise only needs 2,000 B200s and 4 megawatts of power using the company’s new chips. It might result in the artificial intelligence industry using less electricity overall, or it might mean that in a short while, the same electricity will be utilized for operating far larger AI models.
How does a chip become “super”?
The business also revealed the GB200 “super chip,” the second component of the Blackwell line, in addition to the B200. To create a system that, as claimed by Nvidia, gives “30x the performance” to hosting clusters that operate instead of teach chatbots like Claude or ChatGPT, it combines two B200 chips in a single board with the business’s Grace CPU. According to the company, that method also claims to save energy use by as much as twenty-five times.
Having everything on one board increases efficiency and frees up more processing power for calculating the figures that enable chatbots to sing—or, perhaps, converse. By reducing the duration of the chat, the chips can chat more efficiently.
What if I demand more?
With a market capitalization of almost $2 trillion (£1.6 trillion), Nvidia would certainly be thrilled to offer its products. Consider the GB200 NVL72 server rack, which has 72 B200 chips configured and is wired together with over two miles of wiring. Is that insufficient? Why not have a glimpse of the DGX Superpod, which integrates 8 such workstations into a single Intelligence datacenter in a box approximately the dimensions of a container for shipment? Although the cost was not made clear at the moment, it is fair to assume that you are unable to afford it if you need to ask. Even the last version of chips cost a whopping $100,000 or more per unit.
My robots, what about them?
Project GR00T is an innovative base model from Nvidia designed for directing humanoid robots. It is allegedly named after the arboriform alien from Marvel, albeit it is not directly related to the comic book character. The basic AI model that is used to build specific use cases is called a foundational model, like GPT-4 for content or StableDiffusion for visual production. Although these products are the most costly to develop in the entire industry, they are the driving forces for all future innovation because they may be “fine-tuned” to specific scenarios in the future.
Thanks to Nvidia’s foundation model, robots will be able to mimic human movements and understand human speech. This will enable them to learn collaboration, flexibility, and other skills fast. These abilities will be necessary for them to operate, adjust, and communicate with the real environment.
GR00T is paired with Jetson Thor, a system-on-a-chip made particularly to function as a robot’s brain, another piece of Nvidia technology (and another Marvel association). The principal objective is to create a self-driving device that can be programmed to carry out wide activities, such as activities for which it could be more instructed through ordinary words spoken by humans.
A quantum?
Quantum cloud computing is one of the few highly competitive fields in which Nvidia lacks a foothold. The technology is still relatively new, but Microsoft and Amazon have already used it in their products, and now Nvidia is joining in on the action.
However, Nvidia’s cloud will not be linked to a quantum computer. Rather, the company is providing an application that employs artificial intelligence chips to mimic a quantum computer. This way, experts can evaluate their concepts without having to pay for access to the unique and expensive real ones. Eventually, however, Nvidia claimed the system would allow users to connect to quantum computers owned by other companies.
