Nvidia Gb200 Grace Blackwell Superchip © Nvidia
In one year, the value of Nvidia shares increased by more than 240% thanks to artificial intelligence. Indeed, the company markets the famous H100 chips that digital giants and artificial intelligence laboratories are competing to develop their models. And yet, it was only the beginning. During the GTC 2024 conference, the company unveiled a new architecture called Blackwell, which will lead to new breakthroughs in the field of AI.
Based on TSMC's 4NP process, Backwell GPUs feature 208 million transistors, while the old architecture only offered GPUs with 80 million transistors. With this power gain, the new architecture is optimized to train and operate AI models with trillions of parameters, while reducing costs and energy by up to 25 times.< /p>
According to explanations from The Verge, training an AI model with 1 800 billion parameters would previously have required 8 000 GPUs of the old architecture Hopper, and 15 megawatts of power. On the Blackwell architecture, training the same model would be possible with only 2,000 GPUs, while consuming only 4 megawatts. Nvidia also offers a package called GB200 NVL72, for the heaviest tasks. “It combines 36 Grace Blackwell superchips, including 72 Blackwell GPUs and 36 Grace processors interconnected by fifth generation NVLink”, explains the company.
In other words, Nvidia's Blackwell GPUs will be the new essential components in the race for artificial intelligence. Among the companies that will use these new chips are Amazon Web Services, Dell Technologies, Google, Meta, Microsoft, OpenAI, Oracle, Tesla and xAI. “Blackwell delivers huge advancements in performance and will accelerate our ability to deliver cutting-edge models. We are excited to continue working with NVIDIA to improve AI computing,” said Sam Altman, CEO of OpenAI.
“The transformative potential of AI is incredible and will help us solve some of the world’s most important scientific problems. Blackwell's revolutionary technological capabilities will provide the critical computing needed to help the world's brightest minds plot new scientific discoveries,” said Demis Hassabis, co-founder and CEO of Google DeepMind.
Regarding pricing, Nvidia CEO Jensen Huang revealed to CNBC that a Blackwell GPU will cost between $30,000 and $40,000 per unit . He also recalled that Nvidia had to invent new technologies for this chip, and spend nearly $10 billion on research and development.
📍 To not miss any news from Presse-citron, follow us on Google News and WhatsApp.
[ ]
Sitting on a small bench, former Colombian general Henry Torres takes a moment of respite…
Photo: Fabien Deglise Le Devoir In 17 Oregon counties, signs have sprouted in recent years…
Photo: Fabien Deglise Le Devoir At a Lakeview gun show, Bo Gordon, whose county also…
© Tinder The number sends shivers down your spine. In 2020, the Federal Trade Commission…
© Wikimedia Commons Amazon is one of the world's largest companies: from e-commerce to the…
“Euromillions (FDJ) result: the draw for this Tuesday, May 14, 2024, makes a winner in…