Nvidia's CEO Jensen Huang introduced the new Blackwell graphics processing unit (GPU) at its annual GTC conference in San Jose, Calif., stating it would drive the next era of computing.
During the presentation, he also highlighted the Nvidia Quantum-X800 and Spectrum-X800 communication technologies, which were developed in Israel.
Globes reported this year that Nvidia's second largest development center outside of America is in Israel, with some 3,300 employees.
The Blackwell GPU, combined with the new NVLink technology – also developed in Israel – is making waves in AI infrastructure. The company said it will enable organizations to build and run real-time generative AI on trillion-parameter large language models at up to 25 times less cost and energy consumption than its predecessor.
The Blackwell GPU architecture features six transformative technologies for accelerated computing, which will help unlock breakthroughs in generative AI, data processing, engineering simulation, electronic design automation, and computer-aided drug design—all emerging industry opportunities for Nvidia.
"For three decades, we've pursued accelerated computing, with the goal of enabling transformative breakthroughs like deep learning and AI," Huang said at GTC. "Generative AI is the defining technology of our time. Blackwell is the engine to power this new industrial revolution. Working with the most dynamic companies in the world, we will realize the promise of AI for every industry."
Among the many organizations expected to adopt Blackwell are Amazon Web Services, Dell Technologies, Google, Meta, Microsoft, OpenAI, Oracle, Tesla and xAI. The company said Blackwell-based products will be available from partners later this year.
What is Blackwell?
The Blackwell, which the company claims is the world's most powerful chip, replaces Nvidia's Hopper (H100 and H200) GPUs, which have become the go-to GPUs for AI applications. The Blackwell will be available as a standalone GPU or two Blackwell GPUs can be combined and paired with Nvidia's Grace central processing unit to create what it calls its GB200 Superchip, offering up to a 30 times performance increase to the Nvidia H100 GPU for large language model interference workloads, according to the company.
The system is named in honor of David Harold Blackwell, a mathematician who specialized in game theory and statistics. Blackwell was the first Black scholar inducted into the National Academy of Sciences, the company said.
Ian Buck, general manager and vice president of Accelerated Computing at Nvidia, explained during a briefing ahead of the GTC that "today, a new kind of AI is emerging ... one that is even more intelligent, one that has been built not just on a single AI model, but a collection of AI models, called a mixture of expert models… These new models take multiple AI models and have them operating in concert.
"Now, it is not just the query or the question we ask the chatbot, but the additional information that is being fed from the database from your histories, your search engines that go into the model to create that great answer," Buck continued. "We realized we needed to build a new kind of computer, not one that just took chips from Nvidia and put them in PC cards, but one that allowed multiple GPUs to talk together across the fabric called NVLink, to build a bigger GPU, a more capable system."
Nvidia's DGX SuperPOD supercomputer system, also released at GTC, comprises eight or more DGX GB200 systems. These systems feature 36 GB200 Superchips working together as a unified computer. The company said customers can expand the SuperPOD to accommodate tens of thousands of GB200 Superchips per their requirements.
"NVIDIA DGX AI supercomputers are the factories of the AI industrial revolution," Huang said in a statement. "The new DGX SuperPOD combines the latest advancements in Nvidia accelerated computing, networking and software to enable every company, industry and country to refine and generate their own AI."
At GTC, Nvidia also debuted its Nvidia NIM generative AI microservices to empower developers to craft and integrate customized AI applications while retaining their intellectual property rights. They also unveiled healthcare-focused models and microservices to aid drug discovery and advance digital transformation in the medical field. Moreover, a collaboration between NVIDIA and Johnson & Johnson MedTech was announced, focusing on integrating generative AI technology into operating rooms.
The company announced Project GRooT, a multimodal AI that will power humanoids in the future. This includes the Jetson Thor robotics computer with a GPU based on the Blackwell architecture and the DRIVE platform to better connect the automotive industry to the creative AI era.
"The world is built for humans," said Rev Lebaredian, vice president of Omniverse and simulation technology at Nvidia, during the pre-GTC briefing. "General purpose AI for the physical world will be humanoid." Project GRooT includes the world's first human foundation model, which takes language, video, and human demonstration as inputs and combines them with past experiences to generate its next actions."
Nvidia also announced Isaac Manipulator, a platform with foundation models and acceleration libraries to help the world's robot arms become faster and more accurate, as well as Isaac Perceptor, a platform of acceleration libraries and frameworks to help autonomous mobile robots become smarter through 3D-surround, visually based perception.
"These smarter, faster, better robots will be deployed in the world's heavy industries," Lebaredian said.
Finally, Nvidia announced the expansion of its Omniverse platform for developing and deploying physically based industrial digitization applications – "the feedback loop for AIs to enter the physical world," explained Lebaredian – across the world's industrial ecosystems. The company announced Omniverse cloud APIs that let developers integrate core Omniverse technologies into their existing apps and workflows.
Omniverse will allow organizations with a focus on industry to use AI and physics-based 3D simulations to plan before integrating robots into a facility. This ensures that every possible scenario is explored before actual construction begins.
Nvidia, a full-stack computing infrastructure company, was founded in 1993 and opened its Israel office in 2016.
In December, Nvidia and its employees donated $15 million to Israeli and foreign NGOs supporting civilians affected by the October 7 Hamas massacre and ongoing Gaza war.