As the world of high-performance computing is getting together this week in Atlanta for SC24, Nvidia Corp. held a “special address” to showcase the company’s vision for computing and artificial intelligence, with founder and Chief Executive Jensen Huang leading the charge.
Joined by Ian Buck, Nvidia’s vice president and general manager for hyperscale and high-performance computing, Huang (pictured) underscored Nvidia’s central role in transforming scientific research and industries through accelerated computing. The presentation also featured insights from Arvind Ramanathan of Argonne National Laboratory and David Keyes of King Abdullah University of Science and Technology, adding depth to the discussion while emphasizing real-world applications.
Nvidia’s journey in supercomputing
Huang opened the address by discussing the societal importance of supercomputing. “Supercomputers are among humanity’s most vital instruments, driving scientific breakthroughs and expanding the frontiers of knowledge,” he said, which also framed Nvidia’s journey in supercomputing.
Looking back to the launch of CUDA in 2006, he highlighted how Nvidia reshaped computational science, drastically reducing computing costs by a million-fold. He described CUDA as “a computational microscope, a telescope, and a time machine” that empowers researchers to conduct work previously deemed impossible.
Huang cited CUDAx libraries as the keys to this transformation, explaining that they act as “the engines of accelerated computing” in domains such as healthcare, telecommunications and manufacturing.
Huang said these tools have broadened the applications of Nvidia’s hardware and fueled a virtuous cycle of adoption, partnerships and developer engagement. This foundation underscores a broader industry trend: Platforms matter as much as hardware. Companies competing in the HPC market must invest in comprehensive ecosystems that enable seamless integration across applications.
The platform approach has created a massive moat around Nvidia that should enable the company to remain the de facto standard in accelerated computing, including supercomputing and AI. The company makes excellent graphics processing units, but CUDA dramatically simplifies the process of building applications and services that leverage the underlying Nvidia technology. Many industry watchers are looking for one of the other silicon manufacturers to take a bite out of Nvidia, but CUDA keeps the company out front.
Pioneering the AI computing era with Blackwell
As expected, Huang discussed Nvidia’s Blackwell system, the company’s latest AI platform designed to meet the escalating demands of training and inference. He described it as “a giant virtual GPU with 1.44 AI exaFLOPs” that integrates multiple chip types into a liquid-cooled, energy-efficient system. With production facilities underway in the U.S., Mexico and Taiwan, Blackwell exemplifies what Huang called Nvidia’s “extreme co-design philosophy,” where every component is optimized to deliver peak performance.
Huang didn’t hold back from discussing the implications of what Blackwell can enable. “AI will soon power humanoid robots capable of adapting and performing various tasks with minimal demonstration,” he predicted. He added that these developments will reshape industries like manufacturing and logistics, fueling productivity growth and job creation.
Nvidia’s vision reflects a broader shift in computing. AI is not just a tool but a transformative force. By scaling systems like Blackwell, Nvidia positions itself as a cornerstone of this transformation, pushing traditional CPU vendors to reconsider their strategies.
I believe the AI industry is poised for a massive shift. During this time, technology will become ubiquitous and embedded into almost everything we use. In the early 1990s, Cisco predicted that the internet would change the way we “work, live, learn and play. “ AI will have a similar but bigger impact, and Nvidia will be the foundation of that.
Accelerating science with AI
“AI has arrived, and a new computing era has begun,” Huang declared as he turned to integrating AI into scientific workflows. He highlighted applications ranging from drug discovery to climate modeling, where AI accelerates simulations and real-time data analysis. Nvidia’s newly announced CuPyNumeric, a GPU-accelerated implementation of NumPy, enables researchers to scale Python-based projects across GPU clusters without rewriting code.
Buck emphasized its impact, sharing that the LCLS X-ray laser team achieved “a sixfold speedup in data analysis, reducing workflow times from years to months.”
Buck introduced Arvind Ramanathan of Argonne National Laboratory, who illustrated how this technology transforms protein design. Ramanathan’s team used Nvidia’s architectures to train multimodal AI models that integrate natural language and protein sequences.
“We achieved nearly three exaFLOPs of mixed precision runs,” he said, calling the performance “unbelievable” in its ability to accelerate discovery. These advancements make AI indispensable in science, democratizing access to HPC tools and enabling breakthroughs at scales previously unimaginable.
Building a sustainable computing future
Energy efficiency emerged as a recurring theme. Buck pointed out that while GPUs consume more power per server, they significantly reduce the total energy required to complete computations. He cited examples like the Texas Advanced Computing Center achieving “a 110x speedup and 6x better energy efficiency” using Nvidia’s Grace Hopper architecture.
Huang also tied energy efficiency to broader challenges, emphasizing that AI-driven systems like Blackwell will help industries adapt to global problems such as climate change. This focus aligns with the growing demand for sustainable computing practices as energy costs rise and environmental concerns intensify.
One of the frustrations points I’ve had recently is that industry watchers call out AI as conflicting with sustainability. Though it’s true that AI infrastructure is power-hungry, it’s important to understand the alternative. One ChatGPT query might use significantly more power than a search but might return an answer faster.
For example, the PGA TOUR uses generative AI to compile information on players. Without it, it requires loading media guides, searching videos, and other tasks that can take days or weeks to complete compared to the few seconds it takes AI. If used correctly, next-gen infrastructure can be more sustainable than legacy technology.
Collaboration and innovation driving HPC
Buck noted key milestones from the broader HPC community, including the Test of Time Award honoring Ari Kaufman’s 2004 paper on GPU clusters. He said Kaufman’s work laid the foundation for today’s accelerated computing, which now powers more than 70% of the top 100 fastest supercomputers.
David Keyes of King Abdullah University of Science and Technology added context with genomics and climate modeling examples. His team scaled genomic studies to simulate 13 million synthetic patients, enabling countries to conduct population-wide analyses.
“We democratized genome-wide association studies,” he said, emphasizing the potential for intelligent health and agriculture applications.
He highlighted advancements in climate emulation, which reduced data processing from 100-kilometer to 3.5-kilometer resolution.
“We brought the queryable distance down dramatically,” he said, demonstrating the practical benefits of Nvidia’s innovations. These achievements underscore the importance of partnerships between technology providers and researchers.
Some final thoughts
Nvidia’s SC24 Special Address offered a compelling vision of how accelerated computing and AI will shape the future. As the industry navigates this new era, collaboration, transparency and sustainability will be critical to harnessing the transformative power of AI. Nvidia’s leadership in this space sets a high bar, challenging others to innovate and adapt in a rapidly evolving landscape.
As Huang put it: “Let’s build the future together.”
