Chip manufacturing is the “ideal application” for NVIDIA accelerators and AI computing. explained CEO Jensen Huang during a speech at the ITF World 2023 industry conference in Antwerp, Belgium.
For nearly 40 years, exponential growth in CPU power has fueled the entire tech industry, but in recent years the segment has been stretched and slowing as demand for processing power continues to grow. As a result, the energy consumption of data centers has risen sharply. Huang pointed out that NVIDIA helped overcome the crisis by combining the parallel processing capabilities of GPUs with the capabilities of CPUs, becoming the pioneer in accelerated computing. The success was fueled by the work of machine learning researchers who discovered that GPUs can deliver supercomputer-level performance with low power consumption. By optimizing algorithms, NVIDIA hardware helps accelerate applications by 10x to 100x while reducing cost and power consumption by an order of magnitude. As a result, the fields of artificial intelligence and accelerated computing are becoming the determining directions for the development of the technology industry.
Manufacturing advanced chips involves more than a thousand steps that reduce electronic components to the size of a biomolecule, and each step must be completed with near-perfect results. NVIDIA technologies are deployed in different phases, and in March the company announced a joint project with TSMC, ASML and Synopsys in the field of computational lithography, a task that requires the most computational resources throughout the chip design and manufacturing cycle. library NVIDIA cuLitho made it possible to speed up data processing by a factor of fifty, replace tens of thousands of servers with several hundred NVIDIA DGX systems, and thus reduce energy consumption and financial costs by an order of magnitude.
Mr. Huang also spoke about promising AI systems that users can understand, reason and even interact with the physical world – these are robotics, autopilot vehicles and more advanced chatbots. The company also developed its own AI project called NVIDIA VIMA. In particular, it allows you to perform operations on graphical objects according to a text description, while working on the NVIDIA Omniverse 3D modeling platform. Another project is NVIDIA Earth-2 is a digital twin of the Earth whose developers created the FourCastNet AI model that emulates weather conditions up to 100,000 times faster than existing analogues. Finally, the company’s technology helped scientists at the UK Atomic Energy Agency and the University of Manchester build a fusion reactor emulator to simulate plasma physics – here you can test all hypotheses before starting processes in a real reactor.