[ad_1]
Nvidia Corp. Chief Executive Jensen Huang detailed Tuesday how the chip maker expects to accelerate the development of artificial intelligence and next-generation chip-making, introducing software meant to allow foundries to etch smaller transistors faster and a bevy of artificial-intelligence products for businesses.
In the keynote address at Nvidia’s
NVDA,
annual GTC developer conference, Huang introduced the company’s cuLitho software library that improves upon computational lithography, the process in which algorithms are used to improve the resolution of the ever-shrinking transistors that are etched onto silicon wafers. The “cu” part stands for Nvidia’s CUDA parallel computing and app platform.
Nvidia said that third-party foundry Taiwan Semiconductor Manufacturing Co.
TSM,
and electronic design company Synopsys Inc.
SNPS,
will be using the library, and that chip-equipment maker ASML
ASML,
will work with Nvidia to integrate support for graphics-processing units on their lithography software.
“The chip industry is the foundation of nearly every other industry in the world,” Huang said in a statement. “With lithography at the limits of physics, Nvidia’s introduction of cuLitho and collaboration with our partners TSMC, ASML and Synopsys allows fabs to increase throughput, reduce their carbon footprint and set the foundation for 2nm and beyond.”
On a media call, cuLitho architect Vivek Singh could not comment on how Nvidia is pricing the software or on business models, but he noted that the software helps chip makers where it matters.
“We have conducted … various total cost of ownership analyses, and on every single vector that is important — cost, productivity, output, power, space — it is hugely advantageous to use cuLitho,” Singh said.
The software allows 500 of Nvidia’s DGX H100 GPU-based data-center systems to do the same work of 40,000 CPU-based systems, the company said. That results in reduced power and space needs, as well as reducing the environmental impact, and processes that once took weeks can now be handled overnight, Nvidia said.
“The cuLitho team has made admirable progress on speeding up computational lithography by moving expensive operations to GPU,” said C.C. Wei, TSMC’s chief executive, in a statement provided by Nvidia. “This development opens up new possibilities for TSMC to deploy lithography solutions like inverse lithography technology and deep learning more broadly in chip manufacturing, making important contributions to the continuation of semiconductor scaling.”
Similarly, ASML Chief Executive Peter Wennink said the software “should result in tremendous benefit to computational lithography, and therefore to semiconductor scaling.”
Back in September when Nvidia introduced its new “Ada Lovelace” chip architecture, Huang announced that Moore’s Law — the standard that the number of transistors on a chip doubles every two years for around the same cost — was “dead” in that silicon wafers were “a ton more expensive” now. Over the years, Nvidia has developed such an entrenched software ecosystem for its chips that it has prompted some analysts to start looking at Nvidia as a quickly emerging software company.
AI rollouts
The company also released new artificial-intelligence services in its move to democratize AI and become a key leader in its development.
-
Nvidia said it was launching four new platforms for developers to build specialized artificial intelligence models: the L4 for AI video, the L40 for image generation, the H100 NVL for large language model deployment such as ChatGPT, and Grace Hopper for recommendation models. Nvidia said Alphabet Inc.’s
GOOG,
+3.15%
GOOGL,
+3.08%
Google Cloud Platform is an early adopter of the L4 and was integrating it into its Vertex AI machine-learning platform, making it Nvidia’s “biggest collaboration” in AI, Huang said in a meeting with analysts. Grace Hopper and the H100 NVL will be available in the second half of the year, while the L40 is available now, and the L4 is available in a private preview from Google.
Read: Google opens up access to Bard AI chatbot, its rival to ChatGPT
- Nvidia launched DGX Cloud, a service where businesses can get instant access to artificial intelligence models over a simple browser for $37,000 a month, calling it an “iPhone moment of AI.”
- The company also launched its Foundations model-making service that can handle language, images, video and 3-D, with its NeMo and Picasso services.
[ad_2]
Source link
