NVIDIA Research, the company’s research organization established in 2006 and led since 2009 by Bill Dally, is presenting its work at NVIDIA GTC this week in San Jose, California. The group says its focus is on projects with a high “risk horizon” that can still affect NVIDIA products and the wider industry.
“We make a deliberate effort to do great research while being relevant to the company,” said Dally, chief scientist and senior vice president of NVIDIA Research. “It’s easy to do one or the other. It’s hard to do both.”
David Luebke, vice president of graphics research and NVIDIA’s first researcher, said: “Our mission is to do the right thing for the company. It’s not about building a trophy case of best paper awards or a museum of famous researchers.”
The research team works closely with product teams and industry stakeholders under NVIDIA’s “one team” approach. Bryan Catanzaro, vice president of applied deep learning research at NVIDIA, said: “Everybody at NVIDIA is incentivized to figure out how to work together because the accelerated computing work that NVIDIA does requires full-stack optimization.”
NVIDIA Research says its work has helped shape several major technologies. CUDA, launched in 2006, made it easier to use GPU acceleration for scientific simulations, gaming applications and AI models. Ray tracing research later led to the launch of NVIDIA RTX, including RT Cores, and NVIDIA DLSS, or Deep Learning Super Sampling, which uses an AI pipeline to create high-resolution images from a fraction of the pixels.
In AI software, NVIDIA cuDNN was developed as a research project and released as a product in 2014. NVIDIA StyleGAN followed as a visual generative AI model that produced photorealistic imagery. The group also introduced models such as GauGAN, which developed into the NVIDIA Canvas application, and more recent 3D generative AI work like 3DGUT.
For large language models, Megatron-LM enabled efficient training and inference of massive LLMs for content generation, translation and conversational AI, and is integrated into the NVIDIA NeMo platform.
Beyond AI and graphics, the group has worked on chip architecture, electronic design automation, programming systems, quantum computing and networking. A 2012 research proposal led to NVIDIA NVLink and NVSwitch, while 2013 chip-to-chip link research helped create the connection between the NVIDIA Grace CPU and NVIDIA Hopper GPU.
In 2021, the ASIC and VLSI Research group developed VS-Quant, a software-hardware codesign technique for AI accelerators that enabled many machine learning models to run with 4-bit weights and 4-bit activations at high accuracy. The work influenced FP4 precision support in the NVIDIA Blackwell architecture.
This year at CES, NVIDIA also unveiled NVIDIA Cosmos, a platform created by NVIDIA Research to accelerate the development of physical AI for next-generation robots and autonomous vehicles.
Source: blogs.nvidia.com.
Companies can share verified announcements through Newz9’s international press release submission page.
