NVIDIA CEO Jensen Huang took the stage for a main report at the company’s GTC conference and gave some exciting news about the company’s new technologies.
It announced the GeForce RTX 40 series GPUs, which are based on the company’s new Ada Lovelace architecture. Advances in Ada include a new Streaming Multiprocessor, an RT Core with twice the throughput of ray intersections and triangles, a new Tensor Core, 1.4 petaflops of Tensor CPU power, and the latest version of NVIDIA DLSS technology.
Huang called the launch a “quantum leap” forward for creating fully simulated worlds.
He also announced that the next generation Hopper GPU is in full production and will be ready to ship in a few weeks. Grace Hopper combines Grace’s Arm-based data center processor and Hopper GPUs. This will greatly increase the amount of fast memory and will be ideal for recommender systems. Systems with Grace Hopper will be available in early 2023.
Huang went on to announce the DRIVE Thor, which is a processor that can be used for robotics, medical instruments, industrial automation, and artificial intelligence edge systems. It combines a Hopper transformer, an Ada GPU, and a Grace processor.
Another update is the Nemo LLM Service, a cloud-based service for adapting pre-built LLMs (large language models) for specific tasks. According to Huang, LLMs are the most important AI models to date and can be used for unsupervised understanding of meanings and languages.
“Today we announced new chips, new enhancements to our platforms and, for the first time, new cloud services,” Huang said. “These platforms are enabling new breakthroughs in AI, new applications of AI, and the next wave of AI for science and industry.