From NVIDIA: 2024-08-23 11:00:34

Hot Chips 2024 conference will feature NVIDIA engineers presenting on advanced processor technology, liquid cooling for data centers, and AI agents for chip design. Topics include NVIDIA Blackwell platform, NVLink interconnect technology, and Quasar Quantization System for AI computing. The event will also include tutorials on hybrid liquid-cooling solutions and AI models for processor design.

NVIDIA Blackwell integrates multiple chips and systems to power AI across various industries. NVIDIA GB200 NVL72 is a liquid-cooled, rack-scale solution connecting Blackwell GPUs and Grace CPUs. NVLink technology enables high GPU communication for generative AI. Quasar Quantization System accelerates AI computing. AI models assist in building processors for AI.

At Hot Chips 2024, NVIDIA engineers will showcase new architectural details of Blackwell silicon and examples of generative AI models. The conference will also include tutorials on liquid-cooling solutions for energy-efficient data centers and AI models for processor design. NVIDIA engineers are focusing on innovation in data center computing and design for improved performance and efficiency.

Director of architecture at NVIDIA, Ajay Tirumala and Raymond Wong, will explain how technologies like NVIDIA GB200 NVL72 and Quasar Quantization System work to enhance AI computing performance and energy efficiency. GB200 NVL72 offers faster inference for LLM workloads. The Quasar Quantization System supports high accuracy on low-precision models like LLMs and generative AI.

NVIDIA engineer Ali Heydari will present various hybrid-cooled data center designs that improve efficiency and sustainability. Liquid cooling is more efficient than traditional air cooling and allows for more server racks in data centers. Heydari is involved in the COOLERCHIPS program developing advanced data center cooling technologies using NVIDIA Omniverse platform for optimization.

AI models are aiding engineers in semiconductor design by improving design quality and productivity. NVIDIA expert Mark Ren will discuss the use of AI models in processor design, including agent-based AI systems powered by LLMs. These systems can autonomously complete tasks in microprocessor design, boosting efficiency and productivity in the industry.



Read more at NVIDIA: NVIDIA to Present Innovations at Hot Chips That Boost Data Center Performance and Energy Efficiency