Ted Hisokawa
Aug 23, 2024 14:53
NVIDIA engineers to unveil advancements in the Blackwell platform, liquid cooling, and AI-driven chip design at Hot Chips 2024.
At the upcoming Hot Chips 2024 conference, NVIDIA engineers are set to present groundbreaking innovations aimed at enhancing data center performance and energy efficiency. The event, scheduled for August 25-27 at Stanford University, will feature four talks by senior NVIDIA engineers, according to NVIDIA Blog.
NVIDIA Blackwell Platform
NVIDIA’s presentations will highlight the NVIDIA Blackwell platform, which integrates multiple chips, systems, and NVIDIA CUDA software to drive the next generation of AI applications. The NVIDIA GB200 NVL72, a multi-node, liquid-cooled solution, will also be showcased for its ability to connect 72 Blackwell GPUs and 36 Grace CPUs, setting a new benchmark for AI system design. Additionally, the NVLink interconnect technology will be discussed for its high throughput and low-latency capabilities, crucial for generative AI.
Liquid Cooling Advancements
A significant portion of the conference will focus on liquid cooling technologies. Ali Heydari, director of data center cooling and infrastructure at NVIDIA, will present designs for hybrid-cooled data centers. These designs aim to retrofit existing air-cooled data centers with liquid-cooling units, offering a more efficient and space-saving solution. Heydari’s team is also collaborating with the U.S. Department of Energy on the COOLERCHIPS program to develop advanced cooling technologies. Using the NVIDIA Omniverse platform, they are creating digital twins to model energy consumption and cooling efficiency.
AI-Driven Chip Design
NVIDIA is also leveraging AI to enhance semiconductor design. Mark Ren, director of design automation research at NVIDIA, will discuss how AI models improve design quality and productivity by automating time-consuming tasks. These models include prediction and optimization tools that help engineers analyze and improve designs rapidly. AI agents powered by large language models (LLMs) are being developed to autonomously complete tasks, interact with designers, and learn from a database of human and agent experiences. Ren will share examples of AI applications in timing report analysis, cell cluster optimization, and code generation.
These presentations underscore NVIDIA’s commitment to pushing the boundaries of data center computing through innovative technologies in AI and cooling solutions. The company’s efforts aim to deliver unparalleled performance, efficiency, and optimization across various sectors.
Image source: Shutterstock