The new NVIDIA GB200 NVL72 system sets new standards for AI system design by connecting 72 Blackwell GPUs and 36 Grace CPUs.
NVIDIA Grace Blackwell Compute Tray
NVIDIA is set to reveal its latest innovations at the upcoming Hot Chips 2024 conference, where they will present its advancements in data center performance and energy efficiency. As the trillion-dollar data center computing market continues to evolve, NVIDIA’s engineers will showcase developments in the NVIDIA Blackwell platform, liquid cooling solutions, and AI agents for chip design.
During the conference, NVIDIA will highlight how the NVIDIA Blackwell platform brings together multiple chips, including the Blackwell GPU, Grace CPU, BlueField DPU, and CUDA software, to power the next generation of AI across industries. According to NVIDIA, the new GB200 NVL72 system, a multi-node liquid-cooled rack-scale solution, sets new standards for AI system design by connecting 72 Blackwell GPUs and 36 Grace CPUs. This system will enable unprecedented low-latency inference for generative AI workloads, with up to 30x faster inference for large language models (LLMs).
“Be Ready For Blackwell”
“NVIDIA Blackwell is the ultimate full-stack computing challenge,” NVIDIA’s Dave Salvator said in a blog post today. “By uniting these advanced technologies, we’re redefining the standard for AI and accelerated computing performance while dramatically enhancing energy efficiency.”
NVIDIA Quantum InfiniBand Switch
Beyond chip innovation, NVIDIA also focuses on cooling technologies that could redefine data centers. Liquid cooling, which is becoming a key element in supporting large-scale AI workloads, will be a focal point of the discussions. Ali Heydari, Director of Data Center Cooling and Infrastructure at NVIDIA, will showcase hybrid liquid-cooling solutions that can be retrofitted into existing air-cooled data centers.
NVIDIA Spectrum Network 800Gbit Switch
In addition to these hardware advancements, NVIDIA is pushing the boundaries of AI-powered design. Mark Ren, Director of Design Automation Research at NVIDIA, will provide insights into how AI models and agent-based systems transform semiconductor design. These AI tools enhance engineers’ productivity by automating tasks like cell cluster optimization and debugging, making chip design more efficient. “AI agents are unlocking new possibilities by autonomously performing complex tasks that previously required manual intervention,” said Ren.
NVIDIA NVLink Switch
Hot Chips 2024 will take place from August 25-27 at Stanford University and online. Attendees can expect to gain deeper insights into NVIDIA’s latest breakthroughs and how they shape the future of AI and data center technology.
For more information about NVIDIA Blackwell announcements at Hot Chips 2024, visit NVIDIA’s official website.
Engage with StorageReview
Newsletter | YouTube | Podcast iTunes/Spotify | Instagram | Twitter | TikTok | RSS Feed
