In the early morning of March 19, Beijing time, Huang Renxun, founder and CEO of NVIDIA, delivered a keynote speech at GTC2024 and released the latest breakthrough achievements in the field of accelerated computing and robotics, of which the biggest explosive point is undoubtedly the launch of the new Blackwell architecture.
EA402 913-402-000-012 Based on the Blackwell architecture, Nvidia also officially released the world’s strongest B200 and GB200 superchips, as well as the GB200 NVL72, a liquid-cooled rack-level system composed of 36 Grace cpus and 72 Blackwell Gpus.
According to Nvidia
With 1440petaflops(aka 1.4exaflops) of inference power, the Nvidia NVL72 rack-scale AI system is capable of supporting large-scale training and inference tasks and can handle large language models with up to 27 trillion parameters. Compared to the same number of EA402 913-402-000-012 H100 Tensor Core graphics processing units for inference purposes, the GB200 NVL72 improves performance by up to 30 times. In addition, the system can reduce costs and energy consumption by up to 25 times.
GB200 NVL72 liquid-cooled rack server
Maximizing performance is not an easy task, and many factors need to be taken into account. To realize the full potential of the Blackwell architecture, Nvidia’s GB200 NVL72 rack-scale server uses a liquid cooling system. In fact, for Blackwell-architecture chips, liquid cooling systems are not only the best option, but almost a prerequisite for operation.
It is roughly estimated that under full load operation, the EA402 913-402-000-012 GB200 includes two 1200W Gpus and 300W Arm cpus, and the possible total power consumption is about 2700W, far beyond the traditional air cooling capacity. However, with the liquid cooling system configuration, the heat output of the chip can reach 1200W even at full load operation, while achieving a strong performance of 20petaflops. By using liquid cooling technology, Nvidia has been able to continue to break the performance limits of AI accelerators without worrying about the limitations of cooling systems, and it is reported that Nvidia plans to apply liquid cooling technology on a full range of subsequent products.
Server liquid cooling technology
Nowadays, in the context of the increasingly complex task of artificial intelligence training load and the rising power consumption of the corresponding hardware, liquid cooling technology has become the core solution to solve the high power dissipation heat dissipation of artificial intelligence server, which can make the core chip play the best performance level.
As a global leader in fast connection solutions, Staubli has more than 30 years of in-depth research and volume deployment experience in liquid cooling technology, focusing on providing safe, reliable and industry-standard liquid cooling connector solutions for server liquid cooling systems. Driven by the increasing demand for advanced artificial intelligence models such as AIGC, with its technical strength and product reliability, Staubli is able to provide strong technical support and guarantee for the accelerated development of liquid cooling systems in the server field, and contribute to the construction of more energy-efficient AI server systems.