Nvidia has announced a liquid cooled server rack featuring 72 of its newly shared Blackwell GPUs.
The GB200 NVL72 system also includes 36 Grace CPUs and nine NVLink switch trays, each of which has two NVLink switches. That, the company claims, allows for the entire system to act as one giant GPU.
"The fabric of NVLink, the spine, is connecting all those 72 GPUs to deliver an overall performance of 720 petaflops of training, 1.4 exaflops of inference," Nvidia's accelerated computing VP Ian Buck told DCD in a pre-briefing ahead of the company's GTC conference.
"Overall, the NVLink domain can support a model of 27 trillion parameters and 130 terabytes of bandwidth."
The system has two miles of NVLink cabling across 5,000 cables. Buck added that "in order to get all this compute to run that fast, this is a fully liquid cooled design" with 45°C (113°F) coolant going in and 65°C (149°F) coolant coming
"This is what a GPU looks like," Nvidia CEO Jensen Huang said.
The GB200 NVL72 will be available on Amazon Web Services, Google Cloud, Microsoft Azure, and Oracle Cloud.
This article has been updated to reflect the coolant temperatures of the design.
