Late Tuesday, Microsoft Azure shared an interesting image on its social media platform
Microsoft is one of Nvidia's largest customers, and the company often receives products first time around and integrates them into cloud computing and corporate infrastructure. Even Nvidia will listen to feedback from companies like Microsoft when designing future products, especially products like the now-cancelled NVL36x2 system. The image below shows a massive cluster that roughly divides the compute area into one-third of the entire system and devotes two-thirds of the system to closed-loop liquid cooling.
The entire system uses Infiniband networking, which is the standard for GPU-accelerated systems due to its low packet transfer latency. While the details of the system are a little unclear, we can see that an integrated closed-loop liquid cooling system allows the GPU rack to come in a 1U form factor for increased density.
Given that these systems will go into the wider Microsoft Azure data center, the systems need to be easy to maintain and cool. Microsoft data centers are limited in the power and heat output they can handle, so these types of systems are typically designed to meet internal specifications designed by Microsoft. Of course, there are more compute-dense systems, such as NVIDIA's NVL72, but hyperscalers should generally choose other custom solutions that fit their data center specifications.
Finally, Microsoft noted that we can expect to see more details and learn more about the GB200-powered AI system at the upcoming Microsoft Ignite conference in November.