Originally published at: https://developer.nvidia.com/blog/nvidia-contributes-nvidia-gb200-nvl72-designs-to-open-compute-project/
During the 2024 OCP Global Summit, NVIDIA announced that it has contributed the NVIDIA GB200 NVL72 rack and compute and switch tray liquid cooled designs to the Open Compute Project (OCP). This post provides details about this contribution and explains how it increases the utility of current design standards to meet the high compute density…
【 Figure 2. NVIDIA GB200 NVL72 NVLink domain delivering an aggregate AllReduce bandwidth of 260 TB/s】，on this picture ，1 NVSwitch offer 144 port ，maybe 9 NVSwitch is enough
I am following your NVL72 very closely!
I thought your rack thermal design is too much limited, as the following picture shows. With the repeated rumers of overheating, I am more convinced its thermal design is limited up to the point that it does not work, or doesn’t work very well, sorry to say that. Otherwise, the way to have it work is put the rack in a very very cold datacenter, that sounds not good!
Have a look of the attached link https://eviden.com/wp-content/uploads/2024/10/Eviden-brochure-BullSequanaXH3000-HPC.pdf, I believe this is the only solution that can host your NVL72!
