Supermicro Begins Shipping NVIDIA Vera Rubin NVL72 Racks
Published2026.09.29 09:52
Charles Liang, President and CEO of Supermicro, stated, "Customers can now order scalable units and receive systems with end-to-end integration completed, ready for immediate deployment in commercial environments," and added, "Supermicro's DLC-2 technology enables us to provide an integrated solution for the NVIDIA Vera Rubin platform."
It consists of 18 1U computing trays and 9 sixth-generation NVLink switch trays, with scale-up bandwidth reaching 216TB/s.
Per rack, it supports 20.7TB of HBM4 memory and up to 54TB of maximum LPDDR5X memory.
Supermicro tests and validates all racks with the complete cooling system integrated, reducing the interval between on-site deployment and operational startup.
Liquid cooling components include △cold plates △manifolds △hose kits △in-row cooling distribution units (CDU) △in-rack CDU △rear door heat exchangers △facility cooling towers.
In-row CDU supports cooling capacity of 1.8MW per CDU and is configured in N+1 redundancy.
On a single scalable unit basis, it supports 1,152 GPUs and 331TB of HBM4 memory across 16 computing racks, with balanced configuration of cooling, power, storage, and networking.
Supermicro performs all stages directly, from node machine design through large-scale manufacturing of liquid-cooled racks to on-site cluster deployment based on NVIDIA reference architecture.
Comments
Related news
Supermicro Announces Air-Cooled AI Server with AMD MI355X GPUs
Supermicro has announced a 10U air-cooled server equipped with AMD's latest Instinct MI355X GPU. The product delivers 288GB of HBM3e memory per GPU and 8TB/s bandwidth, achieving up to 4x improvement in AI compute performance and up to 35x improvement in inference performance compared to the previous generation. With a scalable architecture compatible with both air-cooled and liquid-cooled infrastructure, it is optimized for large-scale AI/HPC workloads in cloud and enterprise environments.
2025.11.24by 명세환 기자
Supermicro Launches NVIDIA HGX B300-Based Liquid Cooling Solution
Global IT solutions company Super Micro Computer, Inc. (hereinafter Super Micro) has unveiled a new liquid-cooled server solution featuring NVIDIA's latest HGX B300 platform, beginning in earnest to provide highly integrated and efficient infrastructure for hyperscale data centers and AI factories.
2025.12.15by 배종인 기자
.jpg)
SKT Partners with Supermicro and Schneider to Build Modular AI Data Centers at MWC26
SK Telecom has signed a three-party Memorandum of Understanding (MOU) with Supermicro and Schneider Electric at MWC26 held in Barcelona, Spain, to build an AI data center (AI DC). The three companies will jointly review an integrated solution using a "Pre-fabricated Modular" approach, in which servers and equipment such as power and cooling systems are integrally designed in module units, pre-manufactured, and then assembled on-site. SKT will participate in promoting the integrated model based on its operational experience, while Supermicro will take on the role of GPU server and cluster configuration capabilities, and Schneider will be responsible for MEP (Mechanical, Electrical, and Plumbing) design and operational solution services. Specific construction timelines and cost reduction figures have not been disclosed.
2026.03.04by 명세환 기자
.jpg)
Supermicro Unveils Data Center Solution Powered by NVIDIA 'Vera Rubin'
Supermicro, a leader in the AI infrastructure market, has unveiled its data center building block solutions (DCBBS) product family that supports next-generation Nvidia Blackwell Rubin NVL72 and HGX Rubin NVL8, as well as Blackwell CPUs.
2026.03.20by 배종인 기자
.jpg)
Supermicro Unveils Compact Edge Server… Center of Gravity for AI Computing Shifts to the Field
Supermicro has announced a lineup of compact edge systems based on AMD EPYC 4005 series processors. The product lineup consists of a mini 1U box form factor, short-depth 1U rack-mount, and slim tower configurations, targeting environments with significant space and power constraints such as retail, manufacturing, healthcare, and enterprise branch locations. Supermicro stated that the products were designed with both real-time AI inference and general workload processing in mind. Given that the AMD EPYC 4005 series is positioned for small to mid-sized enterprises, hosting providers, and branch office IT environments, this launch aligns with the trend of expanding AI infrastructure deployment beyond data centers to on-site edge locations.
2026.04.15by 배종인 기자
