Supermicro is introducing its 100% direct liquid-cooled rack solution powered by the NVIDIA Vera CPU — a fully integrated system designed to handle the emerging requirements of agentic AI and the most demanding HPC simulation workloads at scale. While the NVIDIA Vera CPU delivers the raw compute and memory bandwidth, the story starts at the rack level with highly scalable architecture. Supermicro engineers the complete infrastructure stack to maximize density, efficiency, and deployability in both HPC and AI data centers.
Supermicro’s Vera CPU rack is delivered as part of its Data Center Building Block Solutions (DCBBS®) — an end-to-end infrastructure framework that spans compute, storage, networking, power distribution, advanced liquid cooling, management software, and deployment services. DCBBS covers every layer of the stack, from server-level integration through facility infrastructure, backed by a dedicated Supermicro team managing the full deployment lifecycle: site survey, project design, rack integration, solution validation, and ongoing support. This is the same framework Supermicro has used to deploy some of the world’s largest liquid-cooled AI factories, including facilities with over 100,000 GPUs.
At the core of Supermicro’s DCBBS for the Vera CPU rack is DLC-2 — Supermicro’s second-generation advanced direct liquid cooling technology stack, engineered for near-total heat capture and full-stack integration. DLC-2 spans every layer of the cooling infrastructure: direct-to-chip cold plates, coolant distribution units (CDUs) supporting up to 1.8MW each, cooling manifolds, rear-door heat exchangers, and cooling towers. It features Supermicro’s proprietary SMC PG25-A coolant, engineered for exceptional chemical and thermal stability across sustained high-density operation.
For the Vera CPU rack, 100% direct liquid cooling supported by DLC-2 is what makes the density possible: 256 NVIDIA Vera CPUs, 22,528 cores, and up to 300 TB/s of aggregate memory bandwidth in a single 48U rack footprint. By eliminating chilled-air overhead entirely, DLC-2 delivers dramatically lower PUE, reduced facility cooling costs, and the ability to deploy far more compute per megawatt — a decisive operational advantage for HPC centers and AI factories running under strict power constraints.
HPC workloads and agentic AI orchestration share more in common than is widely recognized: both are memory-bandwidth intensive, both rely on fast per-core execution, and both demand predictable performance under sustained parallel load. In agentic AI systems, AI models and accelerated simulations run on GPUs, while CPUs coordinate and execute much of the surrounding workflow: : compiling code, invoking tools, querying databases, running sandboxed environments, launching CPU-based simulations, and returning results before the next reasoning step begins. The Supermicro Vera CPU rack is purpose-built to scale this execution layer, supporting more than 22,500 concurrent CPU environments per rack while delivering the memory bandwidth and thermal efficiency that HPC simulation workloads have always demanded. That same CPU layer also runs CPU-based simulation directly — supporting traditional as yet unaccelerated HPC workloads as well as agent-driven simulations that step through physics, test hypotheses, and generate feedback before the next model call.
To learn more about Supermicro’s NVIDIA Vera CPU rack solutions, DCBBS architecture, and liquid-cooling expertise, visit https://www.supermicro.com/en/accelerators/nvidia/vera-rubin
By clicking subscribe, you consent to allow Supermicro to store and process the personal information submitted above to provide you the content requested.
You can unsubscribe from these communications at any time. For more information on how to unsubscribe, our privacy practices, and how we are committed to protecting and respecting your privacy, please review our Privacy Policy.
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。