Compal Debuts Next-Gen Liquid Cooling Solutions at Supercomputing 2026

Compal Debuts Next-Gen Liquid Cooling Solutions at Supercomputing 2026



Compal Electronics (TWSE: 2324) has revealed its cutting-edge AI infrastructure solutions that leverage NVIDIA's latest technological advancements during the Supercomputing 2026 (SC26) event in Chicago, taking place from November 17 to 19 at McCormick Place. Highlighted at Booth #5008, Compal showcases a sophisticated rack-scale liquid cooling system integrating NVIDIA's HGX Rubin NVL8 platform along with the NVIDIA Vera CPU Compute Tray. This combination underscores Compal's expertise in comprehensive engineering and system integration, pulling together various compute modules into fully functional NVIDIA Vera CPU rack systems.

The NVIDIA Vera is recognized as the fastest CPU for autonomous agents and simulations, offering twice the processing speed for sophisticated AI tasks, making it ideal for data-driven factories and quicker reinforcement learning processes. Compal's solutions are designed to respond to the increasing compute density, energy efficiency, and cooling requirements arising from the growing demands of generative and agentic AI applications.

As AI tasks transition from basic large language models to more intricate reasoning and autonomous workflows, data centers must adapt to these changing requirements. Compal is addressing this challenge by providing a combination of GPU-accelerated computing, CPU-intensive processing, high-bandwidth interconnections, and advanced cooling systems, all while ensuring seamless system integration.

The showcased rack-scale system, built on the HGX™ Rubin NVL8 platform, includes the capability to accommodate up to nine direct liquid-cooled OG231-2-L1 servers within a single rack. This setup supports a staggering 72 NVIDIA Rubin GPUs, making it capable of managing large-scale AI training and compute-heavy tasks. Each server is equipped with eight Rubin GPUs and utilizes NVIDIA NVLink™ 6 technology, promising impressive performance metrics. According to NVIDIA's specifications, this eight-GPU setup can yield up to 400 petaflops of NVFP4 inference performance along with an outstanding 24 TB/s of aggregate bandwidth.

Impressively, when compared to the previous NVIDIA HGX™ B200 model, the Rubin NVL8 platform offers an estimated tenfold increase in token throughput under specific AI inference workloads, significantly enhancing processing capabilities.

Additionally, Compal has implemented direct liquid cooling (DLC) alongside a comprehensive power infrastructure and centralized system management. This strategic design not only bolsters deployment flexibility but ensures operational efficiency for data center operators facing power and thermal issues related to high-density GPU deployments. Beyond GPU technologies, Compal also presents a complete NVIDIA Vera CPU Rack solution that caters to agentic AI, reinforcement learning, and CPU-intensive tasks. They are demonstrating the NVIDIA Vera CPU Compute Tray at SC26, which highlights their capacity for comprehensive solution engineering—from individual compute modules to complete rack-scale infrastructures.

The OX130-8N-L1 of the NVIDIA Vera CPU rack can house up to 32 CPU Compute Trays, totaling 256 NVIDIA Vera CPUs, and supports an impressive 22,500 simultaneous CPU execution environments. This configuration allows for enhanced performance, often resulting in about 1.8 times better efficiency compared to NVIDIA's established x86 CPU metrics and possibly quadrupling energy efficiency under set conditions. This high-performance architecture is tailor-made for demanding workloads involving AI agent execution, code processing, tool usage, and data management.

Alan Chang, Vice President of Compal's Infrastructure Solutions Business Group, stated, "As AI progresses from simple model training towards large-scale inference and autonomous applications, clients need more than enhanced computing power; they require all-inclusive, deployment-ready infrastructure solutions. Our integrated approach combines compute, power, and cooling within our NVIDIA HGX™ Rubin NVL8 platform, aimed at establishing integrated system designs for the future of AI infrastructure." Compal’s commitment to continuous collaboration with NVIDIA signifies a forward trajectory in optimizing compute density and energy efficiency, accelerating the application of innovative AI technologies.

With a reputation for exceptional engineering capabilities, Compal is poised to deliver advanced computing solutions ready for deployment across various applications including generative AI, scientific research, and high-performance computing. This strategy resonates with the ethos of SC26, themed

Topics Consumer Technology)

【About Using Articles】

You can freely use the title and article content by linking to the page where the article is posted.
※ Images cannot be used.

【About Links】

Links are free to use.