#United States #San Jose #AI Server #Compal Electronics #NVIDIA HGX

Compal Presents High-Density AI Server Solution at GTC 2026

Compal Electronics made a significant impression at the NVIDIA GPU Technology Conference (GTC) 2026 with the introduction of its cutting-edge AI server solution, the SG231-2-L1, built on NVIDIA’s HGX Rubin NVL8 platform. This innovation highlights Compal's commitment to advancing AI supercomputing infrastructure amidst growing demands for AI capabilities across industries.

Unpacking the SG231-2-L1

The SG231-2-L1 model stands out with its high-density accelerated architecture, incorporating eight NVIDIA Rubin GPUs in a compact 2U chassis. This design not only maximizes computational density but also enhances space efficiency, positioning the server as a feasible option for facilities with limited physical footprint.

Performance Capabilities

Compal’s new server excels with impressive performance metrics, delivering up to 400 petaFLOPS of computational power utilizing NVFP4 technology. This level of performance is particularly beneficial for large language models (LLMs) and other intensive AI training and inference tasks. The high-bandwidth GPU interconnect supported by NVIDIA NVLink 6 further amplifies its performance, achieving an astounding 28.8TB/s of bandwidth between GPUs, thereby facilitating enhanced multi-GPU scalability for complex workloads.

Memory and Cooling Features

Designed to handle memory-intensive AI applications, the SG231-2-L1 supports up to 2.3TB of GPU memory alongside a remarkable memory bandwidth of 176TB/s. To maintain stable performance under heavy workloads, the server features an advanced direct liquid cooling solution, sustaining approximately 24kW of power, thus ensuring efficiency even in demanding operational scenarios.

Rack-Level Deployment and Scalability

The server is engineered for seamless integration into high-density AI racks, offering easy scalability from single nodes to extensive data center deployments. This feature is critical for organizations looking to future-proof their investments in AI infrastructure.

Compal’s booth at GTC also featured an NVIDIA Vera CPU HPM module, further showcasing the company's engineering readiness and alignment with NVIDIA’s vision for heterogeneous computing architectures, which combines CPUs and GPUs for optimal performance.

Strategic Vision and Market Impact

Alan Chang, Vice President of Compal's Infrastructure Systems Business Group, emphasized the shift in AI infrastructure competition from isolated performance evaluations to broader aspects of deployment efficiency. As AI applications continue to expand, Compal is dedicated to enhancing the compute capability and overall scalability of data center architectures. This approach enables clients to adopt sustainable models tailored for next-generation AI workloads.

As a leading manufacturer with decades of experience, Compal not only aims to bolster its technological capabilities but also reaffirms its dedication to maintaining a strong presence within the NVIDIA ecosystem. The showcase at GTC 2026 is a testament to Compal's robust engineering abilities and its foresight in the increasingly competitive landscape of AI infrastructure solutions.

About Compal

Founded in 1984, Compal Electronics is recognized as a leading manufacturer within the notebook and smart device sectors. With a focus on innovation and strategic partnerships, Compal has frequently been acknowledged for its groundbreaking product designs, earning numerous international accolades. Notably, in 2025, it was recognized as one of Taiwan's top manufacturers and has consistently featured on Forbes' Global 2000 list. More recently, Compal has been keen on diversifying into emerging markets including cloud computing, automotive electronics, and smart medical technologies, utilizing its comprehensive hardware and software development capabilities to deliver tailored solutions.

For more detailed information regarding Compal’s offerings and initiatives, visitors are encouraged to explore their official website.