Compal Electronics (Compal; Stock Ticker: 2324.TW), a global leader in technology innovation and system integration, makes its official appearance at Taiwan Expo USA 2025, unveiling its next-generation AI server lineup. The featured systems—SG720-2A, a high-density AI server, and SX420-2A, an ultra-scalable AI platform—demonstrate Compal’s expanding capabilities in intelligent computing, modular engineering, and sustainable data center solutions. These platforms are engineered to meet the rising demands of AI training, inference, and high-performance computing (HPC), reinforcing Compal’s strategic position in the global AI infrastructure market.
SG720-2A: High-Density AI Computing Platform for Scalable AI Models and Green Data Centers
The SG720-2A is built on cutting-edge architecture and supports up to eight AMD Instinct™ MI350X / MI355X GPUs, leveraging PCIe Gen5 and AMD Infinity Fabric™ for maximum performance and low-latency interconnects. It is purpose-designed for AI training, large model deployment, and HPC workloads. Its modular structure allows independent servicing and component upgrades, with hot-swappable PCIe slots enabling non-disruptive network card replacements. To address increasing thermal and power challenges, the SG720-2A offers an optional 2-phase direct liquid cooling (DLC) solution. In collaboration with ZutaCore, Compal integrated HyperCool™ technology, achieving a partial PUE as low as 1.01—up to 5% more efficient than conventional cooling systems—making it a forward-looking and sustainable platform for next-generation data centers.
- Supports up to 8 AMD Instinct™ MI350X / MI355X GPUs
- Optimized for AI, HPC, and large-scale model deployments
- PCIe Gen5 and AMD Infinity Fabric™ interconnects
- Compatible with ROCm™, PyTorch, and TensorFlow
- Optional 2-phase DLC for enhanced cooling efficiency
- Fits both EIA 19” and ORv3 21” rack standards
SX420-2A: Modular, Ultra-Scalable AI Platform for Agentic AI and High-Density Media Workloads
The SX420-2A is Compal’s flagship AI server, built on NVIDIA MGX modular architecture and designed for maximum flexibility and compute density. It supports up to eight NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs and can be deployed in either EIA 19” or ORv3 21” rack configurations, offering unmatched scalability for enterprise AI and research environments.
Each GPU features 96GB of ultra-fast GDDR7 memory and a passive cooling design, enabling reliable performance across demanding AI inference, agentic AI, high-resolution rendering, and media workloads. The platform’s modularity and thermal optimization make it ideal for data centers seeking both power and energy efficiency.
Key Technical Highlights:
- Accelerated by up to 8 NVIDIA RTX PRO 6000 Blackwell GPUs
- 96GB GDDR7 per GPU with passive cooling
- Ideal for agentic AI, HPC, graphics, and media workloads
- NVIDIA MGX-based architecture with flexible configuration
- Supports EIA 19” and ORv3 21” rack standards
As Compal continues to advance its AI infrastructure offerings, these new platforms reflect its commitment to performance, modularity, and environmental sustainability. By integrating with leading ecosystem partners, Compal is accelerating intelligent computing innovations for a wide range of industries and enterprise-scale AI applications.
Exhibition Information｜Taiwan Expo USA 2025
- Date: August 14 (Thu) – August 16 (Sat), 2025
- Venue: Kay Bailey Hutchison Convention Center (Hall A, 650 S Griffin St., Dallas, TX 75202, USA)
- Compal Booth Number: 407、409