MSI Showcases Liquid-Cooled NVIDIA MGX Server and NVIDIA DGX Desktop Supercomputer at Computex 2026

Liquid cooling, NVIDIA Blackwell hardware, and ultra-fast 400GbE networking are at the heart of MSI’s latest enterprise showcase at Computex 2026. The company is presenting the CG681-S6093 liquid-cooled server based on NVIDIA MGX architecture and the XpertStation WS300 desktop supercomputer built on NVIDIA DGX Station architecture. Together, the two platforms cover everything from large rack-scale deployments to deskside development and research workloads.

CG681-S6093 Packs Eight RTX PRO 6000 Blackwell GPUs in a Liquid-Cooled Rack Design

The CG681-S6093 is a 6U server designed for demanding compute environments. MSI pairs dual AMD EPYC processors with support for up to eight NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs. Each GPU can operate at up to 1200W, giving the platform the computing power needed for large-scale training, simulation, and inference workloads.

NVIDIA ConnectX-8 SuperNICs provide up to 8×400Gbps Ethernet connectivity, allowing rapid communication between nodes across large deployments. Fast data movement becomes increasingly important as workloads grow in size and complexity.

MSI has developed the platform around a rack-scale architecture. A 48RU rack can accommodate up to four CG681-S6093 servers, creating a dense computing environment for enterprise deployments. NVIDIA Spectrum-4 SN5600 scale-out Ethernet switches and SN2201 out-of-band switches provide connectivity throughout the rack.

Liquid cooling runs across the entire platform to support sustained multi-GPU workloads. The cooling design helps maintain stable operation under continuous heavy loads while improving overall efficiency inside high-density deployments.

XpertStation WS300 Delivers GB300 Grace Blackwell Power on the Desktop

While the CG681-S6093 is built for data centers, the XpertStation WS300 takes similar computing power and places it on a desk. Built on NVIDIA DGX Station architecture, the workstation targets developers, engineers, researchers, and professionals working with large models and data-intensive applications.

The workstation is powered by NVIDIA’s GB300 Grace Blackwell Ultra Desktop Superchip. MSI states that it supports up to 748GB of coherent memory and delivers up to 7.1TB/s of HBM3e memory bandwidth. Those specifications place the WS300 among the most powerful desktop computing platforms announced so far.

The architecture enables high-speed CPU and GPU data sharing, helping accelerate development, fine-tuning, inference, and other compute-heavy workloads. Larger datasets and more complex projects can be processed directly on the workstation without relying on larger infrastructure.

Dual NVIDIA ConnectX-8 SuperNIC connections provide 400GbE networking, allowing the workstation to connect directly to larger deployments when needed. This gives users access to enterprise-grade connectivity while maintaining the convenience of a desktop platform.

Liquid cooling is integrated into the workstation as well. The design helps sustain performance during long-running workloads while keeping the footprint significantly smaller than traditional rack-mounted hardware.

MSI is demonstrating the XpertStation WS300 at both Computex 2026 and NVIDIA GTC Taipei, where visitors can see practical deployment examples and development workflows in action.

With the CG681-S6093 and XpertStation WS300, MSI is expanding beyond traditional servers and workstations with platforms built around NVIDIA’s latest Blackwell technologies. From rack-scale deployments with eight RTX PRO 6000 GPUs to a desktop platform featuring the GB300 Grace Blackwell Ultra Superchip, the company is targeting organizations that need more computing power, faster networking, and efficient cooling for next-generation workloads.

MSI liquid-cooled NVIDIA MGX server racks and NVIDIA DGX workstation platform designed for enterprise AI and high-performance computing workloads
MSI and NVIDIA liquid-cooled AI systems

Source: MSI

This site uses Akismet to reduce spam. Learn how your comment data is processed.