Danny Weber
18:01 18-11-2025
© RusPhotoBank
At Supercomputing 2025, MSI unveiled NVIDIA MGX and DGX AI servers and workstations with Grace Blackwell chips for LLM training, edge and desktop AI.
MSI, one of the leading developers of high-performance solutions, unveiled a new generation of AI systems at the Supercomputing 2025 show in St. Louis. The lineup spans server and desktop products built on NVIDIA MGX, NVIDIA DGX Station, and NVIDIA DGX Spark architectures. The new platforms tap NVIDIA Hopper GPUs, NVIDIA RTX PRO 6000 Blackwell Server Edition, NVIDIA Blackwell Ultra, and the Grace Blackwell GB10 superchip.
These systems target a broad spectrum of tasks—from training large models and running complex simulations to edge computing and desktop AI development. MSI’s server platforms are optimized for large language models, deep learning, and NVIDIA Omniverse projects. Configurations can include Intel Xeon or AMD EPYC processors and support GPU accelerators up to 600 W, paving the way for substantial scalability. The mix signals a focus on modularity and headroom rather than one-size-fits-all builds.
The model includes two AMD EPYC 9005-series processors, eight PCIe 5.0 FHFL slots, 24 DDR5 modules, eight U.2 NVMe, and eight 400 GbE ports with NVIDIA ConnectX-8 SuperNIC adapters. This configuration delivers high bandwidth for building large AI clusters.
The Intel-based variant features two Xeon 6 processors, eight FHFL GPU slots, 32 DDR5 modules, and 20 PCIe 5.0 E1.S NVMe. The server is tuned for training and fine-tuning models.
A compact option with a single Intel Xeon 6, 16 DDR5 slots, and four double-width GPUs (up to 600 W). It’s a fit for edge deployments and inference workloads.
MSI’s corporate business general manager Danny Hsu said the new platforms enable customers to scale AI workloads and make full use of modern GPUs.
For developers and researchers, MSI introduced the AI Station CT60-S8060—a desktop workstation built on the NVIDIA DGX Station architecture. It uses the NVIDIA GB300 Grace Blackwell Ultra Desktop superchip. With up to 784 GB of unified memory, the station brings DGX-level compute to a specialist’s desk. It reads like a bid to put heavyweight training and prototyping within arm’s reach of small teams.
As of October 15, 2025, the EdgeXpert is available—a compact 1.2-liter system powered by the NVIDIA GB10 Grace Blackwell superchip with 128 GB of unified memory. EdgeXpert targets educational institutions, labs, research teams, and enterprise environments that need a powerful yet compact AI platform.
According to David Wu, who oversees MSI’s custom solutions, EdgeXpert helps narrow the gap between AI research and deployment by providing a secure, high-performance tool for professionals. The positioning underscores a clear push toward capable edge devices without the footprint of a rack.