MSI EdgeXpert Compact Desktop AI Supercomputer

Feature

• NVIDIA® Grace Blackwell (GB10) Architecture:
  – NVIDIA Blackwell GPU & Arm 20-core CPU
  – NVIDIA® NVLink®-C2C CPU-GPU memory interconnect
• 128 GB LPDDR5x coherent, unified system memory
• 1000 AI TOPS (FP4) AI performance
• Full stack solution, hardware & software, designed for AI developers

The MSI EdgeXpert AI Supercomputer redefines desktop AI computing, delivering petaflop-scale performance through the cutting-edge NVIDIA® GB10 Grace Blackwell Superchip—the same powerhouse at the core of NVIDIA DGXTM Spark. Purpose- built for developers, AI researchers, and data scientists, the EdgeXpert empowers local AI development with unmatched performance, scalability, and advanced features—all in a compact, desktop-ready form.

NVIDIA® Grace Blackwell Architecture

• NVIDIA Blackwell GPU & Arm 20 core CPUOptimizes data preprocessing and orchestration to accelerate model tuning and enable real-time inference with greater efficiency.
• NVLink®-C2C TechnologyOffers a seamless CPU+GPU memory model with up to five times the bandwidth of PCIe 5.0, ensuring ultra-fast data access and transfer.

Unmatched AI ComputingPower

• 1000 AI TOPS (FP4) Tensor PerformanceDelivers blazing-fast performance for effortlessly running complex AI workloads at scale.
• 128 GB LPDDR5x, unified system memoryProvides the large memory needed for smooth model development, rapid experimentation, and high-efficiency inference.

Running and scaling LLMs locally, from 200B to 405B parameters

• Supports AI models with up to 200 billion parametersRun LLMs locally for data security, low latency, cost control, and full AI workflow—prototyping, fine-tuning, and inference.
• Stack via NVIDIA ConnectX: Large AI models &PerformanceHigh-performance NVIDIA ConnectX networking enables two MSI systems together to work with AI models up to 405 billion parameters.

Seamless AI Model Scaling from Desktop to Cloud

Leverage NVIDIA’s AI software architecture to seamlessly scale from desktop to NVIDIA DGX™ cloud or other NVIDIA® accelerated data centers or cloudinfrastructures, just with minimal code changes.