MSI EdgeXpert Personal AI Supercomputer 4TB

MSI EdgeXpert Personal AI Supercomputer 4TB (ARM 20-core CPU/AI Blackwell GPU/128GB LPDDR5x/4TB/WiFi 7/BT 5.3/NVIDIA DGX OS) (9S6-C9311-61S) - 3 Year Local Warranty

Extreme AI Performance with NVIDIA Grace Blackwell

The MSI EdgeXpert Personal AI Supercomputer is built on the NVIDIA GB10 Grace Blackwell architecture, combining a powerful Blackwell GPU with a 20-core Arm CPU to deliver up to 1000 AI TOPS (FP4) of performance, enabling advanced workloads such as large language model (LLM) training, inference, generative AI pipelines, and real-time data processing directly on a compact desktop system without relying on cloud infrastructure.

Unified Memory Architecture for Large AI Models

Equipped with 128GB LPDDR5x unified memory, the system allows seamless data sharing between CPU and GPU through high-speed interconnect, significantly reducing latency and bottlenecks, making it ideal for handling large datasets, complex AI models, and memory-intensive workloads efficiently in a single system.

Local AI Development with Massive Scalability

Designed for enterprise and research environments, the EdgeXpert supports large AI models up to hundreds of billions of parameters and can scale further by linking multiple units together, enabling users to build, fine-tune, and deploy advanced AI systems locally with high performance and flexibility.

Compact Form Factor with Data Center-Class Capability

Despite delivering near data center-level performance, the system is housed in a compact desktop chassis, making it suitable for office desks, labs, and edge deployments where space is limited, while still providing powerful compute capabilities for demanding AI applications.

Full AI Software Stack with NVIDIA DGX OS

Preloaded with NVIDIA DGX OS, the system provides a complete AI development environment with support for CUDA, TensorRT, PyTorch, and TensorFlow, allowing developers and enterprises to rapidly develop, test, and deploy AI models with optimized performance and minimal setup.

Technical Specifications

Processor (SoC Architecture)
• NVIDIA GB10 Grace Blackwell Superchip
• 20-core ARM CPU (10 × Cortex-X925 + 10 × Cortex-A725)
• Unified CPU + GPU architecture via NVLink-C2C
• Designed for AI orchestration, preprocessing, and inference

GPU (AI Compute)
• NVIDIA Blackwell GPU architecture
• 5th Gen Tensor Cores + advanced AI acceleration
• Up to ~1000 AI TOPS (FP4 sparse performance)

Memory (Unified Architecture)
• 128GB LPDDR5X unified memory
• Shared between CPU + GPU (coherent memory model)
• 256-bit interface, ~273 GB/s bandwidth

Storage
• 4TB NVMe M.2 PCIe SSD (self-encrypting)
• High-speed storage for datasets, models, and pipelines

AI Performance & Capability
• ~1 petaFLOP AI compute (FP4 class)
• Supports LLMs up to ~200B parameters locally
• Multi-node scaling (2 units) up to ~405B models
• Optimised for training, fine-tuning, and inference

Operating System & AI Stack
• NVIDIA DGX OS (pre-installed)
• Supports CUDA, TensorRT, PyTorch, TensorFlow, RAPIDS
• Seamless scaling to NVIDIA DGX Cloud / data centre

Networking & Connectivity
• 1 × 10GbE RJ45 LAN
• NVIDIA ConnectX-7 Smart NIC (high-speed AI networking)
• WiFi 7
• Bluetooth 5.3

Ports & I/O
• 4 × USB 3.2 Type-C
• 1 × HDMI 2.1a
• Multi-display via USB-C DisplayPort

Form Factor & Design
• Ultra-compact desktop AI system (~1.2L chassis)
• Dimensions ~151 × 151 × 52 mm
• Weight ~1.2 kg

Power
• External power adapter (~280W class)

Cooling & Efficiency
• Optimised thermal design for sustained AI workloads
• High performance per watt (ARM + Blackwell architecture)

Security & Data Control
• Local AI processing (no cloud dependency)
• Full data sovereignty for enterprise workloads
• Self-encrypting SSD support