- Deliver up to 1,000 ai tops (fp4 sparse) and 1 petaflop of compute density, enabling on desk training of llms up to 200 billion parameters and generative workloads that normally require a full data centre rig.
- Run 128 gb of lpddr5x coherent unified memory over a 256 bit, 273 gb/s bus, providing the bandwidth needed for dense model fine tuning of up to 70 billion parameters while keeping latency low.
- Connect via the nvidia connectx 7 smartnic and 10 gbe lan, plus wi fi 7 and four usb c ports with dp 1.4a, giving dual unit clustering at 200 gb/s, fast external storage, and multi display output from an ultra compact 1.2 l chassis.
Description
Step into the vanguard of local artificial intelligence development with the MSI EdgeXpert-34SAU Personal AI Supercomputer Mini PC. Engineered on the breakthrough NVIDIA DGX(TM) Spark platform, this ultra-compact 1.2-liter powerhouse brings data-center class execution directly to your desk. It is powered by the revolutionary NVIDIA Grace Blackwell GB10 Superchip, integrating a high-throughput 20-core ARM processor (10x Cortex-X925 + 10x Cortex-A725) alongside next-generation Blackwell graphics architecture. This potent combination unleashes a staggering 1,000 AI TOPS of Tensor performance (FP4 Sparse), delivering up to 1 PetaFLOP of parallel compute density for training, local inference, and generative workflows.
Equipped with 128GB of LPDDR5x coherent, unified system memory running on a 256-bit interface, the EdgeXpert-34SAU effortlessly hosts, tests, and validates Large Language Models (LLMs) up to 200 billion parameters, and can fine-tune dense models up to 70 billion parameters locally. A massive, self-encrypting 4TB NVMe M.2 SSD ensures lightning-fast data ingestion pipelines and maximum on-device file security.
Designed for seamless ecosystem scaling, the unit features an enterprise-tier NVIDIA ConnectX-7 SmartNIC, enabling dual-system clustering over a 200Gb/s interconnect to double your unified memory and handle frontier models up to 405 billion parameters. Running on the production-ready NVIDIA DGX(TM) OS, it allows researchers and developers to build locally and deploy to DGX Cloud with zero code refactoring. Complete with a 10 GbE ultra-fast LAN port, cutting-edge Wi-Fi 7, and multiple DisplayPort-capable USB-C interfaces, this desktop mini PC completely redefines the boundaries of private, edge-based machine learning.
Specifications
Technical Parameter
Details
Model / MPN
EdgeXpert-34SAU (MS-C931 / 956-C931-101)
Platform Base
NVIDIA DGX(TM) Spark(TM) Platform
Processor / Superchip
NVIDIA(R) Grace Blackwell (GB10) Architecture
CPU Architecture
20-Core ARM (10x Cortex-X925 + 10x Cortex-A725)
GPU Architecture
NVIDIA Blackwell (5th Gen Tensor Cores / 4th Gen RT Cores)
Tensor Compute Performance
Up to 1,000 AI TOPS / 1 PetaFLOP (FP4 Sparse)
Supported Data Formats
TF32, FP16, BF16, INT8, FP8, FP6, FP4
System Memory
128 GB LPDDR5x Coherent Unified Memory
Memory Bus / Bandwidth
256-bit / 273 GB/s
Storage Capacity
4 TB NVMe M.2 SSD with Hardware Self-Encryption
High-Speed Interconnect
NVIDIA ConnectX(R)-7 SmartNIC (Supports 2-Unit Cluster Mode)
Wired Networking
1x RJ-45 10 GbE LAN Port
Wireless Connectivity
Wi-Fi 7 + Bluetooth 5.4
USB Interface
4x USB 3.2 Gen 2x2 Type-C (20 Gbps, DP 1.4a Alt Mode)
Display & Audio 1x HDMI 2.1a, Up to 3x DP via USB-C, HDMI Multichannel Audio
Media Engine
1x NVENC | 1x NVDEC Hardware Accelerators
Operating System
NVIDIA DGX(TM) OS (Pre-installed Full AI Software Stack)
Power Dynamics
140W Superchip TDP / ~240W Type-C External Power Adapter
Chassis Dimensions
151 mm x 151 mm x 52 mm (Ultra-Compact 1.2L Form Factor)
System Weight
1.2 kg