AI Supercomputer Based on NVIDIA® DGX Spark™ GB10 Platform
- 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
- ConnectX-7 networking:
Scales out from 2 to 4 EdgeXpert systems, expanding total VRAM up to 256 GB/384 GB/512 GB
- 3-node cluster: ring topology via 3 QSFP DAC cables (switchless)
- 4-node cluster: star topology via a 200 GbE QSFP Ethernet switch
- Overview
- Product Specifications
- Service
MSI EdgeXpert
Built for Developers. Powered for AI
EdgeXpert is designed from the ground up to build and run AI. Powered by the NVIDIA GB10 Grace Blackwell Superchip, it is a complete local AI platform for agents, multimodal inference, fine-tuning, data science, and edge development, bringing the power of NVIDIA Grace Blackwell to every desk.
- NVIDIA
GB10 Superchip - DGX OS
Ubuntu Linux - 1 petaFLOP
AI Performance - max.128GB
Unified Memory - up to 200 GbE
ConnectX-7 - Desktop
Agent Computer
NVIDIA® Grace
Blackwell Architecture
Delivers a powerful SoC solution combining NVIDIA Blackwell GPU technology with 20-core Arm CPU, offering up to 1 petaflop of AI compute. The GB10 Superchip uses NVLink-C2C to deliver a CPU+GPU coherent memory model with 5x the bandwidth of PCIe Gen 5.
Coherent
Unified System Memory
Features up to 128GB of coherent, unified LPDDR5x system memory. EdgeXpert enables users to run models up to 200B parameters locally. Coherent, unified memory lets multiple models and agents share one large pool for faster communication, lower latency, and rich long-running context.
NVIDIA® AI
Software Stack
Runs DGX OS, built on Ubuntu Linux, and comes preconfigured with the full NVIDIA AI software stack. Developers can hit the ground running with popular tools including PyTorch, Jupyter, and Ollama to prototype, fine-tune, and run inference locally, using the latest AI models on release day.
Scale to Multi-Node Clusters
with ConnectX-7 NIC
Equipped with a ConnectX-7 NIC, the EdgeXpert provides up to
200GbE. High-performance NVIDIA ConnectX networking enables the connection of up to four EdgeXpert systems to work with AI models of up to 700 billion parameters.
Scalable Desktop Solution
With built-in ConnectX-7 networking, up to four systems can be clustered to provide a large memory pool and up to 4 petaFLOPS of FP4 AI compute for larger models and more complex agents.
Up to 200B Parameters
Deploy Out of the Box
128GB Memory
1 petaFLOP FP4 Compute
Up to 405B Parameters
Scale When You Need More
256GB Memory
2 petaFLOPs FP4 Compute
Up to 700B Parameters
Enterprise-Grade Power
512GB Memory
4 petaFLOPs FP4 Compute
Superior Thermal Engineering for
Uncompromising Reliability
Built to withstand intense operational heat, EdgeXpert combines a patented FanDock design, high-efficiency SSD Vapor Chamber cooling, and higher footpads optimized for stacked deployments. Experience sustained speed, mission-critical reliability, and extended system lifespan.
- Overview
- Patented FanDock
- SSD Vapor Chamber
- Elevated Footpads
Thermal Solution: Sustained Performance for Always-On AI
Because AI agents run 24/7, thermal design determines whether peak performance is also sustained performance. EdgeXpert pairs a dedicated SSD Vapor Chamber, patented FanDock airflow technology, and elevated stacking footpads to stay rock-solid above 200W—yielding 10% faster token throughput and running up to 15°C cooler under heavy load.
Accelerated System Airflow
Directs cool air across the vapor chamber, sustaining over 200W of system power without downclocking. The result: approximately 10% faster GPT-OSS 120B inference than the reference design.
Sustained Storage Performance
Uses a copper-nickel plated vapor chamber to lower drive temperatures by 9°C, preventing storage throttling and preserving peak read/write throughput during continuous 24/7 inference.
Optimized Thermal Dissipation
High-clearance footpads create an open airflow channel between stacked units, reducing rear chassis temperatures by up to 15°C and top surface heat by 9.1°C for silent, multi-unit deployments
Run Autonomous Agents Locally, Securely, and Cost-Effectively
Keep your AI models, prompts, sensitive corporate data, and 24/7 inference strictly on-device. MSI EdgeXpert delivers full enterprise compute power behind your firewall—eliminating recurring per-token cloud API fees, network latency, and data leakage risks for completely private, autonomous agent execution.
Applications
Delivering the power of Grace Blackwell in a desktop-friendly size, MSI EdgeXpert is ideal for AI developer, researcher, and data scientist workloads—accelerating edge applications across medical, retail, and smart city environments.
Developers & Data Scientists
AI developers can fine tune AI model or AI inference application in local. High-performance data science at your desk.
Education & Research
Train locally in classroom labs and scale up on school server infrastructure.
Medical
Enables clinicians and researchers to run AI models locally for diagnostics, genomics, and drug discovery.
Robotics & Edge Applications
Fine-tuning edge AI vision transformers for real-time robotic factory monitoring and alerts on NVIDIA Jetson.
Retail
Drive real-time video analytics and automated checkout with local data processing.
Smart City
Uses NVIDIA's Metropolis and Isaac frameworks to power edge AI for smart cities, running real-time computer vision locally.
Resources
- Case Studies
- Videos
- Blogs
Legal AI Suite
Enterprise AI for Legal and IP Governance
AI Machine Vision
Operational AI for the Factory Floor: Memorence Operagents
AI Voice Translation
Secure, Real Time, and Fully Offline Communication
Local Edge Agentic AI
Keeps sensitive data fully within the enterprise perimeter
The 5 Surprising Secrets of This AI Supercomputer
The MSI EdgeXpert is a desktop AI supercomputer built on the NVIDIA GB10 / Grace Blackwell architecture.
The 5 Surprising Secrets of This AI Supercomputer
The MSI EdgeXpert is a desktop AI supercomputer built on the NVIDIA GB10 / Grace Blackwell architecture.
Let the Data Speak: Hermes Agent's Challenge to the AI
As we enter 2026, the AI Agent market is undergoing a seismic shift from "tool" to "automation productivity."
The 5 Surprising Secrets of This AI Supercomputer
The MSI EdgeXpert is a desktop AI supercomputer built on the NVIDIA GB10 / Grace Blackwell architecture.
FAQ
MSI EdgeXpert is a desktop AI supercomputer built on the NVIDIA DGX Spark platform, powered by the NVIDIA GB10 Grace Blackwell Superchip. It delivers up to 1,000 AI TOPS (FP4) of performance in a compact 151 x 151 x 52 mm chassis weighing 1.2 kg, letting developers run large AI models locally instead of in the cloud.
It's designed for prototyping, fine-tuning, and running large language models (LLMs) locally, with applications spanning robotics, smart city vision systems, healthcare diagnostics, genomics, education, and finance. Its full NVIDIA AI software stack supports frameworks like Isaac, Metropolis, and Holoscan for edge AI development.
It has 128 GB of LPDDR5x coherent unified system memory shared between the CPU and GPU via NVLink-C2C, on a 256-bit interface delivering 273 GB/s of bandwidth. This unified design lets a single unit run models with up to 200 billion parameters.
Yes. Two EdgeXpert units can be linked via their NVIDIA ConnectX-7 ports to scale up to models with roughly 405 billion parameters. This makes it possible to expand capacity without moving workloads to a data center or the cloud.
It ships with either 1 TB or 4 TB of self-encrypting NVMe M.2 storage. The self-encryption adds a layer of data security for on-premise AI workloads handling sensitive data.
It runs NVIDIA DGX OS, an Ubuntu Linux-based operating system built on Arm64, with the NVIDIA AI software stack pre-installed.
It includes a 10GbE RJ-45 port, an NVIDIA ConnectX-7 Smart NIC, Wi-Fi 7, Bluetooth 5.3, and 4x USB Type-C ports (up to 40 Gb/s). This combination supports both high-speed networking for multi-unit setups and standard peripheral connections.
Unlike a conventional PC, it uses unified CPU-GPU memory (rather than separate system RAM and GPU VRAM) and Arm-based silicon purpose-built for AI inference and fine-tuning rather than general computing or gaming. It's best understood as a developer appliance for local AI model development, not a general-purpose mini PC.
Product Specifications
- Part Number
- EdgeXpert MS-C931
System Processor
- CPU
- 20 core ARM, 10 Cortex-X925 + 10 Cortex-A725 ARM
- Frequency
- --
- L2 Cache
- --
- CPU Type
- onboard
- Chipset
- onboard
- BIOS
- --
- Tensor Performance
- 1 petaFLOP Al Performance (FP4, Sparse)
Memory
- Technology
- LPDDR5x
- Max. Capacity
- 128GB
- Socket
- onboard
Graphic
- Controller
- Nvidia® GB10
- Graphic Memory
- Share 128GB unified Memory
- Multiple Display
- 4 independent displays
- Display Interface
- HDMI 2.1, 4x DP1.4a via USB-C
Ethernet
- Controller
- 1 x 10GbE
Audio
- Controller
- --
H/W Monitor
- Watchdog Timer
- --
- Smart Fan
- Yes
TPM
- Controller
- Y
I/O Interface
- PS/2
- --
- CANbus
- --
- LVDS
- --
- VGA
- --
- RJ-45
- 1
- PoE
- --
- Audio
- HDMI Multichannel Audio output
- USB
- 4 x USB 3.2 Type-C
- Parallel
- --
- Serial
- --
- HDMI
- 1
- GPIO
- --
- DVI
- --
- DisplayPort
- 3 via USB-C
- Antenna
- 2
- QSFP
- 2
- WiFi
- WiFi 7*
* Wi-Fi 7 functionality depends on local regulatory approval. Restricted bands will be activated via software updates once legal requirements are met in the region.
- Bluetooth
- BT 5.4
Expansion Slot
- PCIe x16
- --
- PCIe x8
- --
- PCIe x4
- --
- PCIe x1
- --
- PCI
- --
- M.2
- 1x M.2 M Key 1TB/4TB SSD with self-encryption
- Mini-PCIe
- --
Storage
- SATA
- --
- CFast
- --
- mSATA
- --
- CompactFlash
- --
Power Supply
- DC Input
- ~240W USB-C with external power adapter
- AC Input
- --
Environment
- IP Rating
- --
- Operating Temperature
- 0 ~ 35°C
- Operating Humidity
- 10 ~ 90%, non-condensing
- Storage Humidity
- 10 ~ 90%, non-condensing
- Storage Temperature
- 0 ~ 60°C
- Vibration
- --
- Shock
- --
Physical Characteristics
- Dimensions
- 151 x 151 x 52 mm
- Weight
- 1.2 kg
- Mounting
- --
Certification
- EMC
- FCC Class B, CE, UKCA, VCCI, BSMI
- Safety
- BSMI
Software Platform
- OS Support
- NVIDIA DGX™ OS
NIC
- Controller
- ConnectX-7 NIC @200 Gbps