MSI has introduced the EdgeXpert, a compact AI supercomputer built on the NVIDIA DGX Spark platform and the Grace Blackwell architecture. Designed for developers, researchers, and enterprise users, the system balances high performance with a small footprint, fitting advanced compute density into just 1.19 liters of volume.
Equipped with a 20-core Arm CPU and NVIDIA Blackwell GPU, EdgeXpert is optimized for local AI workloads, prototype development, and inference tasks, providing a portable alternative to cloud-based compute.
Performance
At FP4 precision with sparsity, EdgeXpert achieves up to 1000 TOPS (trillion AI operations per second). With FP8, performance can scale even further.
The system integrates 128 GB of unified LPDDR5x memory on a 256-bit interface, delivering 273 GB/s of bandwidth shared between CPU and GPU for high-speed data flow.
Storage Options
EdgeXpert offers two NVMe M.2 SSD configurations:
- 1 TB (self-encrypting)
- 4 TB (self-encrypting)
These options ensure secure local data processing, particularly for workloads with strict confidentiality requirements.
Connectivity and Expansion
The rear I/O layout includes:
- 4× USB 3.2 Type-C ports (20 Gb/s)
- 1× 10GbE RJ-45 Ethernet port
- 1× NVIDIA ConnectX-7 Smart NIC
The ConnectX-7 NIC supports dual QSFP56 200GbE links, enabling two EdgeXpert units to be interconnected. In this setup, compute performance doubles to 2000 TOPS, and memory expands to 256 GB.
For wireless, EdgeXpert includes Wi-Fi 7 and Bluetooth 5.3. Video output comes via an HDMI 2.1a interface, supporting multi-channel audio. Hardware-based encoding/decoding is handled by NVENC (encode) and NVDEC (decode), ensuring efficient multimedia processing.
Application Scenarios
MSI positions EdgeXpert as a versatile platform for:
- AI model prototyping
- Fine-tuning large language models (LLMs), up to 70B parameters
- Inference workloads
- Data science tasks, powered by 6144 CUDA cores
This makes it suitable for industries ranging from medical AI (via Holoscan) to robotics (via Isaac) and smart city analytics (via Metropolis).
Software and System Tools
EdgeXpert ships preloaded with NVIDIA DGX OS, eliminating the need for manual setup. Drivers, libraries, and toolkits are ready to use out of the box.
The system includes:
- DGX Dashboard for monitoring GPU utilization, memory usage, and thermals
- Integrated update tools
- Jupyter Lab support for streamlined development and operations
Specifications
| Category | Configuration |
|---|---|
| System Memory | 128 GB LPDDR5x (unified) |
| Memory Interface | 256-bit |
| Bandwidth | 273 GB/s |
| Storage Options |
Option A: 1× 4TB NVMe M.2 SSD (self-encrypting) Option B: 1× 1TB/2TB NVMe PCIe 4.0 SSD |
| Audio Output | HDMI multi-channel |
| Display | 1× HDMI 2.1a |
| Ethernet | 1× 10GbE RJ-45 |
| Smart NIC | NVIDIA ConnectX-7 (dual QSFP56 200GbE, device stacking supported) |
| Wireless | Wi-Fi 7 |
| Bluetooth | 5.3 |
| USB | 4× USB 3.2 Type-C (20 Gb/s) |
| Codec Support | NVENC (encode), NVDEC (decode) |
| OS | NVIDIA DGX OS (preinstalled) |
| Dimensions | 151 × 151 × 52 mm (1.19 L) |
| Weight | 1.2 kg |
According to MSI, EdgeXpert is designed as a cost-effective alternative to cloud AI services and traditional GPU servers. The goal is to lower the barrier to local AI computing with reduced TCO (Total Cost of Ownership) and shorter deployment cycles.
While MSI has not announced official global pricing, Australian distributors (Computer Alliance and Com International) have disclosed retail pricing:
- 128 GB + 1 TB version: AUD $6999 (~USD $4580)
- 128 GB + 4 TB version: AUD $7999 (~USD $5240)
