MSI Launches EdgeXpert MS-C931 Based on NVIDIA DGX Spark: Detailed Review of a Compact AI Supercomputer

MSI has released its own version of the desktop AI supercomputer NVIDIA DGX Spark (formerly Project DIGITS) under the name MSI EdgeXpert MS-C931. The device is built on the NVIDIA Grace Blackwell GB10 system-on-chip, featuring 20 Arm cores (10 Cortex-X925 and 10 Cortex-A725) and 6144 CUDA cores, delivering up to 1000 TOPS in sparse FP4 computations and up to 32 TFLOPS in FP32. The unified LPDDR5x-8533 memory capacity is 128 GB, enabling the launch of models with up to 200 billion parameters and fine-tuning of models with up to 70 billion parameters. The storage is a 4 TB NVMe SSD from Phison. Connectivity includes 10GbE RJ45 based on Realtek RTL8127, Wi-Fi 7 (MediaTek MT7925), Bluetooth 5.3, and two QSFP56 ports for 200GbE from NVIDIA Mellanox ConnectX-7, allowing multiple devices to be combined in a cluster. The chassis, measuring 150×150×58 mm, weighs about 2.5 kg. According to MSI, the cooling system is 15 °C more efficient than the reference design, and the Wi-Fi antenna placement improves performance by 37% in the 6 GHz range. The device runs on NVIDIA DGX OS based on Ubuntu 24.04.4 Long Term Support (LTS). In tests, power consumption under AI load reached 180 W, with SoC temperatures up to 86 °C. Noise is described as low and steady. Pre-installed utilities include NVIDIA DGX Dashboard and AI Workbench. The primary purpose is local development, debugging, and optimization of AI models, data processing, as well as Edge AI solutions (machine vision, speech recognition, autonomous systems) while maintaining privacy.