ASRock Rack has developed a new edge server that leverages NVIDIA’s Thor industrial system-on-chip (SoC), marking a shift toward specialized AI solutions for sensor-heavy environments. This 2U short-depth chassis, named 2UXGI-THOR, targets industrial and medical use cases requiring robust real-time data processing at the edge.
The core component is NVIDIA’s IGX Thor SoC, which integrates 14 Arm Neoverse-V3AE CPU cores with an NVIDIA Blackwell GPU, emphasizing high-efficiency sensor data ingestion and analysis. Unlike typical data center servers focused on x86 or NVIDIA Grace processors, this platform caters to industrial demands such as programmable vision acceleration, sensor bridging, and Camera-over-Ethernet protocols.
At the heart of ASRock’s design lies the IGX T7000 microATX motherboard, which combines the Thor module with 128GB of LPDDR5X memory and a high-speed NVIDIA ConnectX-7 network interface card. This configuration facilitates placement within compact server racks, meeting the spatial and technical requirements of edge deployments.
To amplify AI processing beyond the SoC’s capacity, ASRock Rack equipped the server with an NVIDIA Blackwell PRO video card via a PCIe slot. Customers can select either the PRO 5000 or PRO 6000 GPU, effectively transforming the system into a powerhouse for AI workloads demanding extensive compute performance.
The server offers versatile connectivity options, including a 1GbE RJ45 port via the Thor SoC’s integrated NIC and a pair of high-bandwidth 200GbE QSFP28 ports. These network interfaces enable significant data throughput, critical for industrial environments that manage large streams of sensor data in real time.
This development stands out for adhering to NVIDIA’s MGX modular server specification, which promotes flexible configuration of compute units to suit specific vendor or customer needs. ASRock Rack’s implementation pushes this modularity into the realm of industrial AI edge servers, contrasting typical MGX systems geared toward data center workloads.