Austrian IoT tech specialist Kontron has launched a 3U VPX GPU board designed for AI inference and sensor processing in industrial environments. The VX33211 is intended to facilitate complex data workloads directly at the network edge.

The board is powered by the NVIDIA RTX PRO 2000 Blackwell embedded GPU. It delivers up to 13.78 TFLOPS of FP32 performance and features 3,328 CUDA cores, 104 Tensor Cores for deep learning, and 26 RT Cores for real-time ray tracing. The system includes 8GB of GDDR7 memory providing up to 384 GB/s of bandwidth.

Designed for harsh conditions, the conduction-cooled version complies with VITA 48 standards and operates in temperatures ranging from -40°C to +85°C. The VX33211 uses SOSA-aligned OpenVPX profiles to ensure interoperability with existing systems and includes PCIe Gen4 connectivity for high-speed data transfer.

“The VX33211 brings powerful GPU acceleration to rugged embedded VPX systems, enabling customers to process complex data and AI workloads directly at the edge,” said Sébastien Vitre, Product Manager for Rugged Systems at Kontron.

The board is compatible with Kontron’s existing CPU hardware, such as the VX307C, and is integrated into the company’s HARAKAN-F platform. Typical applications for the technology include radar and sonar processing, electronic warfare, and industrial image processing.