ICP Mustang‑F100: FPGA Accelerator for Real‑Time Deep Learning Inference

The Mustang‑F100 FPGA accelerator, developed by ICP Deutschland, delivers high‑performance, real‑time deep learning inference for industrial applications. Built on the Intel® Arria® 10 GX1150 FPGA and equipped with 8 GB DDR4 memory, the card excels at video and image processing, machine‑vision analytics, and sensor‑data interpretation.

Its reconfigurable architecture allows it to adapt to changing workloads and support multiple floating‑point precisions. The Mustang‑F100 is compatible with a wide range of neural‑network topologies—including AlexNet, ResNet, and YOLO Tiny—making it suitable for tasks from object recognition and facial identification to image segmentation.

Powered by Intel® Enpirion’s integrated solution, the Mustang‑F100 achieves a low thermal design power of under 60 W while delivering up to 1.5 TFLOPs of performance. The integration of the Intel® OpenVINO™ toolkit further accelerates deployment: pre‑optimized libraries, operating‑system images, and pre‑trained models reduce time‑to‑market.

With a compact footprint of 170 × 68 × 34 mm and a standard PCIe Gen3 x8 interface, the Mustang‑F100 is straightforward to embed in inference platforms such as the FLEX‑BX or TANK‑870AI. ICP’s experts assist customers in selecting the right hardware and configuring multi‑card systems via unique card IDs for scalable deployment.