The MAU™ Accelerator is an AI inference accelerator for data centers. It achieves the highest throughput for latency-constrained inference workloads. These features make it well suited to real-time applications such as speech transcription or synthesis, time series prediction or fraud detection in financial transactions. Neural network models developed using popular frameworks such as TensorFlow, MXNet or PyTorch can be accelerated using the MAU Accelerator. Some of the latest accelerator cards from Intel and Xilinx can host the MAU Accelerator and these are available off the shelf and in the cloud today.