High Performance Deep Learning AI Inference - INT8 >200 TOPS, actual throughput 2-10 times of GPU with the same power consumption, super low latency optimal for real-time applications.

Powerful Video Processing - high density H.264 (AVC), H.265 (HEVC) and AVS2 Video Decoding, supported resolutions up to 8K.

General Purpose - support for fast deployment of most mainstream neural networks in FP16, BF16 and INT8, including Detection, Classification, Segmentation, Video Processing, LSTM/RNN, Natural Language Processing (BERT), Recommender System, etc.

Standard Half-Height Half-Length (HHHL) PCIe Add-in Card with 75 W TDP, compatible with all servers without additional power supply.