The A16 design includes NVIDIA Ampere architecture-based CUDA cores, second-generation RT Cores, and third-generation Tensor Cores, delivering the flexibility to support a wide range of compute workloads, including computer vision using the latest H.265 codecs for encode and decode. An A16 has more than double the encoder throughput compared with the previous generation M10 card. Computer vision video applications that require video encoding and decoding should also benefit from the A16 design.
The A16 also has a unique quad-GPU board design that offers more hardware encoders and decoders for computer vision applications compared to earlier validations using the NVIDIA A40 single-board design. The potential performance impacts of these additional A16 hardware encoders and decoders compared to an A40 were a primary research question for this validation of these computer vision workloads.