The Graphcore Bow-Pod64 system is the latest-generation AI accelerator from Graphcore. This is a one-rack system consisting of 64 Bow-class Intelligence Processing Units (IPU) with a custom interconnect. The system provides for an aggregate 22 Petaflops/s of performance in half precision. It has a total of 57.6 GB In-Processor-Memory with a total of 94,208 IPU cores. The system consists of four servers for data-processing.
For more details refer to the POD64 spec
(Figure from https://www.graphcore.ai/products/poplar)
The Graphcore software stack includes support for TensorFlow and PyTorch using the Poplar SDK. The Poplar® SDK is t is the toolchain specifically designed for creating graph software for ML applications. It integrates with the traditional ML frameworks like PyTorch and TensorFlow allowing users to port their existing code to the IPU hardware-specific code. The various components of the poplar SDK stack are shown in the figure. It includes the PopTorch framework which is a wrapper over the PyTorch framework optimized to the IPU hardware. It also enlists the different PopLibs libraries supported, which enables to construct graphs, define tensor data and control how the code and data are mapped onto the IPU for execution.