Technology Forum: Heterogeneous Computing and Composable Architectures with Next-Gen Interconnects

The next step in taking full advantage of heterogeneous computing is the ability to use larger numbers and multiple types of accelerators, each optimized for a particular step in the workflow. In most accelerated workflows, data often must move to and from storage with each processing step. The result is slower than necessary application performance and overburdened networks. The answer is not just faster interconnects, but a more efficient and productive interconnect fabric enabling a seamless passing of data from one processing step to the next and reducing trips to and from storage. For effective heterogeneous computing one needs composable infrastructure with the ability to balance CPU-to-GPU compute ratios, create systems with different types of GPUs, as well as optimal GPU-to-GPU and GPU-to-storage communications, and the ability to scale solutions spanning multiple GPU appliances. GigaIO has developed a next generation interconnect fabric based on PCIe and CXL, called FabreX, to address these challenges.