Hyper-convergence is to integrate large clusters of cloud computing IaaS services into small-scale clusters, which is convenient for deployment in private clouds and enterprise clouds.
The hyper-heterogeneous fusion chip emphasizes the overall optimization of the system stack, which is to optimize the overall system running on the server into an efficient and high-performance single-chip mixed with multiple engines. Hyper-heterogeneous fusion chips can support both hyper-convergence and non-convergence (that is, extreme deconstruction and coexistence of multiple users and multiple systems).
We can simply divide the system into two planes:
Control and management plane: still software running on the CPU;
Computing and data plane: At this moment, CPU, GPU, other types of xPU, and even I/O can be regarded as equal computing engines. They complete the work they are good at and fully interact to form a more efficient and higher performance of a whole system.