Let’s take another look at the DPU-centric architecture.
It needs to be clarified: the CPU computing-centric architecture is essentially control-centric; the DPU-centric architecture is not naturally the same as data-centric; if the DPU architecture implementation is still the traditional SOC, then the DPU It’s still essentially control-centric. To truly achieve data-centricity, the entire system architecture needs to be adjusted on a large scale.
Let’s take a look at what problems are specifically solved in the mixed state of multiple heterogeneous computing centered on DPU?
There is a saying that the DPU is a Switch, it does not do specific functions, but connects many PUs. In this case, the DPU actually does nothing, and such a DPU is meaningless.
Everyone has a consensus view on the DPU, which needs to support virtualization, networking, storage, and secure accelerated processing. In this case, the DPU itself completes a large amount of I/O data computing tasks, reducing the burden on the CPU and significantly improving the performance of the entire system.