What problems does the DPU not solve?
Computing tasks include not only I/O task processing, but also other system layers, and even computing tasks at the application layer. If the DPU is used as a comprehensive computing acceleration platform, the DPU can continue to integrate more acceleration functions.
The DPU replaces the CPU and becomes the central node. “The dragon slayer has become a dragon.” The communication between the CPU, GPU, and other xPUs is still very troublesome, and it is still an isolated island.
If the underlying architecture of the entire system is not updated, the DPU-centric architecture is still essentially control-centric rather than data-centric, and it still cannot achieve an order-of-magnitude improvement in overall data throughput and computation.
There is still the problem of physical space constraints, and the server space is limited, especially 2U or 1U servers, which also need strong enough computing power. And with the popularity of green data centers, the power consumption constraints of a single server are bound to increase. At this time, the problem of independent CPUs, GPUs and various independent accelerators needs to be solved urgently.
From a CPU-centric architecture to a DPU-centric architecture. It will make the DPU the top priority, the key “bottleneck” of the system, and the unbearable weight of the system.