First used to be "Floating Point Unit" coprocessor (Intel 287), then GPU (Graphics Processing Unit),
and now NVidia comes with "Data Processing Unit" to help network communication in cloud / data centers. And similar to previous solution for Machine Learning SDK (NVIDIA CUDA),
for "DPU"s they have now developed "DOCA SDK". This could be part of the reason for very expensive purchase of ARM company.
Amazon / AWS has also developed similar specialized solution for AWS EC2 compute instances
called "Nitro", that offloads network, security and management to a separate board, feeing full CPU and GPU power for user tasks.
What's a DPU?
System on a chip that combines:
- Industry-standard, high-performance, software-programmable multi-core CPU
- High-performance network interface
- Flexible and programmable acceleration engines
very useful description in this podcast interview:
Intel better come with some relevant innovation soon...