Intel Infrastructure Processing Unit (IPU)

2021: Infrastructure Processor Announcement

On June 21, 2021, the company Intel introduced the Infrastructure processor Processing Unit (IPU), a programmable network device designed to offload and release resources () central processors CPU in provider cloudy and service systems. telecommunication

Infrastructure Processing Unit

IPU is a technology category that represents one of the important pillars of our cloud strategy. The infrastructure processor extends our portfolio of SmartNIC and is designed to reduce the complexity of data center management and resource efficiency. told Guido Appenzeller, Intel Data Platforms Group Technical Director

According to the company, IPU is a programmable network device for intelligent management of infrastructure resources at the system level by safely accelerating these functions in the data center.

The infrastructure processor enables cloud operators to migrate to a fully virtualized network and storage architecture while maintaining proper performance and predictability along with the required degree of manageability.

IPU has special features for accelerating modern applications based on microservice architecture in data centers. Research by Google and Facebook showed that from 22% to 80% of the CPU clock can be spent on microservice communication.

With IPU, cloud service providers can securely manage infrastructure functionality, allowing customers to fully control CPU and system memory functions.

IPU offers:

• The ability to accelerate infrastructure functions, including virtualized data stores, network virtualization, and security with dedicated protocol accelerators.
• The ability to offload the CPU cores by migrating the network virtualization and storage functions previously performed by the program to the CPU to the IPU.
• Optimize data center resource utilization with flexible workloads.
• Enable cloud service providers to configure infrastructure deployment at software speed.

The first Intel FPGA-based platform IPUs have already been deployed to a number of cloud service providers, and our first IPU ASIC is under testing. said Patty Kummrow, vice president of Data Platforms Group and general manager of Intel's Ethernet Products Group

Even before 2015, Microsoft first used reconfigurable SmartNIC adapters on Intel servers of various generations to offload and accelerate network modules and storage using services such as Azure Accelerated Networking. SmartNIC allows us to free up processor cores, optimize throughput and I/O for storage, and add other capabilities after deployment. Intel is initially our trusted partner, and we are pleased to see Intel continue to promote a clear industry vision for data centers with an infrastructure processor. said Andrew Putnam, Chief Hardware Engineer for Microsoft

Infrastructure Processing Unit

Intel will release additional FPGA-based IPU platforms and specialized chips (ASICs). These solutions will be provided with a software basis that will allow users to create software products for orchestrating clouds.

Emerging data centers will need an updated intelligent architecture where large-scale distributed heterogeneous computing systems work together and are seamlessly linked to act as a single computing platform. This architecture will help address resource constraints, congested data flows, and incompatible platform security solutions. This intelligent data center architecture will have three computing categories - general-purpose CPU, XPU for application and workload acceleration, and IPU for infrastructure acceleration - all connected through programmable networks to leverage data center resources.