Intel® IPU E2000: A collaborative achievement with Google Cloud
Author(s): Patty Kummrow, Corporate Vice President and General Manager, Network and Edge Ethernet Products Group
Today, Intel officially launched the 4th Gen Intel® Xeon® Scalable Processors, previously known as Sapphire Rapids. You may have caught Google Cloud’s remarks on its collaboration with Intel to deliver compute infrastructure optimized for customer workloads with new levels of price-performance and security. Back in October 2022, Google Cloud shared a private preview of its new C3 machine series — the first virtual machine in the public cloud featuring 4th Gen Intel® Xeon® Scalable Processors and the Intel® Infrastructure Processing Unit (Intel® IPU) E2000, previously known as Mount Evans. Co-designed by Intel and Google Cloud, the E2000 enables high-performance computing and data-intensive workloads and is the first of its kind as an ASIC-based IPU.
With the Intel® IPU E2000, Google Cloud’s C3 machines offload programmable packet processing to IPUs securely at 200Gbps. This frees up the high-performance computing power of the 4th Gen Intel® Xeon® Scalable Processors, allowing the CPUs to tackle customer workloads. Through this unique approach in architecture, Google Cloud’s C3 machines achieve performance gains of up to 20% over their previous C2 machine series.[1] The debut of the Intel® IPU E2000 in Google Cloud’s C3 machines marks a significant milestone in transitioning to the data center of the future, where IPUs become a control point to drive more robust security and stronger networking and storage performance.
IPUs are a Strategic Control Point to Evolving Data Center and Edge Architectures
Intel IPUs are versatile and powerful tools for managing infrastructure from the data center to the edge. The capabilities of Intel IPUs present compelling advantages to moving infrastructure workloads off CPUs and onto IPUs. They provide tremendous value through increased security, task migration, performance acceleration, and feature velocity needed for your evolving data center.
- Increased infrastructure security — increased security is achieved by isolating the tenant and applications from the infrastructure.
- Infrastructure task migration — Moving the infrastructure work to the IPU cores reduces host compute load and frees up the CPU to take on additional workloads. Lower system total cost of ownership is achieved by reducing the number of servers needed and the IPU replacing the need for dedicated network adapters.
- Accelerated performance — Intel IPUs utilize specialized accelerators that enable the CPU to free up cycles by pushing this work onto the more efficient IPU. Running workloads on the IPU reduces system power consumption, leading to lower operating costs and increased sustainability.
- Feature velocity — Intel IPUs facilitate customization through software utilizing highly configurable hardware such as the ability to offload storage configuration and management to the IPU, allowing for disaggregated storage.
Integrating Learnings from Previous Generations
For more than six years, Intel has been in the IPU business collaborating with key cloud service providers (CSPs) to develop market-leading products. With our FPGA-based products, we observed that hyperscale CSPs realized the value of IPUs over various stages. They started by accelerating networking, then moved to accelerate storage, then accelerating security, and finally, to infrastructure processing.
With these observations and integrating the learnings from development of our FPGA-based products, we began a co-development program with Google Cloud to harden the IPU solution. The past few years of this collaboration culminated in the launch of the Intel® IPU E2000 that is now part of Google Cloud’s C3 machine series. The opportunity to co-develop the E2000 with Google Cloud is one of our deepest collaborations. It has been a huge benefit to have the ability to hear the needs and use cases directly from Google Cloud and then respond by building them into the product.
For instance, the E2000 was designed with a leadership programmable packet processing engine to support numerous existing use-cases such as vSwitch offload, firewalls, and virtual routing. The E2000 also includes an Intel® Optane™ derived NVMe emulation engine that enables use cases, such as pooled storage, to provide both cost savings and location/sizing flexibility for the data center’s storage solution. Further, advanced crypto and compression engines enable the E2000 to accelerate infrastructure workloads such as the ZStandard compression algorithm. Finally, the E2000 is designed with a “software and developer first” mindset, supporting well-known software development kits (SDKs) like the Data Plane Development Kit (DPDK) and Storage Performance Development Kit (SPDK), as well as the new Infrastructure Programmer Developer Kit (IPDK), for ease of programming with open source solutions.
Focusing on the Software and Developers through Open Source Solutions
Open source software solutions allow developers to take full advantage of the E2000’s highly configurable hardware and isolated cores. These solutions also factor in the rapidly changing network requirements of data centers as they continue to evolve into the future. By shifting the focus to the software and developers, Intel is at the forefront of the programmable infrastructure transformation.
As a founding member of the Linux Foundation’s Open Programmable Infrastructure (OPI) Project, Intel values and is committed to open software to advance collaborative and competitive ecosystems. The project brings together hardware and software companies to develop a community-driven, standards-based, open ecosystem for the next generation of architectures and frameworks based on IPU and DPU technologies. Through simplification of network, storage, and security application programming interfaces (APIs) within applications, OPI enables rapid adoption and use of IPUs and DPUs in the cloud and data center.
Intel fully supports the development of IPDK as a sub-project of OPI. Advancing these tools with the Linux Foundation fulfills the need for programmable infrastructure across cloud, data center, communication, and enterprise industries and allows developers to innovate and advance technological developments at an accelerated pace.
Compared to proprietary, closed software that locks developers into vendor-specific hardware, open programmable platforms, like IPDK, allow developers to build vendor-agnostic applications for IPUs and DPUs. Should the time come to move onto next generation hardware, developers can easily move their applications to the new hardware without having to rewrite their applications. Open source programming generates the basis of the architecture and frameworks for IPU and DPU software stacks that can be applied across any vendor’s hardware offerings.
Future Collaborations Drive Future Innovations
This is an exciting time as IPUs are changing the way we look at how data centers are designed, built, and managed. Through IPUs like the E2000, Intel gives customers the opportunity to optimize security, flexibility, and performance in their data center. The Intel® IPU E2000 signifies the achievements that come from deep collaboration with our customers, and we look forward to future collaborations with our customers for future innovations.
Notices and Disclaimers:
Performance varies by use, configuration, and other factors. Learn more at www.Intel.com/PerformanceIndex.
No product or component can be absolutely secure. Your costs and results may vary. Intel technologies may require enabled hardware, software, or service activation.
© Intel Corporation. Intel, the Intel logo, and other Intel marks are trademarks of Intel Corporation or its subsidiaries. Other names and brands may be claimed as the property of others.
[1] Based on Google Cloud’s blog post, “The next wave of Google Cloud infrastructure innovation: New C3 VM and Hyperdisk” from October 11, 2022.
