Designing for the Future of System Architecture With CXL and Intel in the ATC

One of the highlights of the recent Supercomputing 2021 conference (SC21) was to see the gathering momentum toward unanimous industry support for the long-anticipated Compute Express Link (CXL) protocol, perhaps the most significant architectural shift in generations. On November 10, two weeks prior to the conference, the CXL Consortium and Gen-Z Consortium announced a letter of intent to identify synergies that, once enacted, will transfer all Gen-Z specifications and assets to the CXL Consortium. SC21 provided a fitting backdrop for the first in-person showcasing of working CXL technology.

For those unfamiliar, CXL is the open, cache-coherent interconnect standard that can increase server memory capacity and bandwidth while dramatically reducing latencies. Developing CXL's technical specifications is the work of the CXL Consortium, an open industry standard group consisting of OEMs and other ecosystem players. But before customers can start enjoying the full potential of CXL usage models to come, those OEMs will embark on a fundamental re-engineering of processors, accelerators, motherboards and many other components – hence the importance of industry-wide consensus and support.

The possibilities are enormous for complex use cases such as high-performance computing (HPC), artificial intelligence (AI), machine learning (ML), cloud computing, and others yet to be imagined. SC21 served as a preview of the great possibilities to come in 2022 as the promise of CXL advances from theory to reality – but having the right system architecture in place will be critical. 

Working with our principal ecosystem partners like Intel , World Wide Technology (WWT) is on the leading edge of development as we prepare to build, test, evaluate and validate all-new CXL-enabled system architectures in our Advanced Technology Center (ATC). Wherever you are in your server refresh lifecycle, now is an opportune time to evaluate the profound improvements in flexibility to be gained from rethinking your system architecture, factoring in the game-changing advantages of a CXL-enabled environment.

System architecture's new divide: before CXL and after

To appreciate the significance of a beyond-the-box system architecture on the world's server systems, consider the global impact of commercial air transport, touching everything from commerce to business and leisure travel, bringing the most remote destinations suddenly within reach. It's that profound a transformational change.

When fully functioning real-world use cases start to proliferate in 2022, their appearance will come none too soon: CXL arrives at a time when technology has struggled to keep pace with customers' need for next-gen performance on every front: faster data processing, greater memory capacity, larger bandwidth, and all of it in a smaller, more efficient footprint using standardized components.

The introduction of PCIe 1.0 and its subsequent versions served to connect the various system components – memory, cards, graphics, accelerators and so forth, but fell short of being able to connect additional components outside the server. Although there were numerous attempts to connect these disparate components in the past, none were deemed successful – until CXL.

CXL leverages the PCIe 5.0 physical layer infrastructure to create a common memory space across the host and all devices, even those outside the chassis. Crucially, cache-coherence means the entire system will have access to the same data as needed. This new interconnect capacity will complement and enable composable disaggregated infrastructures (CDI) solutions such as Liqid and GigaIO and leverage resources located outside the confines of a server chassis.  CXL-enabled systems also will provide the framework for today's most efficient and performant solutions that are designed to support complex workflows such as HPC and AI/ML, even assessing and improving operations in real time.

Conceiving and deploying novel applications for CXL will require a fresh, solutions-driven mindset. And it all starts with the right system architecture design.

Intel's technology portfolio is foundational to realizing CXL

True to its long-established culture of innovation and focus on the future of technology to benefit all, Intel is renowned for developing standards for emerging technologies and then releasing them to the ecosystem so that new products, platforms and architectures can be nurtured in an open environment, free from the constraints of proprietary solutions.

Intel is doing for CXL what it has also done for the oneAPI programming model for acceleration: it created a technology based on open standards, achieved support from its partners, and spearheaded the drive to make it ubiquitous across the industry. Through the years Intel has provided similar leadership to drive open standards and scale for many important technologies, including ethernet, USB and Wi-Fi.

As Intel has recently announced, CXL will be an enabled feature in next-generation Intel® Xeon® Scalable processors, code-named Sapphire Rapids, coming later in 2022. These server processors will feature critical complementary technologies such as PCIe 5.0 support with CXL 1.1 protocol for accelerators and memory expansion in the data center segment. With CXL, systems can leverage resources from various devices (GPUs, FPGAs, etc.) for use by applications as they're needed, regardless of where they reside. Before CXL, those resources would have remained isolated and unavailable. Being based on open standards, CXL supports integration with other complementary products such as Intel® Optane™ persistent memory and storage, graphics processors, networking adapters and more. Architectures will be able to mix and match other standards-based solutions without the limitations imposed by proprietary approaches. 

As a result of our close, longstanding partnership with Intel, we will receive an early release of Sapphire Rapids-based systems to begin building out and testing a CXL demo environment in our ATC.

Tomorrow's technologies – built, tested and validated in the ATC today

WWT's ATC is a one-of-a-kind research and development ecosystem built on a collection of physical labs, virtualized to create a collaborative innovation ecosystem. Its core mission is to design, build and deploy leading-edge technology products, integrated architectural solutions and digital business outcomes, educating and advising to benefit customers, partners, associates and communities worldwide.

This multi-campus incubator for IT innovation consists of four separate data centers, serving as our testing and validation facility for emerging technologies and applications. Over the years WWT has used the ATC to develop usage models for many technologies that are in common use today: SmartNICs composable disaggregated infrastructure, FPGAs, GPUs and more. Now it will explore once-in-a-generation breakthroughs made possible with CXL.

WWT's tight relationships with its partners earns us early access to the latest products and technologies from Intel, leading OEMs and other manufacturers. In addition to Sapphire Rapids, WWT will receive related components from a wide assortment of platforms supporting CXL, reaching beyond the x86 world. In addition, WWT works in cooperation with the CXL Consortium, leveraging its rapid evolution of the cache-coherent interconnect. Owing to this cooperative relationship, WWT customers will soon be able to explore new system designs with CXL technology in action on Intel® architecture in the ATC.

CXL changes everything – learn what it means for your organization

Just as aviation carried humans to an entirely new realm, the Compute Express Link CPU-to-device interconnect will transform how data centers are architected, how their components are engineered, and how customers can leverage their new flexible, high-performance, low-latency capabilities. WWT has watched these developments closely and can provide the expertise necessary to navigate a rapidly shifting landscape. The time to start thinking toward the future and planning your migration to a new computing paradigm is now.