Nearly a decade ago, Facebook started the Open Compute Project (OCP) with Intel as a founding member to create an open hardware movement that met the challenges facing hyperscale data centers. Since then, Intel has driven more than 30 contributions and enablement of more than 100 products and partners across compute, network, and storage solutions from the cloud to the intelligent edge.
The success of OCP is part of a larger shift in the industry. What we see today is that business-critical workloads are evolving, driven by the growth of AI, cloud, edge, network and storage.
Compute and Storage Standards
Workload performance comes from compute, network, and storage optimization across the entire datacenter and even to the intelligent edge. For the agility necessary to meet users’ insatiable demand for data, we are announcing cloud optimized OCP platforms in collaboration with Facebook for our upcoming 3rd Generation Intel® Xeon® Scalable processors with built-in AI acceleration. Further extensions to the unique Intel DL Boost capabilities enhance deep learning training performance by up to 60%.* These platforms span up to 8-socket configurations, providing the foundation for scalable performance across the full breadth of customer workloads, including machine learning training on particularly large models.
We’re also working with Microsoft and Facebook as they develop new OCP Cloud Solid State Drive (SSD) specifications. These two key partners have had a lot of success with Non-Volatile Memory Express (NVMe) SSDs due to the robust ecosystem, improved reliability relative to HDDs, and essential features allowing them to deploy and manage flash at hyperscale. Together we are optimizing computation and storage at scale.
In the realm of connectivity, Intel is actively engaged in the Co-Packaged Optics (CPO) Collaboration, which is defining common standards for switch system building blocks and the interfaces between them. Intel’s leadership in integrating optics with networking silicon is best demonstrated by our industry-first co-packaged optics Ethernet switch with photonics engines from our Silicon Photonics Product Division and the 12.8 Tbps P4-programmable Tofino 2 ASIC from our Barefoot Division. We also played a key role in the definition of OCP NIC 3.0 specification, which is scaling in the market with multiple products, including new Intel Ethernet Network Adapters, at speeds ranging from 1GbE to 100GbE. 100GbE Intel Ethernet 800 Series, for example, is delivering up to 60% improvement in predictability and up to 60% lower latency for a major caching application—memcached—using an Intel technology called Application Device Queues (ADQ).
Furthermore, Intel’s Barefoot Division is announcing support of SONiC, an open source network operating system, for its next generation, 12.8 Tbps Tofino 2 Ethernet switch ASIC. Barefoot has also partnered with Keysight Technologies, who announced a new, cost-effective, dev-ops friendly multi-terabit network tester platform powered by Tofino for data center operators and network equipment manufacturers.
Improving Security Solutions
Security remains at the forefront of data center management. Last year, Intel announced that we were teaming up with Facebook and other partners on Project Cerberus—an open platform for defining security root of trust (RoT) solutions. We’re making great progress collaborating on Cerberus 2.0 specifications, and we intend to contribute the Intel Platform Firmware Resilience (Intel PFR) solution for protecting server systems from risks such as Permanent Denial of Service (PDoS) attacks.
Beyond this, we are actively engaged with initiatives like the Confidential Computing Consortium and partnering closely with Microsoft on the next generation Azure compute blade leveraging our upcoming Xeon Scalable processor with new levels of hardware-enhanced security.
The unprecedented rate of cloud, edge, and network growth is fueling amazing innovation opportunities. To manage this growth and scale, we need to partner as a community to serve the world’s computational needs sustainably. I’ve talked about sustainability before and believe that there's perhaps no greater need than to get to a carbon neutral, even a carbon negative, data center, and we are innovating with the community across network, storage, compute, systems, and even the supply chain, to help make this a reality.
Intel and our cloud partners deeply care about this issue and governing bodies are beginning to demand regulation. So, how can we, as industry partners, help advance sustainable, power efficient data centers? It's more than just a processor or node transition. We must work diligently from the CPU up to the physical infrastructure and innovate at every level: dynamic management of servers to balance performance and power, smart buildings that harness energy from server heat dissipation for reuse, and so much more.
We are actively driving predictive analytics tools to address memory failure and automated remediation to help improve server reliability, but these tools will be extended to help drive dynamic power allocation for more flexible and sustainable performance. Working together on hardware standards with industry groups like OCP is just one part of a larger path towards true sustainability. We deliver value through the open ecosystem and across the full platform lifecycle to drive solutions, learning, and empower our customers to scale. Our OCP solutions provide hardware and software technology that efficiently support the growing demands of compute infrastructure and bring improved performance, manageability, security, and sustainability to hyperscalers.
I’m extremely encouraged by the work that various teams across so many companies are doing to further the OCP community. Intel will continue to improve products and standards to optimize workloads wherever data is moved, stored, and processed. You can review details of our latest OCP contributions and watch keynotes from recent OCP summits to see how this work is progressing. For more information on our CSP solutions, visit intel.com/csp.
* Results have been estimated using internal Intel analysis and provided to you for informational purposes. Any differences in your system hardware, software or configuration may affect your actual performance. For more complete information about performance and benchmark results, visit www.intel.com/benchmarks.