Intel “Purley” and the Road to Holistic Data Center Leadership

By Charles King, Pund-IT, Inc.  July 11, 2017

For decades, generational microprocessor advances defined the cutting edge of IT industry innovation. That was partly due to the linear simplicity of Moore’s Law, and its role in highlighting significantly improved, often entirely new computing features and end user experiences.

Things have changed a bit as manufacturing processes and materials science ran up against the atomic scale barriers that Dr. Gordon Moore and others predicted. But just because processor advances have become harder and more expensive to capture doesn’t detract from their innovative qualities or essential value.

That’s certainly been the case in personal computing markets and solutions, but it also applies to data center-focused vendors and their business and government customers. That point is central to the “Purley” family of Xeon Scalable Processors  that Intel announced this week.

Purley by the numbers and colors

Intel says that its new chips represent “the biggest data center advancement in a decade”, but what exactly does that mean? From a technical standpoint, the new processors mark advances in virtually every aspect of Xeon:

  • Purley is based on Intel’s new core micro-architecture and a lower latency cache memory hierarchy design, resulting in significantly faster application response time and 6X average better performance than prior generation Xeon chips, a notable generational jump.
  • Along with significantly enhanced overall performance, Xeon Processor Scalable processors offer up to 27% more cores and a new Intel high speed interconnect that delivers greater inter-CPU bandwidth. That supports increased VM density and multi-tenancy for cloud environments, along with improved parallelizable workload performance.
  • By adding memory channels to Xeon Processor Scalable solutions, Intel boosted memory bandwidth and PCIe IO bandwidth by 50 percent.
  • Security algorithm performance is up to 2X better than prior gen Xeon chips. Intel technologies, including AVX-512, QuickAssist and FPGAs can be used to accelerate specific applications.
  • Customers can also take advantage of Intel innovations, like VT-x, MBE and TSC (virtualization technologies), RunSure (improved uptime/availability) and VMD (volume management device) to optimize system performance for hybrid cloud workloads and environments.

In addition, Intel recently introduced a new schema for its Xeon products that aims to simplify customers’ buying decisions:

  1. Bronze – Entry level solution. Single socket systems, suitable for light tasks. Offers scalable performance and standard RAS.
  2. Silver – Good mix of performance and efficiency. Single socket systems for moderate tasks. Supports Intel TurboBoost and Hyperthreading. Delivers scalable performance at low power and standard RAS.
  3. Gold – Great option for mainstream applications. Supports 2-4-socket systems, up to 22 cores, three uplinks and advanced RAS.
  4. Platinum – Best technology for enterprise workloads. Ideal for 2-, 4- and 8-socket systems. Supports up to 28 cores, three uplinks, up to 1.5TB of 2666MHZ DDR4 memory and highest accelerator throughput.

Some initially pooh-poohed this approach, but the fact is that Intel understands the inherent complexity of IT markets and the growing involvement of non-technical personnel in buying decisions. If the company’s classification shorthand makes it easier for customers to understand new solutions and be assured that they’re getting the right silicon for the right job, so much the better.

Holistic data center innovation vs. server-centricity

Another issue to consider in this launch is the degree of difference between Intel’s market position and those of other data center silicon vendors. That point wasn’t sufficiently, let alone fully addressed during AMD’s recent launch of its new EPYC solutions. The company deserves kudos for reconsidering its approach to processors and systems and developing intriguing, even innovative solutions for data center and cloud applications.

Understandably, the EPYC launch was compared favorably to AMD’s 2003 introduction of its Opteron processors which did a great deal to reset market and customer expectations around data center silicon. However, much of the highly positive analysis comparing AMD’s new offerings with Intel Xeon neglected a critical point: the fundamental shift in market position and penetration between the two companies that has developed in the past decade.

That’s not just related to the estimated near-99% volume share that Intel enjoys in servers but also includes the company’s sizable positions in storage controller and networking switch silicon, and its substantial efforts in memory and SSD storage. Finally, it would be a mistake to neglect Intel’s continuing investments in specialized yet complementary data center solutions, including its FPGA SOC technologies, Xeon Phi for applications requiring massive parallelism and vectorization and Intel Silicon Photonics optical networking offerings.

These and other Intel advancements have enabled the company to become the silicon vendor of choice for a wide range of commercial data center systems and applications. But Intel’s focus on innovation extends into considerably loftier, high performance and technical computing areas. For example, Intel Xeon and Xeon Phi power 464 of the world’s fastest supercomputers highlighted in the most recent list. The company and its OEMs are also deeply involved in artificial intelligence and related machine-learning and deep-learning development.

That poses significant challenges for AMD but what about other vendors? Two that come immediately to mind are IBM’s POWER platform and efforts around ARM-based data center silicon and related hardware. Of the pair, POWER is far readier for prime time, especially when one considers the efforts of the 300+ vendors/members of the OpenPOWER Foundation. That is likely to accelerate in the coming months with the launch of IBM’s new POWER9 technologies, along with strong assists from tens of thousands of POWER-focused ISVs, developers and services professionals.

Promoters of ARM-based data center silicon have long promised more than they delivered but that could change as players, including Applied Micro, Cavium and Qualcomm get into the fray. Focusing on customers that rely more on internal developers than commercial software, like telcos and cloud should help ARM vendors. However, that emphasis mirrors AMD’s tight focus on EPYC for servers and is unlikely to undermine Intel’s broader strategy.

Final analysis

This is not to suggest that Intel is unbeatable or that the company’s current leadership positions are permanent. The fact is that data center owners, like organizations and people in every other IT market, are constantly on the lookout for a competitive edge and the next, new thing that might provide it. So, it’s hardly surprising when emerging players and offerings garner a flurry of excited interest.

However, just as uniqueness doesn’t signal a knack for continuity, newness doesn’t always indicate long term sustainability. According to an Intel spokesperson, over the past decade the company has delivered a 41X improvement in single socket performance. That’s evidence of a dedication to innovation that has a tangible, positive impact on businesses and their customers. But at the same time, Intel’s other data center developments have also paid substantial dividends.

In other words, the new Xeon Scalable Processor family demonstrates how much Intel has achieved and how far it has come from being a leading maker of server chips to becoming an eminent builder of holistic, Xeon-based data center platforms. Some competitors will take a run at one or a handful of those areas, but it’s hard to think of any that could pretend to emulate, let alone surpass Intel’s continuing achievements.

© 2017 Pund-IT, Inc. All rights reserved.