Off the Shelf or Custom Tailored Compute?

By Rob Hayes - Intel VP, Data Center Group, GM, Data Center Group Strategic Planning

Common off the shelf, general purpose computing solutions deliver increasingly faster performance and lower total cost of ownership (TCO) compared to fixed function appliances.  Intel continuously improves the performance per TCO of general purpose computing through our investments in Moore’s Law and we have “perfected” the concept of designing a great processor and manufacturing millions of them (per day!) exactly the same.  As more industries have been attracted to general purpose computing on Intel Architecture, the system requirements and workload attributes have grown more diverse, creating tension with the general purpose “one size fits all” model.

Historically, enterprise IT consumed the majority of server CPUs – driving a specific set of requirements.  More recently, developers of public cloud services, high-performance computing applications, and telco services also seek to benefit from the standard high-volume economics of Intel’s ecosystem and gain the agility of quickly deploying software-based services.  These new applications have differing requirements for performance, I/O, security, manageability, etc. making it difficult to serve all user needs with one product line.  The challenge is how to enable great performance for a range of workloads and the ability for customers to integrate their unique value into IA platforms without sacrificing the economic benefits of common off the shelf silicon.

To address this challenge, Intel has been building new capabilities to tailor our general purpose products into workload-optimized derivatives with various combinations of processor cores, accelerators, I/Os, and custom logic.  Some examples:

  • Cores:  Intel has designed a family of processor cores that are each optimized for different types of workloads – integer and floating point, scalar and vector, latency sensitive and high-throughput processing.
  • Accelerators:  We created a variety of workload accelerators to offload the CPU of routine functions like compression, encryption, graphics, media transcode, packet inspection and filtering, and others.
  • I/Os:  We offer a broad range of standard interfaces such as PCI-Express, Ethernet, SATA, DDR, USB, etc. as well as proprietary high-speed and memory-coherent interfaces for tight integration with the processor.
  • Custom logic:  For customers who want a custom-tailored product, we can integrate their unique value into custom CPUs and/or ASICs through our SoC design methodology and Intel Custom Foundry design tools and manufacturing services.

These capabilities allow Intel and our partners to leverage the IP building block designs and manufacturing processes from our general purpose product line to create workload-optimized and custom products on Intel Architecture that provide optimal performance for the desired application.  The best of both worlds.

This blog originally appeared on The Data Stack.