Implications for Investing in a New Microprocessor: Essential Checklist

The Total Cost of Ownership for AI Workloads

White Paper Details

Pages: 8
Release Date: 3Q 2019


Register or Log In to download the free white paper.

As new innovative microprocessors are being brought into the market, one of the biggest drivers is artificial intelligence (AI), especially machine learning (ML) workloads, but other AI use cases exist as well, from processing traditional analytics applications on big data to high performance computing (HPC). When enterprises are evaluating whether to make use of a new chip, they are often making considerable investment decisions. For example, a cloud-based product or service with a global audience may require millions of data processing operations per hour. Providing the right processor in the data center is a major undertaking in terms of costs, resources, and effort.

This Ovum white paper, written in partnership with Tractica, is designed to help users of such chips or those considering the need for such chips, when making large investment decisions for supplying data centers or massively distributed edge devices and to give pointers on what should be evaluated beyond basic performance statistics.

Key Questions Addressed:

  • What factors should be considered when determining whether a new microprocessor is needed for AI workloads?
  • What are some of the conflicting requirements for AI workloads?
  • What are the top 11 criteria upon which a new microprocessor should be assessed?

Who Needs This Report?

  • Semiconductor and component manufacturers
  • Service providers and systems integrators
  • End-user organizations deploying AI systems
  • Industry associations
  • Investor community

Table of Contents

  1. Ovum View
    1. Summary
    2. Recommendations for the enterprise
    3. The complex choices to be traded off when choosing a microprocessor
    4. What to look for in a new chip: detailed criteria
      1. Performance statistics
      2. The microprocessor’s instruction set architecture
      3. Availability of benchmarks
      4. AI use cases
      5. Maturity of software stack
      6. Power requirements
      7. Memory efficiency
      8. Chip connectivity
      9. Algorithm change cycle speed
      10. Security of chip
      11. Total cost of ownership
  2.  Appendix
    1. Further reading
    2. Authors