RESEARCH BRIEF: Moving Beyond CPUs in the Cloud: Will FPGAs Sink or Swim?

General-purpose server processors are reaching “diminishing returns” limits, as performance-per-watt improvements slow and workloads become more specialized. Certain workload classes are open to acceleration by compute offload or alternative (non-CPU) architectures including digital signal processors (DSP), graphics processing units (GPU), field programmable gate arrays (FPGA), and custom logic. While these accelerators historically have been attached to CPUs via offload interconnects, they increasingly are being integrated onto system-on-chip (SoC) designs. As these technologies mature, Moor Insights & Strategy believes that datacenter workloads deployed at scale will use application-specific acceleration models.

Table of Contents

  • Executive Summary
  • Application Acceleration in the Datacenter
  • Are FPGAs in the Datacenter Ready for Prime Time?
  • Key Players to Watch
  • Call to Action
  • Figure 1: Application Acceleration Implementation Models

You can download the paper here.

Companies Cited

  • Altera
  • ARM
  • Baidu
  • Bing
  • Catapult Systems
  • Convey Systems
  • Dell
  • HP
  • IBM
  • Intel
  • Microsoft
  • OpenPOWER Foundation
  • SRC Computers
  • Xilinx