A Survey Paper on Processor Workloads

Abstract

Processors are ubiquitous in today's world. Almost any electronic device contains a processor, and their computing power ranges from a few thousand to millions of instruction per second. Processors are built for many different purposes, and so their capabilities vary greatly. Even within a specific area, processors are different enough to make it difficult to compare them. Benchmarks have been developed to measure performance of processors and rank them among their peers. Much research has been conducted to design and develop benchmarks capable of reflecting a processor's performance.

This paper surveys benchmarks in five major processor families: general purpose processors, graphics processors, network processors, embedded processors, and processors in supercomputers. This paper also covers some basics about benchmarks.

Table of Contents

• 1. Introduction
• 2. Benchmark Background
  
  • 2.1 The Origin of the word Benchmark
  • 2.2 Types of Benchmarks
    
    • 2.2.1 Addition Instruction
    • 2.2.2 Instruction Mixes
    • 2.2.3 Kernels
    • 2.2.4 Synthetic Programs
    • 2.2.5 Application Benchmarks
  • 2.3 Benchmark Metrics
• 3. Benchmarks for Desktop Computers
  
  • 3.1 Whetstone
  • 3.2 Dhrystone
  • 3.3 Fhourstone
  • 3.4 SPEC CPU 2006
  • 3.5 SiSoft Sandra
• 4. Benchmarks for Supercomputers
  
  • 4.1 LINPACK
  • 4.2 LAPACK
  • 4.3 Lawrence Livermore Loops
  • 4.4 NAS Parallel Benchmarks
• 5. Benchmarks for Graphic Processors
  
  • 5.1 3Dmark06
  • 5.2 SPEC viewperf 9.0
  • 5.3 3D games
• 6. Benchmarks for Network Processors
  
  • 6.1 CommBench
  • 6.2 PacketBench
• 7. Benchmarks for Embedded Processors
  
  • 7.1 EEMBC Benchmarks
• 8. Summary
• 9. References
• 10. List of Acronyms