SURVEY OF HIGH-PERFORMANCE CPU ARCHITECTURES
Processor Size Feature Advantages
AMD Opteron 64 bits
  • Advanced HyperTransport
  • Peripheral interconnect
  • ccNUMA memory architecture
  • Arm ARM10 32 bits
  • Return-stack caching
  • Recognizes indirect branches using register 14 to eliminate pipeline flush on subroutine return
  • Broadcom BCM1400 64 bits
  • Quad 64-bit processing cores
  • Triple Advanced HyperTransport/SPI-4 links
  • Quad multiprocessing
  • Extensible ccNUMA architecture, high-speed serial packet support
  • Cyan eCog 16 bits
  • Lockable cache
  • Programmatic control over caching, allowing small routines to be maintained in cache
  • Cygnal C8051F120 8 bits
  • Prefetch buffer with 256-byte instruction cache
  • Single-cycle instruction access from flash memory
  • IBM PowerPC 64 bits
  • Copper and silicon-on-insulator technology
  • Low power, small die size, improved performance
  • Intel Itanium 64 bits
  • EPIC Explicitly Parallel Instruction Computing
  • Level 3 cache
  • Parallel instruction processing
  • Better support for data-intensive applications, such as database servers
  • Intel Xeon 32 bits
  • Hyper-threading
  • Dual processing power on one chip
  • Intel Pentium 4 32 bits
  • Hyper-pipelining
  • Execution trace cache
  • Rapid execution engines
  • Twice the length of prior Pentium pipelines, faster instruction and load/store execution
  • Stores decoded micro ops, eliminating decode delays for cached instructions
  • 1/2-cycle instruction execution times
  • MIPS MIPS64 64 bits
  • SIMD floating-point instructions
  • Accelerates processing of large data streams, eliminates the need for a separate DSP
  • Motorola HCS12 16 bits
  • Three-word instruction queue
  • Single-cycle instruction access from flash memory even during loops
  • Motorola PowerPC 32 bits
  • RapidIO interconnect
  • High-speed peripheral interface
  • SuperH SH-5 64 bits
  • Four-way integer SIMD
  • Accelerated multimedia processing
  • TI OMAP 32 bits
  • CPU/DSP dual processor
  • Each processor handles appropriate algorithm
  • Ubicom IP2202 8 bits
  • Memory-to-memory architecture
  • Efficient packet processing