slide 4 of 14

General Comments

• Combination of host + attached processor is heterogeneous
• None of this is standardized
• Need to write "magic" machine-specific code
• Can get supercomputer performance on specific tasks
• Cost/performance can be very good
  • E.g., FPGA does genetic sequencing faster than high-end Cray
  • E.g., 64 PE ICE DRE delivers 7.7 GFLOPS for ~$120K