Memory and Cache

• This is a combination of chapter 5 and B.9
• I will skip around a bit to accommodate the activity for 12/3
• Latency is the time between when a memory element receives a data request and when that data is available.
  • This is the minimal time to access data.
  • This is not a constant in all forms of memory
    • For electronic devices yes (ram, cache, ...)
    • For mechanical devices no (spinning hard disk drives)
  • In general, this is the MINIMUM time to get data
    • Larger chunks will take longer (transfer rates)
    • And other parts of the machine may reduce the speed.
      • Bus configuration
      • Bus contention
      • Load
• A block is a chunk of memory.
  • We will define this better later, but this will do for now.
  • BUT NOT as an answer on a test.
• To start
  • Registers operate (in fact partially determine) at the speed of the processor.
  • Cache is somewhat equivalent to a larger set of registers.
    • Built at different levels and different speeds.
    • For now, assume one layer.
    • And assume that it is at the speed of the memory portion of the pipeline (~200 ps)
  • Memory is much slower and much larger.
    • Again, many different technologies.
    • Assume that a block of memory can be transferred in 10ns
    • More on this later, but for now assume 10ns range.
  • Hard Drives are MUCH LARGER and MUCH MUCH slower
    • I am having some trouble finding real detail, but this page suggests in the 10-50μs range for personal SSD drives.
    • For HDD assume 2-10ms
  • You should read section 5.2, you really should.