Dual-Channel Architecture

1
Dual-Channel Architecture

• Brad Traina
• April 7, 2009

2
Memory Hierarchy
3
Physical Memory

• Physical memory, also known as RAM, stores data
  for use by the CPU
• There are many types of physical memory, the most
  common used today being SDRAM

4
Random Access Memory

• Random Access Memory (RAM) stored data is to be
  accessed at random
• Any piece of data can be returned at a constant
  time, regardless of its physical location
• RAM is volatile, meaning all information is lost
  when the power is shut off

5
Who Invented RAM?

• In 1966, Robert Dennard invented DRAM (dynamic
  random access memory)
• He received a patent for his idea in 1968.

6
Computer Memory

• A computer's memory (RAM) is a temporary storage
  area for data that needs to be accessed at high
  speeds.
• This can help programs run more efficiently.
• The processor can only work as fast as data can
  be passed to it from the CPU.
• If the CPU is waiting on data from the memory, it
  is not working at its full potential. The memory
  needs to keep up with the CPU.

7
Single-Channel Memory

• Single-Channel memory is like a funnel that feeds
  data to the processor through a single pipe
• Data is transferred 64 bits at a time
• This can create a bottleneck

8
Dual-Channel Memory

• With dual-channel memory it is like having two
  funnels feeding data through two pipes
• With two channels, data can be transferred 128
  bits at a time

9
Memory Controller

• The memory controller manages the data flow
  between the processor and the memory
• Data gets sent to the memory controller which
  decides whether to store it in memory or send it
  to the processor

10
Memory Controller
11
Single-Channel Memory

• Single Channel memory can be though of as a one
  lane street.
• Data needs to flow both in and out of memory on
  this one lane street.

12
Dual-Channel Memory

• Dual-Channel memory doubles the maximum amount of
  data that can be transferred between the CPU and
  the memory
• We now have a two lane road capable of sending
  and receiving data at the same time.

13
Peak Bandwidth

• Peak bandwidth describes the maximum amount of
  data processed by memory in a given time period.
• It is calculated by the following formula
• Memory Speed Number of bytes transferred per
  channel Number of channels

14
Performance

• Notice that these transfer rates are theoretical.
• They are calculated assuming that a data transfer
  will occur each clock cycle.
• In reality, no CPU or memory controller will be
  transferring data 100 of the time.
• For this reason 100 theoretical performance
  increase does not translate into a 100
  performance increase of your computer.

15
Synchronous Dual-Channel

• For the best dual-channel memory performance you
  must use identically paired memory
• In order for memory to be considered Identical it
  must have
• The same capacity
• The same speed
• The same latency
• This setup is referred to as synchronous
  dual-channel. It is the standard when you think
  of dual-channel memory.

16
Asynchronous dual-channel

• Asynchronous dual-channel memory is quite new,
  and provides less performance gain when compared
  to synchronous dual-channel
• In order to use this you must have memory
• With the same speed
• With the same latency
• In this case the capacity of the memory does not
  matter

17
Triple-Channel Memory

• The new intel core i7 processor supports triple
  channel memory.
• The memory controller is now built into the
  processor.
• It is capable of transferring 192 bits at a time

18
End
19
Resources

• http//en.wikipedia.org/wiki/Dual-channel_architec
  ture
• Everything You Need to Know About Dual Channel
• By Gabriel Torres
• http//www.hardwaresecrets.com/article/133/1
• http//www.devhardware.com/c/a/Memory/Dual-Channel
  /
• Intel Dual-Channel DDR Memory Architecture White
  Paper
• By Infineon Technologies and Kingston Technology
• http//www.kingston.com/newtech/MKF_520DDRWhitepap
  er.pdf