Bienvenido! - Willkommen! - Welcome!

Bitácora Técnica de Tux&Cía., Santa Cruz de la Sierra, BO
Bitácora Central: Tux&Cía.
Bitácora de Información Avanzada: Tux&Cía.-Información
May the source be with you!

Monday, June 16, 2008

Memory probs

Memory Access vs. CPU speed

How fast your memory can catch up with your CPU's speed in this GHz era?

Memory Access

Figure: Memory Access Speed Marks: Speed measured in MegaHertz (MHz). Hz=1/Second.

Bus is the data path in a computer. Nowadays, all the CPUs are measured in Gigahertz (GHz). Memories just can races to MHz. How do they work together?
CPU
CPU is the place where all the actual computing is done. The higher speed performs faster result in computing. However the rest of components (so far) can't catch up the speed with CPU due to the technology limitation. Chipset is actually governing how information travels between CPU and the other components. On above graphic, we example CPU clock is 133MHz to communicate with other components, but the CPU core speed is actually 2.65GHz (133MHz x 20 Core/Bus Ratio. Different CPU models have different
core/Bus ratio range, some are from 4 to 20, and some are from 2 to 15.
Front Side Bus
Chipset contains several "controllers" to govern different information. The controller which establishes the information flow between memory and CPU is North Bridge. Data speed between CPU and North Bridge use the CPU clock and be called Front Side Bus (FSB). FSB should equal to CPU clock, but nowadays we read/hear about 400MHz/533 FSB those are actually the efficiency by dual channels and double data rate. For example, the FSB is 133(MHz) x2 (dual channels) x (double data rate) = 533 MHz (virtual working efficiency).
Memory Speed and data speed
Memory speed also measured in memory clock. DDR (Double Data Rate) will twice the data transaction. Nowadays most of bus wide are 64 bits base that means the bus can path 8 Bytes data (64 bits/8 = 8 Bytes). How to calculate data the memory can carry through in one second? Taking an example of DDR 333 (which is actually 166MHz x 2) to multiply 8 Bytes (64bits' Bus) will get 2656MB/sec.
Memory performance comparing with FSB
If we calculate the data traveling speed between North Bridge and CPU, we have to use the efficiency of FBS to multiply bus wide's data. For example, the 133MHz CPU Clock (FSB is also 133MHz) which virtual working efficiency FSB between CPU and North Bridge is 533MHz (133MHz x 2 x2) multiplies 8 Bytes (64 Bits' bus) we will get 4264MB/sec. Comparing the data speed of DDR333 can carry through is still far behind the 533MHz FSB can take. So, there always has room to improve the speed of memory.

No comments: