AMD CPU head to head - AM2 vs S939!
Author: Grga Curkovic
Date: 23 May 2006

Today AMD launched a new socket, the AM2 and a new line of AM2 processors. The difference between these and the "old" CPUs is the memory controller, which in the new processors works with DDR2 memory, while in the "oldies" works with DDR1 memory. The rest of the CPU is pretty much the same. Intel started using DDR2 memory a while ago, but at the time it really didn?t seem like DDR2 had a performance advantage over DDR1 so nobody was worried about the fact that AMD was still using DDR1. The general opinion was that DDR2 performance will match DDR1 performance once DDR2 reaches PC2-6400 speeds (400MHz with two bits transferred per clock, often incorrectly described as 800MHz memory), but we had some nice results with DDR2-667 modules. Last year we saw a big war between memory module makers and records were being broken almost every week. That was a clear signal that DDR1?s time has passed. Lately we saw modules going as high as DDR2-1100, while almost all memory makers introduced DDR2-1000 modules. With both leading CPU makers using DDR2 memory we can expect more affordable prices, bigger volumes and higher availability, which is good for the end user.

To really understand what DDR2 brings over DDR1 we need to know how memory actually works. Although most computer enthusiasts know how memories work, we decided to scratch the surface a bit and let an average user know what he?s buying for his money. All the memories we use as system memories today (and those include SDRAM, DDR1, DDR2, DDR3, RAMBUS?) are DRAMs (DRAM stands for Dynamic Random Access Memory). The most interesting part of this abbreviation is the "Dynamic" which means these memories need to be refreshed periodically and the time they are able to hold information is not infinite. The time a cell is able to keep information is measured in nanoseconds, and the cell needs to be refreshed periodically to keep the information. The basic cell that can store one bit of information consists of a transistor and a capacitor. The charge in the capacitor stores the information while the transistor is used as a switch that let?s us read or store information in the capacitor.

Due to many electric effects the capacitor looses charge, and that?s why it is needed to periodically refresh information. Individual cells are organized into rows and columns. The memory bus consists of two basic parts, the address bus and the data bus. Every time memory is being accessed we need to set the address we are accessing on the address bus, then let the memory know weather we want to read or write information and the read or set the information from/to the data bus. Every of these steps takes some time and that is what latency is. For practical reasons latencies are measured in number of clocks needed for certain actions, so latencies that are printed on memory modules are relative to the memory frequency. The time (in seconds) needed for 2 clocks at 200MHz is the same as 3 clocks at 300MHz, so if you have a DDR-400 memory capable of CAS 2 (Column Access Strobe) and a DDR-600 memory capable of CAS 3, the elapsed time for both will be the same thus latency measured in seconds will be the same.