System Memory Specifications and What to Look For
Table of contents
Given a specified memory type, there are still tens or even hundreds of products from different manufacturers available. They are of course different, both in features and in price. Like many other computer products, some of these features may not be important to certain users depending on their applications and requirements, so it could be helpful to have some general understanding of the basic memory specifications to help you figure out which features really matter to you.
Generally speaking, the larger the capacity the more programs you will be able to run simultaneously (as long as your motherboard and operating system supports it). The capacity you need should depend on your requirements: For most home users, any more than 3GB of memory will not result in any performance gains – at least for now.
Memory speed is a little bit complicated as there are two measurements: operating frequency (or more accurately, the transfer or data rate) and bandwidth. Bandwidth can be regarded as how much data the memory can transfer at any one time. To illustrate, imagine cars on a highway. If the highway has several lanes, traffic moves faster. If the highway only has one lane for both directions, traffic can move very slowly. Higher bandwidth means more lanes for your data traffic.
SDRAM rated PC100 and PC133 work at 100MHz and 133MHz and provide 800MB/s and 1066MB/s bandwidth respectively. An easy way to determine bandwidth is to multiply the frequency of the RAM by how many bytes the memory can transfer in one clock cycle. SDRAM uses a 64-bit architecture. Eight bits equals one byte. Therefore, SDRAM transfers eight bytes per clock cycle. Multiply eight bytes by the RAM frequency and you get the bandwidth in megabytes.
Ex. SDRAM rated PC133 works at 133MHz x 8 bytes ~ 800MB/s
RDRAM utilizes the same rating system as SDRAM - a PC800 RDRAM operating at 800MHz, provides a bandwidth of 1600MB/s. The bandwidth isn’t calculated the same way, because RDRAM uses a 16-bit architecture, which yields two bytes instead of eight.
Ex. RDRAM rated PC800 works at 800MHz x 2 bytes = 1600MB/s
Things are different when it comes to DDR, DDR2 and DDR3 memory. Initially, DDR used the same rating system as SDRAM, e.g. PC266. This has now been changed to DDR266 instead, which still means the memory works at 266MHz, providing 2100MB/s bandwidth, which is where the PC2100 designation comes from. DDR400 memory, by the same rule, is called PC3200 for its 3200MB/s bandwidth. The same rule applies to DDR2 and DDR3 memory, for example, the DDR2 533 is also called PC2 4200 or PC2 4300 but 'PC2' is used here to refer to DDR2 instead. DDR3 1066 can be called PC3 8500. 'PC3' here also refers to DDR3.
The bandwidth we refer to here is for single channel scenarios. Dual channels are becoming industry standard for motherboards. When memory is used in dual channel mode, the bandwidth doubles - for instance, dual channel DDR2 800 provides 12800MB/s bandwidth as opposed to 6400MB/s for single channel DDR2 800.
DRAM stores data in a matrix of columns and rows just like a spreadsheet. If you need to find information in a particular cell, you look for the correct column and then the row. DRAM works the same way. It finds the information by sending signals to the memory in the forms of a Column Address Strobe (CAS) and a Row Address Strobe (RAS).
CAS Latency (or CL), it is the amount of time it takes between a CAS signal assertion and the initial transfer of the data stream. The CAS Latency is measured in clock cycles. For example, a CAS Latency of 2 or CL2 means the data is available 2 clock cycles after the CAS signal prompting. As with any latency parameters in the computer domain, a smaller CAS Latency value means better performance.
There are other memory latency parameters as well, such as tRCD (Row-to-Column Delay), tRP (RAS Precharge) and tRAS (minimum bank cycle time), these parameters affect memory performance as well, but generally not as much as CAS Latency. We often call all these latency parameters "timing" as well, and a "loose timing" means high latency parameters, in contrast to "tight timing".
There are many memory modules shipped in the form of multi-channel kits – these are essentially multiple identical memory modules in one package, one for each channel. This type of product is designed for motherboards/systems that support multi-channel mode, with two 64-bit wide channels to provide double the bandwidth of single-channel memory systems. Identical memory modules are preferred for multi-channel usage because this symmetric architecture causes less compatibility problems and delivers the highest performance. A dual channel kit product marked as "1GB (512MB x 2)" means it contains two identical 512MB memory modules.
Almost all manufacturers add heatspreaders to their high-end memory products for better cooling. The amount of surface area exposed to air is what really matters when you are trying to dissipate heat, and that is where the heatspreader comes into play – it simply spreads heat over a larger surface area than the memory chips, thus providing more efficient dissipation. This really makes sense for high-end products since cooling is always vital in high performance rigs.