It has been a buyer’s market for premium DDR2 memory lately. Your choices range from modestly overclockable modules, which command only a slight premium over value memory, to bleeding-edge modules with heat spreaders that can hit clock speeds as high as 533 MHz (DDR2-1066) and require voltages higher than 1.8 volts. Recently, we’ve seen DDR2 modules rated to use as much as 2.3 volts from Corsair.

What are the advantages of premium RAM memory? High-end DRAM chips are bin-sorted, a term used by manufacturers to refer to sorting chips that can run at different speeds into "speed bins" which hit high frequencies and voltages. These chips can run at lower latency timing than standard DRAM. Finding exactly the right timing settings can be daunting. It’s often a trial-and-error process that involves many test, reboot, and reset cycles.

Most standard or value-grade memory won’t run at the lower latencies at which premium-grade memory runs. For DDR2, that currently means latencies lower than the standard 5-5-5-15 timings.

If DDR2-800 5-5-5-15 doesn’t make sense, then read on to learn how to decode it and why it should matter to you. But first let’s clarify how memory works.

Memory accesses don’t happen in one step. Memory is laid out on a chip in rows and columns, which require repeated pulses of electricity, referred to as "strobing," to reach each location. When memory is accessed, each cycle of strobing takes a fixed amount of time, explained as follows:
• tCL - Column address strobe (CAS) latency; or the number of clock cycles required to access a specific column of data. (The initial t refers to time.)
• tRCD - Row address strobe (RAS)-to-CAS delay; or the number of clock cycles needed between a row address strobe and a column address strobe.
• tRP - RAS precharge; or the number of clock cycles needed to close one row of memory and open another.
• tRAS - The number of clock cycles needed to access data in a specific row of RAM.

Now let’s break down the particular DRAM label above. 800 is the effective clock speed in megahertz. That’s the actual clock speed multiplied times data per clock cycle (200 MHz [for DDR2-800] X 4 [4 samples for DDR2 per clock cycle]). DDR2-800 has a maximum bandwidth of 6.4 GBps. "5-5-5-15" is referring to a tCL of 5, tRCD of 5, tRP of 5, and tRAS of 15.

Because latency is measured in clock cycles, the smaller the numbers, the better. Therefore less time is required for memory accesses. The time is measured in nanoseconds, with a typical system making millions of memory accesses every second. Latency and memory speed trade off with each other. For example, the same DDR2-667 memory module can run at 333 MHz with latencies of 5-5-5-15, or at DDR2-533 speed at 266 MHz with latencies of 4-4-4-11. Since higher clock frequencies represent smaller time intervals, the total time is basically the same for both these speed settings.

The serious overclockers out there will want to push their memory speeds up to 1,100 MHz or more. Overclockers also need to run their memory at higher voltages, so it’s important to have a good motherboard with a beefy voltage regulator, and a BIOS setup that lets you tweak your memory settings. For an example of a motherboard that allows very high memory clockings, look up the eVGA nForce 650i Ultra motherboard.

What you get with premium memory is the ability to run at higher clock speeds and lower latencies. If your applications are sensitive to memory performance, premium memory can pay off. Games, 3G rendering, and media transcoding are all sensitive to memory latencies. Most user’s regular web browsing, office applications, and streaming media typically are less sensitive and will work just fine using standard RAM memory.