As if trying to compare all the variations of one generation of memory weren't bad enough, there's no better example of the kind of uncertainty that exists than that created by AMD when they incorporated both DDR2 and DDR3 controllers into their AM3 processors. An AMD user can buy a new AM3 CPU and put it into their existing AM2+/DDR2 motherboard, or they can opt to upgrade their board to take advantage of faster DDR3 memory. But again, which is better, and will it make a noticeable difference?
In one of our recent memory reviews, the OCZ Platinum XTE PC3-16000 kit, we compared how memory running at different speeds and timings performed. Sure you can purchase two sets of hardware yourself and run both through an array of benchmarks, but it would be nice to know which is better before buying. Fortunately there is a relatively simple formula we can use to determine this.
In the chart below we list several common varieties of DDR, DDR2 and DDR3 memory. Before we get into comparing the differences, let's explain some of the column headings.
The Name / Rate is the specification name assigned to a memory module by its manufacturer. The number corresponds with the theoretical bandwidth speed in MB/s that the memory bus should be capable of. This is often rounded to the nearest hundreds, and may change from manufacturer to manufacturer. For example, one brand of PC3-10666 may be the equivalent of another brand of PC3-10600.
The JEDEC specification is the standard formalized by the Joint Electron Device Engineering Council to ensure compatibility across manufacturers. Not all advertised speeds have been ratified by the council; some ratings are advertised by manufacturers to indicate overclocked memory that exceeds the actual specification.
The memory frequency is self-explanatory. This is the clock speed, measured in Mega Hertz, that the memory is capable of operating at. Due to the Double Data Rate nature of DDR SDRAM, data can be carried on both the rising and falling cycles of the clock signal. Thus the JEDEC spec is rated at twice the actual operating frequency.
The clock cycle is not typically advertised. It is calculated simply by taking the reciprocal of the clock speed. For example, 1 / 100 MHz clock speed = 10ns clock cycle. The lower the number, the less time each cycle takes to complete.
The CL or CAS Latency is typically the primary timing value advertised by the manufacturer, and represents the number of cycles the memory controller has to wait from the time it sends a signal to the module until the data is ready to send back. Although other timings such as TRAS, TRCD, etc. are also important, they often match the CL, which is the first number listed in the timing series. For example, 8-9-9-24 would indicate a CL of 8 cycles.
The Latency is the result we get when multiplying the Cycle time by the number of CL cycles. This number can then be used to compare memory modules of varying speed and timings to see how much latency is inherent at the given specification.
In this way we can determine for example, that DDR3-1600 CL7 memory is not quite as fast as DDR3-1866 CL8, but is actually slightly faster than DDR3-2000 CL9 due to its lower overall latency.
We can also compare modules from two different generations, such as DDR2-1066 CL5 and DDR3-1600 CL8, in which case the DDR2 memory is actually a bit faster. If you'd like to compare a module that is not listed, simply calculate the latency as described above using the memory operating frequency and number of CAS cycles.
Will you notice a difference between 9ns and 10ns? No, absolutely not. It would barely show up in benchmarks. But this can be a useful tool if you're considering whether to spend extra money on higher speed, slower timing memory. In the listed examples above, the latencies are all within +/-1½ns of each other. It would make more sense to go with whatever option is the least expensive.
You can also see that with modern DDR3 memory, most of the higher speeds are all within 8-9ns of latency, the difference of which is indiscernible to human perception. As memory speeds continue to get faster and faster, the variation between the different speeds and timings makes less and less of a difference.
Hopefully this chart will provide some insight and de-mystify the concept of speed and timing selection when choosing memory for your system. Obviously this is not meant to be the end-all, be-all guide for memory; there are other timing values besides CL to be concerned with. But it provides an easy method for comparing two different sets of memory.
