Over the years, computer speeds have been going up at a very rapid pace. For the past few years, the trend has been that computers double in speed after roughly 18 months. For RAM specifically, this is not true at all. Why? What is causing the RAM speeds to evolve at a much slower pace than the CPU, Cache or even CD-ROM drives? To answer this question, we have to take a look at the two main RAM performance factors: clock speed and latency.

This new RAM module has higher latency than some models from 3 years ago.
© 2001 by www.mushkin.com

RAM clock speed has kept pace with the Computer's main bus. They do not need to be any higher because if they do go any faster, they will be outpacing the different components that need to use it. Also, in order to improve clock speed greatly, some other factors suffer. Since too much clock speed does not necessarily give more performance, the memory will be even slower because other parts had to be simplified in order to accommodate the clock speed, and performance is reduced. Improving the clock rate by a huge amount has been attempted successfully by Rambus (their memory will be talked about later in the memory segment), and it has proven that it does not make any difference.

If it is not clock speed that slows something down, it has to be latency. Latency is the time that it takes for the memory to start sending data to other parts such as the CPU. The higher the latency, the lower the memory performance. Latency is very important, especially because information stored from the RAM is needed in many short bursts. This is because Cache data is checked first, and if the required data is not there, then the CPU checks in the RAM. Many times, the required data is in the Cache, because the Cache does store the most widely used information. Because there are so many short bursts, every time this short burst of data starts, the CPU has to wait for the data to be sent.

If data was sent in very long bursts, latency would not be much of a factor to memory performance. If you do not understand the concept of latency, think of this. Drag races are very short, sometimes as little as a quarter of a mile, or a third of a kilometre. It is therefore very important for the cars to accelerate very quickly, because chances are that they will already finish the race before they hit their top speed. This is why latency is important, because there are many short bursts of data, and long streams of data that are uncommon.

If a car was racing for 300 kilometres in a straight line, acceleration would not be important, top speed would be much more important. The same would be true if there were very long bursts of data, latency would not be a very big factor.

With the current way the computers are set up, latency is very important. Over the past few years, latency has become worse and worse, something which is very uncommon to computer evolution.

As you can see, BUS speed has gone up 8X from the 486 era. CPU speed is 60X larger than the 486 era. Latency is 10X worse than in the 486 era. To make up for the bad latency, engineers are making much higher RAM speeds, and adding a larger amount of memory. Think how fast memory would be if all this was done, and latency was as good as it was 5-10 years ago.