Table of Contents
Purpose of RAM
SDRAM and a Problem
On one hand, we have Intel with RDRAM, an implementation that has been designed and marketed to go hand-in-hand with the Pentium4 processor. On the other hand, we have AMD and VIA with DDR technology, which is most commonly seen with AMD Athlon processors.
Both implementations solve the bandwidth problem, and have competitively co-existed for almost a year now. However, both are aiming to become the new standard in consumer level computer systems, and it is expected that only one will stand the test of time.
The New Standard of Performance
AMD and VIA have taken a very different approach with DDR technology. While RDRAM represents an entirely new memory technology, DDR-SDRAM is nothing more than an evolution of the existing SDRAM technology. Like RDRAM, DDR transfers data on both the rising and falling edges of a clock cycle. Its name is very representative of what it is — Double Data Rate SDRAM. Since it is otherwise nearly the same as SDRAM (same frequency, same bus width), it is easy math — DDR-SDRAM is capable of twice the bandwidth as the old SDRAM.
A second concern with RDRAM is its higher latency. Because it only operates on a 16-bit bus, the commands sent to the RAM are forced to be serialized. If you have 64 people trying to get through 16 doors, some people are going to have to wait in line, and that is exactly what we see with the commands sent to RDRAM. This is usually only a very small problem, but if a large series of small chunks of data are requested from RDRAM, its performance suffers. Since most applications benefit greatly from the high bandwidth, this has not yet become a large problem. The concern exists that this is a design flaw that will eventually come back and hit RDRAM.
Lastly, RDRAM suffers from high temperatures. Wonder why RDRAM is equipped with metal plating? Think of that as a low level heatsink. In fact, early designs of RDRAM called for active convection cooling, meaning a heatsink and fan for your ram! To work to solve this problem, the modules on RDRAM power down when not in use. In fact, only ONE module is powered up at any given time. The rest sit in standby mode, where they produce far less heat. The biggest drawback to this solution is that it takes nearly 100ns for a module in standby to power back up, which certainly does not help with the latency issue already present in RDRAM.
While I seem to have a lot to say about the weaknesses of RDRAM, DDR-SDRAM is not a perfect solution either. Because RDRAM is an entirely new technology, and DDR is a mere adaptation, it should not surprise us that RDRAM is much more efficient. While the old SDRAM was about 75% effective in utilizing the bus, DDR takes a drop and can only offer 65% efficiency. Compare this to RDRAM, which is at 85%, and we see a very poor efficiency! Luckily for DDR, this seems to be its only major flaw, and is often overlooked since DDR is still capable of the same effective bandwidth as RDRAM.