This article is more than 1 year old

Wham, bam, thank you RAM

Reg reader on the job

AMD and Intel - who said what?

We're obliged to Charlie Demerjian for this mini-thesis on RDRAM vs DDRRAM. I suggest you make a nice cup of tea before wading in:

I assume you already know this, but in case you have ANY doubts, here is the answer for the reader who wrote in about the RDRAM vs DDRRAM.

Anyways, lets go on to the technical nitty gritty. PC800 rambus runs at 400MHz, double pumped, on a 16 bit (2 byte) wide bus. Overall, the bandwidth is 400MHz*2*2bytes, or 1600MBps, or 1.6GBps. PC700 and PC600 are slower, although neither runs at the speed the names would lead you to believe. Toms Hardware (www.tomshardware) had two excellent articles on rambus a while ago. I highly recommend that anyone even vaguely interested or confused read these, they are a good starting point.

DDR runs at 100 or 133MHz, double pumped, 64 bits (or 8 bytes) wide. 100MHz*2*8bytes=1600MBps, whole 133MHz*2*8bytes=2100MBps. The magic is that DDR transfers 4 times as much information per clock, while rambus runs at 4 times the clock rate.

On paper this is an even match, but there are several things that skew the arguement in favor of DDR. The first is cost. Go to www.pricewatch.com and look around. Since we are not argueing cost here, I will let this one slide for now. Next is latency. The time it takes to read a single byte at random is MUCH higher on a rambus system than on a DDR system. This is a fundamental flaw of rambus. The P4 can hide it fairly well, but you can apply those same trics to DDR is you so choose.

Because DDR's latency is fairly low, there hasn't been much call to do so. All things being equal, in random reads, DDR will most likely beat rambus. After latency comes pin count. Rambus wins this one handily. The number of pins needed to implement a rambus channel is about 1/4 of the pins needed to run a DDR channel. The more pins, the more money it costs to design and manufacture the motherboard. This makes rambus mobos cheaper right? Well, that leads us to the last point, which is speed (although a different speed than the numbers above). Since rambus runs its memory at 3 to 4 times the speed of DDR (400 vs 100 or 133), it is much harder to design. Much MUCH harder. Ask the people who designed the intel i820 chipsets, or the motherboards that use it (*cough* 2 rimms per channel *cough*).

When a company like intel repeatedly recalls, delays, and eventually kills of thier flagship product, you have to wonder about that product.. Intel, for all the run I make of it, probably has one of the finest and deepest pools of engineering talent available in the semiconductor industry. If they can't figure it out, there have to be problems. DDR motherboards are much less of a problem, although still not easy.

Overall, DDR has more bandwidth, less latency, and is easier to design for. The only way rambus is faster is because some motherboards use 2 channels (i840, i850) for twice the raw bandwidth. If you implemented a dual DDR chipset, it would be MUCH faster than a rambus mobo (single or dual)

More about

TIP US OFF

Send us news


Other stories you might like