RDRAM - Explain please

manic111

manic111

manic111

Soldato
Joined
1 Jul 2007
Posts
21,941
Location
Various
Just a quick question really. I'm in the process of helping dad upgrade a really really old PC (Dell Dimension 8100) with a little more RAM. It happens to take RDRAM (yes, I know it costs the earth and isn't financially a particularly clever upgrade!), and I noticed that it runs at 800mhz.

Now, the DDR2 RAM in my current PC also runs at 800mhz (2x400mhz if you take the double data rate into account). I guess all I really want to know is why current DDR2 has only relatively recently surpassed 800mhz, and a couple of years back this speed of RAM was fairly unheard of, when my dad's 7 year old PC was running this speed back then?

Or is it a case of new architecture leading to more efficient, if "slower" chips, as has been the case with C2D versus P4?

EDIT - only just realized how "commanding" the thread title sounds - wasn't meant to, was meant to have a smiley face next to it. Sorry, here's your smiley! :D
 
Last edited:
It's a licensed tech, too expensive for many with few benefits at the time.

Intel began subsidising at one point, but stopped it after a while.
 
There was some stuff about it being fast at the time, but someone or something managing to compete with a slight tweak to SD RAM and therefore negating the expensive and complicated nonsense that is RD RAM. Or something.

I have no idea really :D
 
Funny enough I found an old mobo with some RDRAM on it the other day. I remember paying a fortune for some more of this stuff when I was upgrading the system.
Intel were backing it with the old socket 423 P4s but did a u-turn and dropped it for the cheaper DDR when it came along.
 
Rimms are serial (like SATA, USB and PCI Express), this allows very high clock speeds, and pretty good data transfer, but has a downside... latency. When transfering a very small amount of data (64bits for example), a parallel bus like DDR can transmit all 64bits in a single clock tick, but rimms only transfer 16bits at a time (it has 16 serial lanes, a bit like PCIe x16), so it takes 4 cycles to transmit the data.

In terms of bandwidth 800mhz rimms are about the same as 200DDR (100mhz), with higher latency than the DDR. 800DDR is 4 times higher bandwidth than 800mhz rimms and much lower latency.

Rimms were better than the old SD-Ram (although not by that much), but it never stood a chance against DDR which is superior in virtually every respect.

Parallel interfaces are difiicult to implement at high speeds, hence rambus's attempt to push a serial memory bus, however with carefull design, motherboards can implement a very fast parallel memory bus it just costs a little more, but the memory sticks themselves can use cheap low clock speep parts, which pretty much compensates for any additional cost of the parallel interface.

PS DDR is just an evolution of SD-Ram, which in term was an improvement over fast page ram etc etc, its not that far different from the ram in the earliest of PC's just a lot faster, Rimms/Rambus on the other hand was a revolutionary jump, but it was too expensive due to greed of rambus, and quickly beaten in performance by DDR anyway. Just like P4 was a revolutionary design compared to Pentium III, but at the end of the day, P3 was a better design, and thats why Core is an evolutionary upgrade of P3.. Netburst's been tossed out of the nearest airlock. (Not saying netburst will never return, it had some good concepts, but it hit issues with material technology, which at the moment cant be resolved, so intel went back to P3)
 
Last edited:
Its not the technology I had issue with, it was the company as a whole... lets sue everyone who uses DRAM just for the fun of it. I advise people not to use/be associated with anything that is RAMBUS, they are just a bunch of lawyers with no respect for engineers.
 
For the early generation P4 CPU's RDRAM was the only memory solution worth going for.
SDRAM could not provide the bandwidth that the CPU so badly wanted, RDRAM although expensive could.
There was no DDR based solution for the P4 at this stage - so it was either a cripled SDRAM based system or a flying RDRAM solution.

The first DDR chipsets for the P4 also couldn't offer the same bandwidth that RDRAM could.
It wasn't until the first dual-channel DDR based systems were available for the P4 that RDRAM was no longer the best option available.
 
Oh right, so the pricing etc is down to the fact that it's licensed to one company? Quite interesting all this, always nice to learn :D
 
I stuck with my P3 until DDR was available on P4. Rambus never resolved the latency issues, so even though the early DDR boards had slightly lower bandwidth than rimms, they also had lower latency (as long as you bought low latency 2-2-2-5 ram).

Single channel DDR333 wasnt that far off filling the demands of a 400mhz FSB P4, and by the time we had dual channel 400mhz DDR, paired up with the northwood 'c's 800FSB rimms were dead, they just couldnt keep up at any reasonable price.

Due to the price difference, you could often afford to have twice as much ram on a DDR based P4 compared to Rambus, so even with single channel, as long as you used PC2700 instead of the PC2100 the extra ram pretty much made the DDR system faster overall. In a money no object system Rambus was struggling to show a real advantage over single channel PC2700, and no advantage at all over single channel PC3200.
 
kleox64 said:
Its not the technology I had issue with, it was the company as a whole... lets sue everyone who uses DRAM just for the fun of it. I advise people not to use/be associated with anything that is RAMBUS, they are just a bunch of lawyers with no respect for engineers.
Click to expand...

Actually it was (probably ... court cases still proceeding) worse than this ... the allegation is that RAMBUS participated in the DDR2 standards process and "guided" the result into technologies where they had patents. This isn't unusual but the normally if you do this you're obliged then to license the patents at "reasonable" rates. However, RAMBUS dropped out of the standards committee before the point where (to avoid, I think, antitrust laws) they'd have to declare their patents and thus they claim they were not constrained against suing anyone who used the resulting standard ... meanwhile everyone else sees it as a clear "ambush" and are suing RAMBUS for antitrust. This has been running in the courts for about 10 years.

Meanwhile, collateral damage in the beginning was Intel who'd signed an agreement with RAMBUS which basically obliged them to only support RDRAM for a period of time around the launch of P4 ... this was one of the factors that opened the window of opportunity to AMD!
 
Duke said:
I remember RDRAM used to run very very hot.
Click to expand...

yeah, memory chips were only active one-at-a-time, and they got extremely hot, which is why they needed the heatspreaders. the inactive chips were heatsinks for the active ones :p
 
You must log in or register to reply here.
Back
Top Bottom