RAMBUS

Late this year and into next, the great RAMBUS/DDR debates will begin. This may bemuse many of you who thought RAMBUS was dead and buried everywhere except RAMBUS stock sites and Intel press releases, but no, RAMBUS is our Jason. 🙂

Here’s what you need to know about this in the upcoming months:

Practically none of you have bought RAMBUS because if you go to the store, it will cost you four or five times more than PC133 SDRAM. This is true. What is not generally realized is why it costs four or five times more. It’s not because it now costs four to five times more to make.
The real reason why RAMBUS is so expensive is because hardly any of it is being made, and just about all of it is going into Dell and other OEM computers. Memory manufacturers nowadays sell the vast majority of RAM to OEMs at fixed prices set at contracts usually lasting a few months.
Anything left over gets sold on the spot market, where the price is whatever the market will bear. If there is lots of RAM around, the spot price is usually less, sometimes much less, than what the OEMs pay for it. If there isn’t much RAM around, the spot price is more, sometimes much more, than the OEM contract price.
Do you remember RAM pricing doubling after the Taiwan earthquake? It wasn’t because Taiwan made all that much RAM, and the earthquake put that in peril. Rather, the earthquake happened just when people were buying memory for Christmas season, and since the big OEMs had first priority and wanted more for Christmas, too, there was the fear there wouldn’t be much RAM around for everyone else to build computers with. So prices doubled for a while. They increased, too, for the big OEMs, but nowhere near as much, about 25%.
As soon as it was seen that the earthquake didn’t hurt things too much in Taiwan, prices drifted downward.
With RAMBUS, just about all the supply is going to the OEMs at about $200 per 128Mb OEM stick. Yes, that’s still more expensive than SDRAM, and this price does roughly correspond to how much more RAMBUS now costs to make compared to SDRAM. But $200 is a lot less than $500. The RAMBUS folks are hoping to get the real cost differential down from 50% to 10% by next year.
A lot more RAMBUS will be made in the next couple quarters, but it will still not be much compared to the amount of SDRAM being made.
What you should be keeping your eye on is how next-generation RAMBUS compares to DDR. There are a lot of RAMBUS advocates out there who holler, “RAMBUS is here now! DDR is vaporware!” Well, the quad-pumped RDRAM that Intel is talking about putting into the Willamette motherboard is just as much vaporware at the moment as DDR main memory. That RDRAM is to current RDRAM as DDR is to PC133.
It’s one thing to say something is vaporware now. It’s another to act like if something isn’t around now, it never will be. I’m sure we’ll see both six months from now, and then the next battle begins. When you see comparisons, make sure they’re comparing apples to apples, not old stuff to new stuff.
Which will be better? I have no idea, and nobody else talking will for quite a while, either.
On the whole, memory manufacturers don’t like RAMBUS. It’s much easier and cheaper to convert current SDRAM capacity to DDR than it is to make RDRAM capacity. They don’t want to gamble spending a lot of money building capacity based on the hope that Intel will succeed this time around in imposing RDRAM as the standard. On the other hand, if Intel does manage it, they’ll lose a lot of profits while they’re trying to catch up.
So what do you do if you’re a memory manufacturer? You hedge your bets. You neither stake the company on RDRAM, nor ignore it. You build at least a little RAMBUS capacity so you’re not caught completely short-footed, and you wait to see what happens.
On the other hand, Intel very much wants to impose RDRAM as the standard. If they do, they’ll make billions of dollars in stock profits from their big investment in RAMBUS. That’s a very powerful incentive. On the other hand, Intel can’t sell any CPUs if they only run with RDRAM and not enough RDRAM is around. That’s a very powerful disincentive.
So what do you do if you’re Intel? Pretty much what you’re doing now, which is do everything to push RAMBUS for Willamette short of committing suicide. You make it sound like Intel is absolutely committed to RDRAM, but you never, ever burn your bridges.
If you read what Intel executives say, they’re very cute. They’ll say RAMBUS is better, that they don’t want DDR for Willamette. But they never say that Willamette won’t run with DDR. They can’t. Foster will be the server equivalent of Willamette. Its motherboards will be bus-compatible with Willamette, and it will use . . . . DDR.
Everybody is hedging their bets on this one. AMD will start off with DDR, but they have a RAMBUS license, and they’re hiring RAMBUS engineers, too.
This is a chicken-and-the-egg situation. Memory manufacturers won’t commit to RDRAM unless they have demand, but there won’t be demand unless there’s a reasonably priced supply available.
Intel couldn’t impose RDRAM with the 820/840 because there wasn’t any around. Even with demand a tiny fraction of the overall computer market, there isn’t enough right now.
When Intel gets serious about Willamette rollout in early 2001, though, we may have a different story. If there’s a reasonable supply at a somewhat reasonable price for the Willamette models, people will buy it. That probably will get the snowball rolling as everyone joins in and ramps up capacity as Willamette production ramps up over the course of 2001. RAMBUS use could then spread across the whole PC spectrum.
If there isn’t that reasonable supply at a somewhat reasonable price next year, though, we’ll have deja vu all over again.
So if you want a heads-up on what’s going to happen, try to find how many of these new-generation RDRAM sticks (not the individual memory chips, RDRAM advocates love to toss that number out and not tell you to divide by 8 to come up with the number of sticks) are being made towards the end of year, and how much capacity is out there. If you see around 3 million sticks being made a month come next February or March, and they don’t cost too much more than DDR, RDRAM has a very good chance. If that number is a million or less, RDRAM is in real trouble.