Hi Dave B; You wrote: "Looks like a couple % points. Basically a dead heat." with regard to the difference between PC800 RDRAM and PC133-CL2 SDRAM.
Wow! You mean I can pay 3x as much for my RDRAM memory, and it will only be a couple percent slower than a PC133 system? What a joke!
It turns out that with the current processors, a couple percent difference in memory timing is about all the difference you are going to get. The difference between PC100 CL3, PC133 CL2 and PC133 CL3, for instance, while 33% in frequency, is only a few percent in typical benchmarks. See, for instance:
anandtech.com
In other words, big differences in memory result in very little perceived difference to the user. That's the way it is. [Why is this way, you ask? Very simple. Why should the engineers who 5 years ago speced the FSBs currently in use have speced them to allow for high bandwidth memory that wasn't then available? All things get improved slowly, and at the same time. No reason to put more effort into a subsystem of system than will be needed by the overall system, in its time. Now it is a new time, and new FSBs have been speced. In other words, just the usual march of technology.]
RDRAM just isn't that big of a deal, nor is PC133-CL2. So why should the motherboard makers go to great effort to make RDRAM compatible motherboards when you can get about the same performance out of the run of the mill?
You then wrote: "So it appears that the Rambus chip-to-chip interface makes PC100 memory (which is the core of PC800 RDRAM) perform about as well as PC133 SDRAM (with a 33% faster clock). Not too shabby to get the performance of the next step up in speed. When they move to a 133Mhz core in RDRAM (as Samsung shows on their roadmap), it appears we'll get the performance of 177Mhz SDRAM.
It is true that as of today, PC133 is the high bin for SDRAM. But it is also true that at this time, PC800 is the high bin for RDRAM. In fact, PC800 was until quite recently a very rare result. In any case, PC133-CL2 is already half the price of even PC-600 RDRAM. So where's the big price reduction? RDRAM needs to be no more than about 50% more than SDRAM in order for design engineers to consider it for even niche applications. For the mainstream to use RDRAM, the total system costs need to get within 5%. But RDRAM makes for more expensive motherboards, so that means RDRAM, to survive as a mainstream memory, needs to get cheaper than SDRAM. Not going to happen. No way. Rambus shareholders should think about DDR and SDRAM royalties instead of dreaming of a resurgence of RDRAM.
I mean really. Intel won't have the P4's PC133 chipset out until 3Q01. Don't you realize that this schedule implies that Intel has come to the conclusion that as of 2H01, RDRAM will still be an inexpensive, rare, hard to use, and general pain in the butt boutique memory? That's 12 months from now! What more do you have to see in order to realize that Rambus' dreams of having RDRAM as the mainstream memory are dead, dead, dead? What would it take? Does Sony have to announce a DDR based playstation? How's about if Sun announces that they are porting the MAJC over to DDR? Come on, help me here, what would it take? This technology is so dead that it is rotting, and beginning to stink. After what we've seen the last six moths, a designer would have to be nuts to put RDRAM into price competitive designs for production over the next few years.
Sure Samsung intends on making RDRAM at a 5% cost adder (in 2002, as you neglected to mention), but that doesn't mean that the price to the consumer will be only that much higher. First, RDRAM currently carries a higher royalty cost than DDR or SDRAM. (This is something that may change, but, then again, it may not. In any case, it is a relatively small percentage of the total cost.) Second, RIMMs are quite a bit more expensive than DIMMs. Third, Samsung has quite a monopoly going. The other memory makers can't make RDRAM as cheaply as Samsung, and right now, they don't look like they are exactly falling over themselves to burn the research dollars required to beat Samsung. So Samsung is very likely to continue to charge high prices for RDRAM, and higher prices for RIMMs. This is just good business sense. Fourth, it is now clear that DDR will outship RDRAM in 2000 by around 2 to 1 in terms of bits shipped, and will be out shipping RDRAM by 3 to 1 at the end of the year. (I.e. DDR is ramping faster.) This means that DDR is now the lead technology for the next mainstream memory interface, not RDRAM. This means that DDR will always be cheaper than RDRAM. This means that RDRAM will never, ever catch up. The race is over. DDR won, dude, get used to it. Since DDR will be produced in much higher volume, it will also be priced (to the customer) much cheaper. Sure Samsung may get their cost of manufacturing RDRAM to within 5%, but the customer (and that term includes Dell), is not going to get RIMMs within 20% in price of DIMMs.
Continuing, you wrote: "Also, in the section on Tomorrow's Trends, the line about DRAM speed not mattering is changed to "DRAM speed will be key for system performance as well as CPU/Chipset/FSB!" Oops, Carl."
Let's think about this. Current (Intel) processors are CPU or FSB bound, or both. When they improve the FSB speeds, it is likely that memory speed will have more of an impact on system performance. Right now, PC133 is "like a couple % points" faster than RDRAM. (Your own words.) PC133 CL2 is about to be supplanted by PC266 CL2 DDR SDRAM, which is going to be much faster.
Okay, now that we have the facts, we can look at the consequences. When Intel goes to a faster FSB, the PC133 CL2 memory systems are going to pull even more ahead of the RDRAM systems. Understand? Making the memory system more important to system performance is going to make the differences between memory systems more important, but PC133 is faster than RDRAM. And PC266 is faster than PC133, so those DDR systems are really going to crush RDRAM in the P4 and dual Athlon systems. It is not going to be a matter of a couple percent, it is not going to be close, it is going to be obvious to everybody that you will want your P4 hooked up with DDR. Meanwhile, what does RDRAM have to improve memory speeds with? Samsung wants to go from 800MHz to 1.06GHz around the end of 2001. Too little, and way, way, way too late.
The RDRAM story is broken now, and it is no longer possible to deny it. Go pray for DDR royalties.
-- Carl |
