Intel seems to be in a comfortable place in the processor market right now.
They do not appear to have much pressure from AMD or other processor manufacturers (as opposed to chipsets, where matters are more comptetitive). I think Intel will basically stand pat for a while, pricing and all, while taking care of the PIV optical shrink, and improving its production efficiencies for the fall.
However, you don’t necessarily have to make your processors faster to make your machines faster. You can do that by getting rid of bottlenecks in the machine, and that’s what I think Intel’s next big move is.
Memory in Intel machines is a different story. Intel is moving away from RDRAM/RAMBUS architecture in favor of DDR SDRAM. RDRAM is showing more and more flaws as speeds increase. Memory efficiency is dipping all the way down to 70% of theoretical maximums, while DDR is running comfortably in the 90% range.
The latest PIVs can use more memory bandwidth than DDR or even “old” RDRAM can provide. Cutting edge memory, whether DDR or RDRAM is still bleeding, as manufacturers are still unable to get good yields on DDR400/PC1066, keeping prices high and progress slow.
Intel knows this. They know that their PIVs are hungry for memory. They know the mass market will not understand the memory bottleneck and instead will blame the CPU. So they most likely will not release faster PIVs until there is cheap enough faster memory to support them.
Intel knows people have generally rejected RDRAM. They know single-channel DDR is insufficient for the PIVs current needs, and DDR-II is a couple years away.
So Intel and the other PIV chipset makers plan to squeeze two lemons together and make lemonade with dual channel DDR boards.
It looks like the first dual-channel DDR boards will support at least DDR333. DDR333 has a theoretical bandwidth of 2700Mb/sec. Multiply that by .93 (the level of efficiency we’re seeing today), and we get 2511Mb/sec. Not quite RDRAM standards.
Put two of them together though, and we’re looking at a maximum of 5022Mb/sec. To be sure, interlacing will lower efficiencies, but there’s a big cushion between 5000MB/sec and the 3300MB/sec PC1066 is providing. Additionally, we will see DDR400, and sooner than we’re likely to see PC-more-than-1066. Further down the road, there’s nothing to prevent dual DDR-II boards from keeping Prescotts and its successors fat and happy memory-wise for the next number of years.
So once dual DDR comes into place, it’s a relatively cheap way to keep the PIV happily fed for a while, and Intel can release faster PIVs without fear of slowness due to memory starvation.
RDRAM, on the other hand, has been running in dual-channel mode in PIVs all along. Quad-channel being impractical, the only path to increased speed is increased frequency, and RDRAM’s bus width presents a problem. It operates at a bus width of 16 bits compared to DDR’s 64bits. While that means fewer traces on the motherboard, the price for that is that RDRAM must operate at four times the frequency of DDR just to stay even with it.
Current PC800 dual-channel RDRAM runs twice as much RAM twice as fast to stay theoreticallly even with one stick of DDR400. If the DDR competition becomes two sticks of DDR400, any dual-channel RDRAM system would have to go from PC800 to PC1600 just to stay even. Don’t hold your breath for that.
RDRAM technology also is more expensive to implement. The memory requires more responsive chips (read expensive) and yields on PC1066 chips remain low. Motherboard require more expensive chips, too. The DRCG/ICS (RAM clock generator) chips must run faster to generate the high frequencies required by RDRAM. These chips cost more for the board makers to buy. They also produce more heat, which makes them poor overclockers.
Higher frequencies also require better-designed and shorter motherboard traces. This severely limits the location possibilities of the RAM slots on the Motherboard.
All in all, things look bleak for RAMBUS Memory.
I believe that Intel is backing the right horse with dual DDR. DDR memory has a bright and sunny future ahead of it; RDRAM doesn’t.
DDR is currently slower than RDRAM but that will change with Dual-DDR. If you are looking to buy a new motherboard and want the best technology that will last you a while, I believe that dual-channel DDR is the best choice.
Should you wait for Dual-DDR? Can you wait until October or November?