We know Intel has a good deal of sober conservatism about itself in respect to memory: since PC133 SDRAM, we can’t remember a standard Intel would support in its chipsets first of all other companies. PC1600, PC2100, PC2700 – they all appeared in Intel’s chipsets with a certain time lag. This was presented as an advantage, and quite reasonably: we all know that the company spends a lot of its own time and effort to polish off and improve specifications for stable work of the end-system. Intel also conscientiously approaches the certification of memory modules to meet these requirements.

Still, we shouldn’t forget that Intel has those leader-of-the-industry airs and thinks it can single-handedly choose the way the industry must go. After the Rambus affair, Intel kept low for a while, but time has passed and PC3200 became the next standard the company exerted its muscle upon. Intel supposed that DDR II memory type was going to enter the market soon, somewhere at the end of 2003, and intended to put its stake on it.

Of course, Intel didn’t bother about PC3200 DDR at all: why get busy with it in such a situation? Even the next increase of the system bus frequency (667MHz FSB) was intended to match the 5.4GB/s bandwidth of dual-channel PC2700 DDR.

So, Intel spent the whole year 2002 arguing that PC3200 was no viable standard: it had problems with stability, no one needed such a bandwidth and so on and so forth. Traditionally, VIA and SiS opposed Intel, but not with loud meetings as in the times of the battle over RDRAM. They just built PC3200 support into their chipsets.

The support was unofficial, though, as the industry was much of a mess then: Intel was strongly against PC3200 and JEDEC itself was doubtful, whether it was appropriate to polish the specification into a standard. Thus, there were no official unified parameters for PC3200 modules: memory modules makers as well as chipsets and mainboards manufacturers were groping their way in the dark. It even appeared that mainboard manufacturers had to certify individually specific memory modules for specific mainboard models.

Another thing adding to the overall picture was that not all companies had equal technical facilities at the beginning-middle of 2002. Some of them, even such majors as Micron, just didn’t have a manufacturing technology ready to produce acceptable chip yields. Of course, it was not within their interests that JEDEC would ratify the standard and allow their more technologically advanced competitors to profit this way. As a result, the PC3200 standardization process was sabotaged and delayed in many ways. Moreover, there were problems with timings, which might lead to a situation when the maximum working PC3200 configuration would consist of two one-sided DIMM modules (compare with the ordinary configuration of three double-side modules).

Moreover, we can’t help agreeing that Intel was right then: there really was no need in higher memory bandwidth than the bandwidth of PC2700. The system bus bandwidth of Pentium 4 (533MHz FSB) was then equal to 4.3GB/s, which was clearly above the capabilities of single-channel PC3200, but got fully covered by dual-channel PC2100. Athlon XP processor, even if we take Barton with 333MHz FSB, was also quite OK with PC2700. So, from this point of view, PC3200 did seem unnecessary.