The first is that they are supporting DDR-2 , which requires a totally new design (Rev. F vs. the old/trustworthy Rev. E featured on 939/DDR1 platforms). The second is that AMD is trying its hardest to get Fab 36 up and running with 65nm, so it can have a somewhat suitable answer for Intel's upcoming monster, the Conroe family of products. The third is that the Opteron series of chips is also going to go DDR-2, not to mention a socket change in the very near future.
When all of these factors are put together, it is starting to look like AMD is mightily production constrained. Not only does the firm have to support the older Rev. E. products, but it has to go over to the new Rev. F models, not to mention make more space in their current Fabs to switch over to 65nm production. And let us not forget that AMD will start to work on making Fab 30 into Fab 38, which will feature 65nm and 45nm production on 300mm wafers.
Let us throw another wooden shoe into the factory cogs. The Rev. F cores are approximately 20 per cent larger than their earlier Rev. E brethren. When considering such things as lower yields due to larger die sizes (greater chance of defects per centimetre squared), not to mention more dies being lost around the edges due to chips being square on a round wafer, AMD is looking at between 30 and 40 per cent fewer good dies per wafer, depending on the maturity of the yields.
Let us further the cause by setting said wooden shoes on fire before throwing them into the works. AMD is also coming out with its new Energy Efficient products, which will also take up production space. These EE products, thanks to AMD's APM software, will have their mixes during fabrication specialized so they are enhanced over regular processors when it comes to power consumption and heat production. This again will take up space and possibly negatively affect overall yields.
Opterons continue to gain popularity in the server space, and the change to the Socket F architecture and DDR2 will enhance this even more. Opterons sport 1MB L2 cache per core. So, if AMD is already production constrained, and it expects Opterons to get another boost from this change, where are the dies going to come from? Since a Rev. F 1MB L2 is the same from both the desktop perspective and the server market, AMD is chopping off the 1MB L2 desktop cores at the knees and shuffling all 1MB L2 production to Opteron lines. This, of course, does not affect the FX-62, but those are a high margin/low demand chip, so they do not negatively impact Opteron sales.
So, to maximize profits without impacting production to a great degree, AMD has felt it necessary to get rid of the lower margin desktop 1MB L2 cores. This change affects both Rev. E and Rev. F production, as all 1MB L2 cores will be allocated to the Opteron lines. This is not a huge deal, as the performance differences in the vast majority of applications did not overtly show the benefit of the extra L2 cache in a desktop setting. AMD is able to pare down its product offering for the desktop, and keep from confusing consumers about the differences between say a X2 4200 and a X2 4400.
Keeping the 512KB L2 cores for the desktop will help to relieve production pressure. µ
Sign up for INQbot – a weekly roundup of the best from the INQ