After speaking with a large cross section of motherboard and memory manufacturers, the general consensus at this year’s show is that, initially, DDR3 is going to be too expensive for most enthusiasts.
In a dismal state of affairs, very few memory manufacturers have samples and even fewer motherboard manufacturers have working demonstrations. Given that Bearlake (the first chipset to support DDR3) is coming out very soon, we would have expected a more prominent DDR3 presence at the show.
It seems that DDR3 will hit the channel with the effect of a mosquito to a train. At the moment, there are only two DDR3 chipsets due, Intel’s P35 and X38, and both of these are upper-mid to high-end chipsets that won’t have a huge market penetration.
DDR2 has just recently ramped up and overtaken DDR on the sales front, as users upgrade from old AMD systems and even older Intel systems. This means that DDR2 is now finally affordable and there are performance improvements over DDR.
Add all of this to the fact that making DDR3 chips is not just an update of DDR2; instead, it’s a whole new manufacturing process and that means prices are going to be sky high.
On the motherboard front, the Bearlake chipset will still roll out, but most manufacturers are going to be shipping a DDR2 only model, DDR2/3 combo and DDR3 only. Having only one type of memory onboard allows the board makers to tune the system very well, especially for overclocking, but a combo board allows for more choice.
We talked to Elixir, a company that produces Nanya memory chips, and it had a very sorry tale to tell us about manufacturing DDR3. Firstly it needs to be fabbed on a 70nm process – this isn’t a simple optical shrink like moving from 90nm to 80nm is. Instead, it’s a complete back to the drawing board retooling process. This reduces the size of the DRAM chips, but because it is such a young process, yields are still relatively low in comparison to DDR2.
This is only made worse by the fact that density is currently the same as DDR2, and a single side of DDR2 512MB is made from two cores per package stacked on top of each other. In DDR3 they are one above the other making the whole package physically larger. With companies (that use these memory chips to make their modules) wanting to shift 4GB kits in the ankle high wake of Vista, it’s just adding more stress to the situation.
But that isn’t all though, as it looks like we’re definitely going to see the same problem we had with the DDR to DDR2 transition, in that low latency DDR2 will be faster than any of the first DDR3 modules. This leads to having a rather unexciting and pointless product for the next 6-12 months while DRAM manufacturers get used to the new process. Of course, if something special happens from the DRAM manufacturing industry, DDR3 could take off quickly, but all of the current signs point to that not happening.
Discuss in the forums