VIA had plenty to show off at Platform Conference, all of which we covered in our VIA Roadmap.

Memory is always a large part of the show at Platform Conference. This time was no exception. Representatives from Hyundai, Micron, Elpida (formerly NEC-Hitachi), Samsung, Infineon, JEDEC, and even Rambus (for the first time) were in attendance. Here is a look at some of the issues in memory today.

The memory bandwidth roadmap for SDRAM-based PC systems is well mapped. DDR memory is starting to appear. DDR333 memory will be next, then the move will be made to DDR-II. Signaling speeds will rise, and bandwidth will rise. The problem is, latency is not falling as quickly as bandwidth is picking up, and latency is extremely important to performance. Seeing that problem clearly, several companies have brought out competing technology ideas on how to reduce memory latency and increase system performance. Micron, Elpida, and EMS have competing solutions. Their goals are, more or less, to have their technology become the industry standard.

Micron is proposing that chipset makers build embedded DRAM (eDRAM) onto chipsets. Micron has already shown a chipset with 8MB of built-in L3 cache, proving their technology can work, and claiming up to a 15% performance boost. eDRAM has the advantage of, like all caches, reducing latency to main memory quite dramatically when there is a cache hit. However, building large amounts of eDRAM onto a chipset may make the chipset much more expensive. We will be impressed if Micron can implement the technology without raising chipset costs significantly.

Elpida and EMS are each proposing their own cached DRAM technologies. Elpida is proposing Virtual Channel Memory (VCM) and EMS is proposing Enhanced DDR (E-DDR). Each method introduces fast SRAM onto DRAM in order to decrease latency. VCM's primary plusses are its potential for the highest performance and its low power usage. VCM's primary downside is the amount of die-space it consumes, which increase component price.

E-DDR's primary strengths are its low-die space overhead and, from the benchmarks we have seen, it performs faster than VCM in most cases. E-DDR's primary weakness is that the technology comes attached with royalty payments. As we have seen with the trials of Rambus, memory makers are not big fans of paying royalties.

We expect eDRAM or similar technologies to appear in high-end chipsets, but we don't expect eDRAM caches to become a PC standard, at least not anytime soon. eDRAM's huge advantage is that a chipset maker and a motherboard designer can make design choices to go with eDRAM without having to worry about where the memory industry will go. However, it can boost performance , but the performance will likely cost money.

Right now, we think the die size penalties of VCM will keep it from being adopted anytime soon. If any latency reducing technology becomes widely adopted in the next year or two, it will be E-DDR. It holds the potential to improve system performance without increasing DRAM prices significantly. However, we fear that royalties may negatively effect E-DDR's adoption.