Platform Conference is essentially a small gathering of the PC industry, particularly geared for Intel-alternatives such as AMD and VIA. It is used by companies including AMD, VIA, Micron, Hyundai, ALi, and many others to introduce new technologies, start initiatives, deliver messages to developers, evangelize, and push their platforms forward. From DDR and ACR to Athlon multi-processing and the future of Ethernet, Platform Conference is a window into the future of the PC platform. We attended the recent Platform Conference, held in the Silicon Valley Convention Center, to see where those other than Intel are trying to lead the PC platform. Incidentally, representatives from Intel also attend the conference and sometimes ask very interesting questions during Q & A sessions.

AMD and Tyan showed, for the first time ever, an early revision of a board based on AMD's 760MP chipset. The board was displayed in AMD's booth, not running in a system. As we covered in our recent VIA Roadmap, the AMD 760MP chipset will have to use two EV6 busses to connect two processors, which will likely make it a relatively costly chipset to buy and design a board for. Currently, Tyan is the only motherboard maker we know of who is designing a board. AMD and Tyan are trying to get the board out before the end of Q2 2001.

AMD also had a presentation on their LDT (Lightning Data Transfer) technology. AMD plans to use LDT as the I/O connection for future northbridges. AMD expects LDT to provide the bandwidth and low latency they say is necessary in future multimedia-oriented machines.

LDT will also be used to interconnect AMD northbridges on 4-way+ multi-processor systems. In AMD's 4-way+ designs, each northbridge will serve two processors and connect to local memory for those two processors. The processors will be able to access "distant" memory of other processors as well, but not as quickly as they can access local memory. This architecture is called a Non-Uniform Memory Access Architecture (NUMA). You can contrast NUMA with Intel's method of sharing memory between more than two processors.