In the article linked by Ed Sammons, here is an interesting quote:
<The EV-6 bus is a point-to-point bus, which "makes the electrical environment much, much cleaner," Meyer said. "With Intel's approach, once you put more than one drop on a line you don't have a transmission line anymore, you have a stubs. So you end up with ringing and reflections that make the electrical design much more difficult. I would expect that this 200-MHz bus to be simpler than a 133-MHz Slot 1 bus, because electrically it is a much cleaner environment.">
This is very weird. AMD opted for a point-to-point design, rather than a multiprocessor bus. Meyer is going off about the advantages of P2P over MP bus, but what about the disadvantages? How expensive is that north bridge going to be when you want more than one processor in your server? After all, for every processor, you'll need at least 180 pins (72 data, 36 address, plus request, control, power/ground pins).
I already know that AMD went towards P2P because they had no choice, since AMD is borrowing Digital's technology and the P2P was part of the package. But I know guys who used to work for Sequent, and they talked about huge big-iron servers with up to 20 processors on one bus. ONE BUS! At least Intel has decided that four on one bus is enough.
My point? It's still unclear to me what the advantages of multiple P2P connections are compared to a multiprocessor bus. We've already seen the benchmarks which show 4-way Xeon servers being twice as powerful as 2-way Xeon servers, proving that the MP bus isn't a bottleneck. The P2P connection might be simpler electrically, but it's going to be murder on the north bridge when it has to support four of those 180+ pin connections.
Tenchusatsu |
