Let us start out with the blunt reality, as defined now, CSI won't beat Hypertransport/Ecstasy, and was not engineered to do so. OK, the marchitecture folk tried to engineer it to do so, but that version died a horrible early death, long before the Whitefield debacle.
CSI was started with a "mandate" to beat HT at everything, speed, bandwidth, per pin bandwidth, power usage, and everything else under the sun, not under Sun. From everything I heard, the engineers shot way too high, and achieved few if any of these goals. Hubris hit. So, what do you do in this case? Drink heavily and then start over.
So, Intel started over, and the new CSI had the design goal of 'adequate', and it looks like it hit that target squarely. It isn't going to be a world beater, and has vastly better RAS features than the current HT implementation, beating it here and there. As it stands, it is about equal to a 1600 front side bus (BUS), and Caneland should have 4x 1333 links, so it won't improve on that much. It will be vastly cheaper to implement though.
The problem with CSI is that Intel is beating the current implementation with a product that is set for 2008, maybe, if the stars align right and Venus is in Uranus. I believe the tech is ready, but the first two lead off vehicles, Tukwila the elder and Whitefield were canned long ago, and the current standard bearer isn't looking like it will wow people either should it survive the events of mid-2006.
So, if AMD sits completely still, and Intel actually executes, then CSI will probably beat it. The new HT is set to come out in Q2, and there is at least one other upgrade planned for 2007, one that should match most if not all of CSI's RAS features. AMD is not sitting still.
In the end, CSI is aiming for a moving target with something that is currently adequate. If it comes out on time, as long as AMD doesn't utterly implode, it won't stand much of a chance. It will still be head and shoulders about the current FSB (bus). µ
Sign up for INQbot – a weekly roundup of the best from the INQ