Wednesday, September 27, 2006

The Empire strikes back

Intel has fired a shot back at AMD in the battle for the hearts and minds of third-party chip makers and OEMs.

News of proposed extensions to PCI Express called Geneseo are technically quite different that what AMD is doing with coherent HyperTransport (cHT). But functionally, they are a shot to AMD's heart.

Both efforts aim to link a coherent CPU with co-processors such as future high-end graphics chips and acclerators for XML, security and networking. But Geneseo is by no means a coherent version of Express. At best it will only provide a locking scheme and coherency hints to smooth performance. It also proposes fine grained power management support.

So what's new? AMD tries to do something the right way, and Intel tries to steal the wind from their sails with "good enough" technology. --rbm

1 comment:

Anonymous said...

Not sure which analogy is best - a pea shooter or Little Boy Blue trying to plug all of the holes in the dike. Each approach has flaws. Each approach has benefits.

Intel will claim the ubiquity of PCIe makes it the preferred high-volume approach.

AMD will claim HT provides the superior performance technology to maximize customer value.

Problem with Intel's argument is AMD has a high-speed PCIe offering in addition to HT. Intel has PCIe and makes it clear that they will remain closed, proprietary, and a poor partner as evidenced by their actions - CSI is simply off limits.

Who will win? AMD is tangible today while Geneseo is vaporware for at least 2-3 years. There is a lot to be said for establishing a beachhead. If, and it is a big if, AMD keeps licensing reasonble for Torenza and they can create tool chain and product streams from partners then they can create significant inertia. With IBM, Cray, and others announcing various hybrid offerings, customers will see real products more quickly.

This does not mean PCIe is out of the game. Quite the contrary given one can attach accelerators or any I/O via PCIe. Solutions do not need the minor enhancements to PCIe (other than marketing mystique, Intel code names are meaningless to customers and the industry)to sell. PCIe 2.0 at 5Gbps is a nice performance boost. PCIe 3.0 at 10Gbps will provide yet another nice performance boost. Both signalling rates will be enough for a subset of the application space so both will be successful. So the success of the PCIe approach isn't in question. What is in question is whether the dollars will flow more quickly and in greater numbers in the AMD-based approach? That is the challenge for Intel who has obviously been missing on the financial side and laying off thousands to compensate for poor whatever in its business model

 
interconnects