Saturday, September 22, 2007

AMD Triple Core : more thoughts

Alright, so what exactly do I think of the Triple Core processor... is it a good idea?

Well, my gut reaction is no. The processor matrix from AMD currently has at least 8 active or semi active sockets (754, 939, 940, AM2, AM2+, AM3, Socket A, Socket F). Many of these sockets have similar named processors with major brandings of Athlon64, Sempron, AthlonX2, and AthlonFX.

However, with Quad-Core, AMD is introducing a new brand name, Phenom, which will also cover triple core processors. If AMD leaves Phenom for Quad and triple cores only, then there will be a clear product separation.

Going beyond the gut reaction, this is a surprisingly good move on AMD's part since it's one that Intel can't match. Intel is still reliant on the aging Front-Side-Bus architecture and has no direct-connect system put on the market. The result is that Intel can only do dual core and quad core chips on their architecture.

AMD, however, has HyperTransport, which is a Direct-Connect architecture, and also has the memory controllers on the die of the processor itself. The result is that processors don't have to fight for memory allocation, and can have direct access to their own memory sockets.

Developing cheap motherboards for Triple-Core will also be fairly easy, just remove one of the memory sockets and tracings from an established quad-core motherboard. On the mid-high range of motherboards traced for Quad-core, the triple core just slots right in.

The result is that vendors shipping SMP enabled Linux's will be able to offer higher performing systems right "now" (now being relative to Triple Core reaching OEM) on AMD systems than on Intel Systems. As SMP aware consumer software comes on line, the Triple Cores should start to outpace their dual-core predecessors, even on Microsoft products.

My analysis is that Intel is going to have to shift strategies again, and very quickly. Intel already had to admit that AMD had the right idea on IPC (instructions per clock) being more important than raw speed. Now Intel is faced with the possibility of leaving a market segment unanswered due to not having a Direct-Connect Like Architecture and an on-die memory controller.

While I feel sure that Intel is quite capable of shoving their memory controller onto a processor die, they'll run into a rather large problem. Their current power per watt ratings are going to skyrocket in accommodating the inclusion of the memory controller.

There is also the problem that whether or not Intel wants to admit it, their Direct-Connect Like Architecture is still 2 to 3 years out from mass production. I'm not going to say that Intel's only option is to join the Hypertransport Consortium, but it's one of the few options I see that will allow Intel to remain competitive.

The largest question then is whether or not AMD can feed channel demand for TripleCore processors... We'll know the answer to that in time.

No comments: