Actually they are plenty of reasons not to have Hyperthreading
on smithfield. Heat is a certainly an big issue, but there are
lots of complicated design problems as well.
You'd have a 4 thread systems, but if asymmetric structure,
2 threads (one one core) share ALU and FPUs etc, share cache, share pipeline space, and two don't. Now the electronics has to
decide which two threads share the same core, so as make good
us of the available resources, this is a bitch of a decision problem, get it right and you gain maybe 5-10% performance (about the maximum gain from HT on a single core system), get it wrong and you thrash the cache, and stall the pipeline, and lose maybe 30% or more performance. So theres a big design problem and a lot of extra electronics to add, for something that will only help on a few occasions. Getting HT and dual core to work together well would problem need to be designed from the ground up, and if done well (i.e. with a much better HT than in current P4's) and maybe with a wider superscalar core, would be a very nice design. But Intel's a long way from that. They were rumours (on aces and cpu_architect) though that the K9 (now pushed to K10) might have this sort of design.
P.S.
Hi all, I'm new to SI but a sometime poster to the Yahoo AMD thread. I do tech speak, I don't do bitching! |
