dual cores instead of smp

[Nightingale]

but heres the one thing that bothers me. How much will a dual core chip for running in quad mode cost? Would the chip have to be built like the 8xx opterons with multiple hyperlinks? Might not be able to afford quad opteron until 2006 unless they are really competitive.

 

[gustav]

You know dual cores could be just two cores acting as one, or working together as one. Just because it has two cores, doesn't mean it will act as though its a smp rig with one dual core cpu or as a quad with two dual core cpus. The best thing to do is wait and see how the technology looks when it comes out, instead of speculating what will and wont happen and how it will act, because really, we don't know.

 

[zachj]

All those that praise the low heat output of the 90nm G5 need to take a serious look at the new 2.5GHz PowerMac G5 . . . They had no CHOICE but to implement water cooling. 90nm will lower heat output at the low end. But all three major manufacturers (IBM, Intel and AMD) have made it clear that they end up with very close to a runaway heat monster when they crank up speeds. So I think anyone with dreams of overclocking these things might want to seriously reconsider . . . At low speeds, I agree, these probably won't be problematic when it comes to cooling. They might run a tad hotter than the highest-end model today if they follow the heat output of current mobiles. My fear is that they'll have difficulty ramping these things up, due to cost, complexity, and yields. I hardly think AMD is going to have an easier time than Intel in making fast 90nm processors IN BULK. Add in another core, and the chances of yielding a good crop at competitive speeds seems pretty small. Given that, prices will most certainly be higher than current models, and considering the price of materials and die size, it's going to cost more to make, too. I'm not suggesting I have any clue what the prices are going to be, but it only makes sense. What's the size of a dual core 90nm die? If it's sharing cache, it'll obviously be smaller than twice the size of a single core, but I can't imagine that it will be less than the current die size.

Z

 

[liquidtrance123]

but heres the one thing that bothers me. How much will a dual core chip for running in quad mode cost? Would the chip have to be built like the 8xx opterons with multiple hyperlinks? Might not be able to afford quad opteron until 2006 unless they are really competitive.

i would think that each DIE would have its own hypertransport link and the dual cores would act somewhat the same as hyperthreading. Executing 2 actual threads at once and shoving both the results through 1 hypertransport link.

just a guess tho

 

[mateo]

I recall some mention that a dual-core chip would still share a memory controller, so its not like you're duplicating two entire dies- the heat issue shouldn't be so extremely bad as some may think. I think this is also how the claimed drop-in compatibility with current board would be maintained.

More importantly, I think its important to watch how apps evolve. Since dual-core will become the standard, and more programs implement multiple threads, a 2-way/single dual-core SMP rig may turn into what a single rig is now in terms of multitasking- if you run more than one of these multi-threaded programs at once, CPU utilization will be eaten up a lot faster than if they were single-threaded like they are now. That's one thing that is luring me toward dual dual-cores in the future, because for really smooth multitasking, you essentially have to stay one step ahead of the processor usage curve.

 

[Audioaficionado]

On a NUMA board you wouldn't get full memory access for each core but only the processors. I might go dual core in an SFF 64 Hammer box. It's lot like HT in a P4 but better. Any games written for HT should also work well in true SMP.