Calculting Multi-core processing power to compare systems...

Will the ability to produce a seemingly unlimited number of cores on a CPU now stop the CPU speed upward spiral? I was shopping for CPU's and got to thinking that we might see multi-core cpu speed go not much higher then the current cap of 2.33mhz. Maybe stop at around 3mhz?

It seems to keep the heat down this make sense. We might even see a decrease in individual CPU speed can be used to offer an overall increase in processing power once applications are converted to true multi-threaded software to take advantage.

I have not sat down to sketch out the numbers but there has to be an equation that will give us a way to calc the processor equivalency for yet to be announced 9000-core systems. Something that shows if a "16-core single processor system" running at 1.2mhz is faster then a "dual quad-core system" running at 3.0mhz...that is an example I just pulled out of the air but hopefully you get the idea.

It's gonna take time and it looks like it may get harder to wade through marketing hyope for a while...at least until people get their hands on these new cpu's and there are MB's to support multiple multi-core CPU's it really is just a thought experiment...

Sorry this happens when I can't sleep...

 

I don't think you could reliably develop any equations to model the efficiency of future processors, simply because (short of industrial espianoge) we have little to no idea what the specs of the architecture will be. Without specific diagrams of the architecture, I see no way to make any predictions of this sort. I could of course be wrong though, that's just my opinion.

 

Well, theoretical processing power is simple. Two cores are exactly twice as fast as one, assuming the cores are all identical. And, again assuming all external factors to be equal, two cores running at 1.5GHz are exactly as powerful as one core running at 3GHz. In theory.

In practice there are a lot of unknowns though. First, memory speed and bandwidth, cache access and such. Will the memory controller be able to keep all the cores fed? Will some of them have to idle because their siblings are hogging all the memory bandwidth? Obviously, that's not an issue on a singlecore system.
Then there's cache coherency. If core A saves data to memory, it will be placed into its cache. If core B now needs to access it, it has to wait for A to clear its cache (to ensure coherency), which causes extra wasted time. Again, not an issue on singlecore systems. And of course, the cores are rarely identical. There are major differences between Intel's and AMD's cpu cores, as well as their memory controllers, cache speed and sizes and so on. So it's really impossible to come up with a useful and general equation for what you want.

Next, there's the programming side of it. You can't just assign arbitrary code to each of the cores. Almost all games are still basically singlethreaded (some have tacked a few more threads on top of it, but I don't know of any games that have been designed from scratch with multithreading in mind, where everything is written to take advantage of it.
That means most games can only utilize one, maybe one and a half core. If you have 20 cores, well, it won't give you any extra performance.

So even if we found some equation that told you the relative performance of different systems with different numbers of cores, it still wouldn't work because it wouldn't take the actual software into account.

(Apart from that, there's still a very real limit on the number of cores per CPU. We're only slowly moving to quad-core, and it'll be a few years before we get much above that. There are still physical limitations on the maximum possible size of a CPU, so there's only room for a certain number of cores. Unless you switch to an architecture with much simpler and smaller cores, like the Cell has done. But even they still only have 9 cores.

 

They broke the 3mhz barrier a LONG time ago, think you mean GHZ( someone was gonna mention that sooner or later,might as well be me)

AMD just recently launched a dual core 3GHZ chip
http://www.computerworld.com/action...Basic&articleId=9011562&source=NLT_HW&nlid=51

Clock speed isnt the big factor anymore, and hasnt been for quite some time, its all about what you can acomplish per clock as opposed to how many clocks.