I really, really need to upgrade my P4...

It actually has 6 physical cores, but will run 12 independent threads.

At a price point of over 1Grand I do not see it happening any time soon.....I am quite content with my i7Core 920.....

Me too - no budget this year, just got the Canon 550D with good glass, so no more big investments for now at least :-)

/Ulf

jabloom, you are correct: 6 cores, 12 processes with hyperthreading. My bad.

Fred

The 6 cores have 2 physical interfaces to each core. There is a full set of registers, stacks, and control flags for each of the interfaces to the 6 base cores.

Why do I feel like my P4 and I have moved back into the Z-80 days ?? :-)

mikey

What's frightening is that I remember so many of the Z-80 opcodes.

I need to make room for new knowledge (even though I think Z-80 sales are still going good, including Zilog's Z-80 compatible "fully pipelined 24-bit eZ80 with a linear 16 MB address range").

That CPU really was a breakthrough in its day, and if Zilog hadn't decided to make the Z-80 100% 8080-compatible except for one improved status bit, Intel could have gone out of business.

Yeah, but I can still write Z-80 assembly language in my sleep (well 8080 in octal anyway - used to hand assemble all my code before I even had a floppy drive to support an assembler). From a hardware designers point of view though, the Z-80 had a couple of quirks (timing and voltage level on the clock pin for example). Ah, well, time marches on !

mikey

"The 6 cores have 2 physical interfaces to each core. There is a full set of registers, stacks, and control flags for each of the interfaces to the 6 base cores."

But if you want to count the processor as having 12 "cores", then you would also have to halve the overall clock speed to 1.67 GHz.

And why would you cut the clock in half? This is not the old Pentium 4 hyper-threading scheme and the processor itself is nothing like those old cores.

If anything you would increase the core speed as each set of registers can be preloaded with the next instruction which helps to improve predictive branching performance substantially.

Each core can run parallel instructions, in the advent of the paths being through different subcomponents, and you end up with a pretty good gain in performance.

The performance restriction in any multi-core CPU is the single path to/from memory which requires any core to wait until memory fetches are done, when one of any of the cores, has to access external RAM.

The single biggest reason Intel added the second set of cores was an effort to keep the base core busy as much as possible. Anytime the CPU's have to access system RAM, the CPUs starve. Having another set of registers loaded and ready to run while slow system RAM is being accessed is another step up in performance.

The overall problem with these mutli-core CPUs is you hit a point of diminishing returns due to the single path to/from system RAM. Bus contention becomes more and more of a factor in performance as more CPUs attempt to access system RAM. Where the line is drawn will usually depend on the actual application.


Side note: I loved writting Z-80 assembly. It was a beautiful language. It was amazing what you could do in 64K of RAM. My next favorite, to write assembly for, was the Motorola 68000 series. It was also very elegant and easy to debug.

Each physical core can do a specific number of operations per second. If you run two threads in parallel on that physical core, you get half the efficiency of running one thread.

Since you covered some of problems with having multiple cores, you might also want to read this short note from a Microsoft tech guy, about how software and OSes have not kept up with multicore hardware:

http://www.fudzilla.com/content/view/18168/1/