I hope one day that multi-core cpus develop the ability to run all cores together as one "super core" in order to process intensive single-threaded work loads. Of course this would require a zero latency connection between all cores. Being able to configure a set number of cores for this task would be even better.
(08-08-2017, 11:21 AM)vram74 Wrote: [ -> ]I hope one day that multi-core cpus develop the ability to run all cores together as one "super core" in order to process intensive single-threaded work loads. Of course this would require a zero latency connection between all cores. Being able to configure a set number of cores for this task would be even better.
That's already how modern "superscaler" CPUs work already - each core as multiple pipelines and can often run multiple instructions at once.
But there's a limit to that scaling, as you can't run 2 instructions at the same time where once depends on the result of another, and there's a limited number of instructions that the CPU can read "ahead" trying to find something to run without a dependency.
And then they can't be scaled up infinitely, as we're already getting to the point where clock speeds are so high that the distance a signal can travel is a complex factor in CPU design, plus the switching time required for each of the transistors it has to travel through - for example, assuming the electrical signal travels at 1/10th of the speed of light, in the time of a 3ghz clock tick it only travels 10mm. Just propagating the clock through the chip is a topic of intense research and engineering expertise.
This is also one reason why there are multiple levels of cache - a small, fast cache that can be "close" to the cpu, and a larger, slower cache that's literally "further away"