Is there any way to virtually create a single core out of many cores?
Fool a program into thinking that it is only using 1 core but multiple cores are actually putting in work?
>>58601020
What do you thing an OS scheduler does?
>>58601044
The idea would be to improve performance in older software that only seems to use 1 core anyway
Just disable everything but the coldest running core and clock it to the moon
>>58601084
to be honest I'm not looking for alternate solutions
>>58601114
Well then you're shit out of luck famalam
>>58601020
you have two pipes, how do you stuff a single thing through both of them without separating it or merging the pipes?
>>58601128
Increase pressure
(ie, disable all the other cores and clock one so high it draws the same power as the others would anyway)
>>58601063
There are some ways to do that, but they are very limited, and I have never seen it done on a binary. Its generally done at compile time with OpenMP to automatically find parallelizable loops and such.
The hardest thing would be to gaurantee temporal order of operations which rely upon eachother. In the worst case on a binary it would require solving the halting problem.
If you can gaurantee a halting state for a loop, then you might be able to thread it, although often you wouldn't due to safety constraints. In the end, I estimate it wouldn't make many improvements, and just the algorithm deciding if its canidates solveable would likely be more computationally intensive than the task its trying to optimize.
>>58601020
Not really. Efficiencies in parallel computing need to come from the software. It needs to know how to use multiple cores/threads properly. Otherwise you're just getting concurrent processing which is what OS scheduling does. Basically performing work on different cores but not at the same time.
>>58601146
shit doesn't work like that brother, you clock the whole chip up its going to get hot as shit no matter what and that's not really the thing OP was asking about
thanks for the discussion lads
reading up on it myself it seems like an abandoned idea
>>58601182
You don't clock the whole chip up, you disable all the cores except the 1 that runs the coolest and then start throwing power at it until said coolest running core is hitting 85-90c.
I had a 2500K that could do 5.1ghz on all 4 cores and 5.6ghz on the best.
Not what OP is asking, but it's a way to make a non-multithreaded program run faster.
I think Intel is working on this right now.
The most basic way of thinking about it is 'inverse hyperthreading'.