Not sure where to ask this at but..

[bart1975]

I am fairly certain this will be easy to answer by many but I must still ask.

Why is there no software that gives you the ability to take unused cores and create a profile for games that will treat 2 cores as 1.

For instance some games are only optimized for dual core. In theory would it be possible to make a program that emulates and combines the unused 3rd or 4th cores as if they were part of the original 1st two cores.

Such as with core 1+2 the game would recognize as core one but witht he performance of both cores and so on.

 

[yinan]

Probably because of the huge amount of overhead it would have.

 

[bart1975]

What kind of overhead do you think this kind of software would have? I assume it would be enough to negate this kind of software from being useful.

 

[Veliko]

What kind of overhead do you think this kind of software would have? I assume it would be enough to negate this kind of software from being useful.

If I am reading your post correctly this software would sit between the game and the processor.

The game would normally talk straight to the CPU saying "I need this info calculated but only one core". The software catches this before it hits the CPU and changes it to say "Calculate this info but using all the cores available".

The CPU calculates it using all the cores but the game won't understand it. The software then takes the calculations, translates it back to single-core speak and gives it to the game.


I have no idea how much overhead that would create but you can possibly compare it to using Google translate - the messages get mixed up in the translation so you have to spend extra time cleaning it up.

 

[AyashiKaibutsu]

Single threads assume everything they do is in order... If it was just a matter of using a program to seamlessly make two cores work together windows would already be doing it.

 

[bart1975]

I was just thinking it would be possible to do since it seems video cards do something similar with SLI/X-Fire.

 

[SunnyD]

Intel was working on this a while back. It simply doesn't provide enough efficiency to make it worth while. About the only thing that can be done with such tasks are splitting threads across cores and doing full on branch prediction across cores, which is mostly a waste of time.

 

[Cerb]

I was just thinking it would be possible to do since it seems video cards do something similar with SLI/X-Fire.

Not at all. The work your video card is doing already scales out well in the range of thousands of 'threads' on up (billions is not out of the question, and you can remove the quotes for nVidia GPUs). As such, distributing it across a few GPUs is not hard work (keeping all the GPUs coherent is the hard part). To see this kind of problem run on CPUs, check out Apache performance (near linear scaling per added core per static page request).