I'm using a laptop(g73jh) that is running a Core i7 and Win7. It does not have a bios option for toggling HT on or off. I am curious if anyone knows of a way to disable HT if you don't have access to it in the bios. Googling this mostly brings up forum arguments about why disabling HT is a good or bad idea and that's not what I'm looking for.
I have reason to believe that Windows is not properly scheduling tasks to the processor cores in a way that achieves maximum performance. Example of what I'm talking about: If there's three threads of work to do, Windows is putting two of them on one processor core and one on another. In essence, it is favoring loading up the logical threads before loading up the physical cores in some scenarios. Obviously, three threads being worked on by three different cores will get much more work done than if two of them were on logical threads on a single core. If things were working optimally, Windows would prefer to put work on the actual processor cores before starting to put two threads on a single core.
The benchmark I ran that made me think this was occurring was wprime. With the program set for 4 threads and 32M, I get a score of 22.4 seconds. With it set for 8 threads and 32M, the score is 15.6 seconds. Surely that kind of performance gain isn't reasonable to expect from hyper threading alone. Something must be wrong when it is running 4 threads; IE it's not scheduling them to the four cores.
I have reason to believe that Windows is not properly scheduling tasks to the processor cores in a way that achieves maximum performance. Example of what I'm talking about: If there's three threads of work to do, Windows is putting two of them on one processor core and one on another. In essence, it is favoring loading up the logical threads before loading up the physical cores in some scenarios. Obviously, three threads being worked on by three different cores will get much more work done than if two of them were on logical threads on a single core. If things were working optimally, Windows would prefer to put work on the actual processor cores before starting to put two threads on a single core.
The benchmark I ran that made me think this was occurring was wprime. With the program set for 4 threads and 32M, I get a score of 22.4 seconds. With it set for 8 threads and 32M, the score is 15.6 seconds. Surely that kind of performance gain isn't reasonable to expect from hyper threading alone. Something must be wrong when it is running 4 threads; IE it's not scheduling them to the four cores.
Facebook
Twitter