Testing bulldozer with affinity option in task manager

[Thefumigator]

Not trying to find a way of making a "slower than expected" chip "faster" but I just came with the idea that if the scheduler in windows is not optimal for bulldozer, maybe you can help it out by setting on which cpus to run certain process.

I do so when opening multiple instances of convert XtoDVD for example, on my phenom 9550 just open 3 instances and give a cpu to each, leaving one core unused for my other regular trasks. This makes my system almost as responsive as if I weren't doing anything, of course a raid 0 also helps.

Do you think this could work out? lets say, trying several instances of x264 or whatever, and assign a core to each, or maybe a pair of cpus (a module actually). Does anyone with a bulldozer at hand wants to try this and compare?

 

[Inceptor]

It would have to be a module for each instance, otherwise you'd run into the scheduling problem.
IF Windows, in that situation, can have the module work on two threads...
I'm not sure of the technical details of the scheduling problem... if there's only 1 module to work with, would it allocate workload as if it's dealing with a dual core cpu? Not sure.

 

[Thefumigator]

It would have to be a module for each instance, otherwise you'd run into the scheduling problem. (in bulldozer)
IF Windows, in that situation, can have the module work on two threads...
I'm not sure of the technical details of the scheduling problem... if there's only 1 module to work with, would it allocate workload as if it's dealing with a dual core cpu? Not sure.

actually yes, if windows have 2 cpus to work with (virtual + real, or both real, or whatever), then there shouldn't be a scheduling problem.

The technical details of the scheduling problem, to make it short: it happens when a process has several subprocesses, and the way windows assign each subprocess to certain core. This is impacting bulldozer performance, it seems.

Example: if a process runs in CPU #0 and that process has a subprocess running at CPU #1, then there is no problem (assuming CPU #0 and #1 are in the same module).

The problem triggers when a process running in CPU #0, for example, has a subprocess running in CPU#3 or whatever, where its in complete different module.

Actually, in IMHO AMD should have drop the module concept and design the entire processor as a whole module with 8 cores inside, with the "shared resources" available for every core, but hey, I'm not an engineer, its just an idea of course.

 

[erocker]

This has been discussed in many of the other "Bulldozer" threads. We do not need another thread.