Hardware vs Logical core assignments

[dblue]

I´m overclocking and have a failing core on quad core 775 cpu.
I want to direct a process to this core but it seems windows 7 keep changing the core number each boot.
I´d like to know if a physical core can be assigned to always run a process... can this be implemented?
or ...Can logical Cores be remapped so they always match same hw core on each boot?
...it would be a nice selling feature for overclockers..
thank you!

[BenYeeHua]

Er....
I try to understand first

I´m overclocking and have a failing core on quad core 775 cpu.

I am overclocking the quad core 775 cpu, but I having a core that fail to using it.(mean damaged)

I want to direct a process to this core but it seems windows 7 keep changing the core number each boot.

I want to assign a process to this core but it seems windows 7 keep changing the core number each boot.
---
Er... what?

I´d like to know if a physical core can be assigned to always run a process... can this be implemented?
or ...Can logical Cores be remapped so they always match same hw core on each boot?
...it would be a nice selling feature for overclockers..
thank you!

Why you need to using a core that damaged?
What do you means?

And as I know, the windows 7 will not changing the core number...
Core 0=Core 0 on that CPU, whatever you has reboot windows 7, its is still the same.
Or you means sometimes it will fail to detect that damaged core, and only showing 3 core?

You must explain the meaning of "a failing core" first
A core that damaged?
Or...
---
And the 775 is not having HT(logical Cores)....
If the core is damaged, why not disable it?
---

I´d like to know if a physical core can be assigned to always run a process... can this be implemented?

If you need to let him running a process, just by running a software and set the default CPU affinity to that core.

[dblue]

By failing core I mean that one that is not bearing stress of overclock on prime95

I want redirect selected processes to it so if something goes wrong it most probably will affect a known process... also redirecting processes there windows somewhat avoids usage.. choosing a more free core

I think windows changes core number because prime95 changes the failing core (multiple tests) on every boot... and when Prime95 stops the thread/core it agrees with windows core numeration.

[BenYeeHua]

I think is the volt is not enough for 4 core, or the temp is too high.
Did you had disable spread spectrum and other?
And did you test the CPU only, not the Ram?

PS:The windows will not changing the core number, because why it need to do that?

[edkiefer]

Prime95 is set number of threads not cores, windows will try an spread the threads across all free cores .

Other apps probably will work better with affinities .

[Jeremy Collake]

I´m overclocking and have a failing core on quad core 775 cpu.
I want to direct a process to this core but it seems windows 7 keep changing the core number each boot.
I´d like to know if a physical core can be assigned to always run a process... can this be implemented?
or ...Can logical Cores be remapped so they always match same hw core on each boot?
...it would be a nice selling feature for overclockers..
thank you!

Wow, that's an unusual situation. I've never heard of anyone damaging a core. Of course, I don't keep up with the overclocking community regularly. How do you know that Windows is remapping the core to a different logical processor #? Are you certain a particular core is damaged, as opposed to the entire CPU? (edit: or other problem)

Assuming a damaged core is the case, the solution you really need is at the BIOS/UEFI level. You need to disable that core there. However, I'm assuming your BIOS/UEFI doesn't allow such, else you would have done this . Otherwise, there isn't much I can do. If Windows wants to change the logical core assignments around, I can't stop that.

[BenYeeHua]

Prime95 is set number of threads not cores, windows will try an spread the threads across all free cores .

Other apps probably will work better with affinities .

No, I has tried, after you running >2 worker, it will showing that.

Setting affinity to run worker on logical CPU #1

Setting affinity to run worker on logical CPU #2

So the threads is locked with 1 thread per Core.
----

I've never heard of anyone damaging a core.

I has heard of anyone damaging a core.
When they are unlocking the core(AMD), sometimes the unlocked core will damaged.
And sometimes I saw that, their CPU-Z is showing 1 thread, 1 core while the CPU is 2 core.

If Windows wants to change the logical core assignments around, I can't stop that.

Never heard this.
Because the temp of core is showing the same core that using.
Like CPU 0 is 100% CPU usage, the temp of that core is showing higher.

[edkiefer]

yes, you can't have more threads than cores as it runs test in parallel but if you run 1 or 2 threads on  4 core it will divide threads .
It could be he has issue with 1 core getting to hot an throttling down , though this would be easy to see with any of 3rd party temp apps .
Some bios you can also set each core clock, though I don't really see this as a necessity unless there a issue . you can also disable turbo and have fixed load clocks too .

[BenYeeHua]

Except he is running less than 4 threads(quad core)
I just testing Prime95, when you start the Torture Test with worker less than the number of core, it will divide the threads.
But if the worker is equal to the number of core, it will locking the threads per core.
You can test it by running equal threads first, then you stop 2 of them.
The CPU usage is 2 core 100%, 2 core 0%.
----
Ya, I think is the core getting too hot, and the stability is going down.
And I having the 4400+ x2 CPU running with 1.15V for 2 years.
After I knowing the basic of overclock, I test it with Prime95.
Wow, in 0.1 second, it stop as it detect error.
And I haven facing any BSOD in 2 years.

So I set the volt to 1.5V, and it is not detect error anymore. 

[Jeremy Collake]

I think I may have initially misunderstood the actual problem the user has here, or request being made. If the requirement is to dynamically select an appropriate CPU based on some metrics, then that is doable, but when you take that to its furthest extent, you end up with a duplication of what the CPU scheduler already does (select the best core for a particular thread, and change only as required or desired in later time slices).

If you want it to more aggressively switch a thread to a different core in order to limit thermal emissions, something I believe the scheduler will already do (swap a CPU bound thread to a different core in time, just to round out the load on the CPU), then there's no good answer for that right now.

The best thing would be to keep your CPU in a state to where it can *always* handle one or more CPU bound threads running continuously indefinitely. If your system does not pass this test, then I wonder if the extra few % gain in performance gained by overclocking or unlocking cores is really worth it.