AMD Admits "Stars" in Ryzen Master Don't Correspond to CPPC2 Preferred Cores

[Kaotik]

So basically: Best Core is absolute best core as validated by AMD from the factory.
Preferred Core is relative best core assigned by Windows Scheduler.

AMD gonna update ryzen master in the future to match Windows Scheduler.

Is that everything.

At least pretty much everything.

 

[bug]

So basically: Best Core is absolute best core as validated by AMD from the factory.
Preferred Core is relative best core assigned by Windows Scheduler.

AMD gonna update ryzen master in the future to match Windows Scheduler.

Is that everything.

Not really. Best core is what AMD marks from factory.
Preferred core is what AMD picks/exposes through CPPC.

It's really AMD disagreeing with themselves.

 

[Oberon]

Not really. Best core is what AMD marks from factory.
Preferred core is what AMD picks/exposes through CPPC.

It's really AMD disagreeing with themselves.

They're not disagreeing with themselves, they made a decision to show something different in Ryzen Master than is reported by other utilities.

 

[bug]

They're not disagreeing with themselves, they made a decision to show something different in Ryzen Master than is reported by other utilities.

There are no other utilities. CPPC is part of ACPI which is part of the BIOS/UEFI. Utilities will build on top of that, but the ACPI tables still come from AMD.

 

[Chrispy_]

Sell it and buy an Intel. You'll feel better. Right?

If not better, at least he'll be warmer.

 

[Oberon]

There are no other utilities. CPPC is part of ACPI which is part of the BIOS/UEFI. Utilities will build on top of that, but the ACPI tables still come from AMD.

Right, and those utilities are reporting the CPPC values. Ryzen Master is not. There is no conflict.

 

[Kaotik]

There are no other utilities. CPPC is part of ACPI which is part of the BIOS/UEFI. Utilities will build on top of that, but the ACPI tables still come from AMD.

Yes, but like AMD has explained, Ryzen Master doesn't report that. It reports the best core based on it's properties at factory, but due nature of how Windows scheduling works, it isn't necessarily the best core in Windows (because really Windows wants 2 best cores, not just one, since it passes the single threaded load between 2 cores for power management etc) - CPPC2 reports the best 2 cores within same CCX, because 2 best cores overall could be in 2 different CCXs which would hurt performance when the thread is thrown around. For the same reason 3rd and 4th best cores are always marked on the same CCX as the best 2 even if they're actually the worst 2 cores on the CPU, it's better for performance to fill one CCX before spilling on to the other CCX.

 

[bug]

Yes, but like AMD has explained, Ryzen Master doesn't report that. It reports the best core based on it's properties at factory, but due nature of how Windows scheduling works, it isn't necessarily the best core in Windows (because really Windows wants 2 best cores, not just one, since it passes the single threaded load between 2 cores for power management etc) - CPPC2 reports the best 2 cores within same CCX, because 2 best cores overall could be in 2 different CCXs which would hurt performance when the thread is thrown around. For the same reason 3rd and 4th best cores are always marked on the same CCX as the best 2 even if they're actually the worst 2 cores on the CPU, it's better for performance to fill one CCX before spilling on to the other CCX.

They can report whatever they want, I was only saying this confusion is entirely on AMD, not on Microsoft. It's really no biggie, they should have paid more attention and use more appropriate terms than "best" and "preferred", but now that boat has sailed.

 

[Kaotik]

They can report whatever they want, I was only saying this confusion is entirely on AMD, not on Microsoft. It's really no biggie, they should have paid more attention and use more appropriate terms than "best" and "preferred", but now that boat has sailed.

Actually it's on Microsoft if on anyone. Microsoft is the one who decided Windows schedulers needs to use 2 cores to optimize power management etc, AMD or Intel or any other CPU manufacturer doesn't require that nor do (all of) other schedulers in other OSes. If they'd use just one core instead of 2 for single threaded workloads, it would be the same best core Ryzen Master reports at the moment.

 

[bug]

Actually it's on Microsoft if on anyone. Microsoft is the one who decided Windows schedulers needs to use 2 cores to optimize power management etc, AMD or Intel or any other CPU manufacturer doesn't require that nor do (all of) other schedulers in other OSes. If they'd use just one core instead of 2 for single threaded workloads, it would be the same best core Ryzen Master reports at the moment.

Sure, sure...

 

[Kaotik]

Sure, sure...

Could you please elaborate on your logic?
These are undisputed facts which anyone can check:
- AMD lists in Ryzen Master the best cores per CPU and per CCX based on their properties at the factory
- All OS schedulers don't require 2 cores for single thread load
- AMD doesn't require 2 cores for single threaded load (nor does Intel or any other CPU manufacturer)
- Microsoft requires 2 cores for single thread load

And somehow, it's AMDs "fault" when they report those best cores per CPU and per CCX based on their properties at the factory instead of reporting the best 2 cores within same CCX as "best cores" because Windows wants 2 cores for 1 thread?

 

[bug]

Could you please elaborate on your logic?
These are undisputed facts which anyone can check:
- AMD lists in Ryzen Master the best cores per CPU and per CCX based on their properties at the factory
- All OS schedulers don't require 2 cores for single thread load
- AMD doesn't require 2 cores for single threaded load (nor does Intel or any other CPU manufacturer)
- Microsoft requires 2 cores for single thread load

And somehow, it's AMDs "fault" when they report those best cores per CPU and per CCX based on their properties at the factory instead of reporting the best 2 cores within same CCX as "best cores" because Windows wants 2 cores for 1 thread?

Ryzen Master shows the two cores per CCX designated as "best" by AMD during manufacturing. That same AMD then goes to expose "preferred cores" through CPPC, but those are not necessarily the same as those designated as "best" (again, by AMD). If that all makes sense to you, you're a better man than I.

 

[Kaotik]

Ryzen Master shows the two cores per CCX designated as "best" by AMD during manufacturing. That same AMD then goes to expose "preferred cores" through CPPC, but those are not necessarily the same as those designated as "best" (again, by AMD). If that all makes sense to you, you're a better man than I.

The reason they show different cores via CPPC(2) is the fact that Windows requires 2 cores instead of one, and due the way AMD built Zen those two need to be in same CCX for optimal performance (and further, preferrably not next to each other physically to optimize temperatures). Windows is just too big to ignore, in fact you really have to make most of your decisions on these things solely on how Windows works, because it's used by most machines (outside servers and supercomputers).
However, that doesn't change the fact that what Ryzen Master currently shows is correct for what it's saying it's showing, they are the best cores just like it says. In Windows that's just not relevant for anything but finding highest possible frequencies, aka HC overclockers looking for record frequencies which is done on just one core. AMD never claimed they're showing anything but the potentially highest clocking cores based on CPUs physical properties.
If you'd use OS whose scheduler wants just that one core for one thread, what Ryzen Master currently shows would be optimal (and next 3 threads would be assigned to same CCX as the best one even when they're not the 2nd, 3rd and 4th best cores on the CPU, because spilling to another CCX hurts performance)

 

[Naito]

So you're saying Microsoft should have had a CCX aware scheduler, before AMD launched the CCX?

No, but having said that, they should have (and probably did) worked together to optimize prior to launch. We can only speculate how the Windows scheduler works/doesn't work, but I would say it'd be naive to think that years of Intel dominance didn't play at least some part in shaping it.

Now that even AMD can't point out the "best" core accurately

Again, no. It seems to be a miscommunication on their behalf. Ryzen Master shows only a rank given based on the quality of the core within the CCX. One could assume metrics used are voltage, maximum clock, heat, etc. CPPC2 considers not only the physical qualities of the core, but also which CCX the core resides, the location of the core within the CCX, cache access, etc.

 

[ORLY]

I've just one question: if there are 2 CCX (0 and 1), will using CCX1 over CCX0 cause some kind of performance penalty?
Just wondering, because in my CPU the best core is located on CCX1, but Windows chooses to fully load CCX0 first and only then it loads CCX1, so may be it does so because not to load CCX0 first and to load CCX1 instead will cause some performance penalty?..

 

[Kaotik]

I've just one question: if there are 2 CCX (0 and 1), will using CCX1 over CCX0 cause some kind of performance penalty?
Just wondering, because in my CPU the best core is located on CCX1, but Windows chooses to fully load CCX0 first and only then it loads CCX1, so may be it does so because not to load CCX0 first and to load CCX1 instead will cause some performance penalty?..

No, different CCXs are equal to each other. Loading one CCX fully before spilling to next is better for performance. In your case, 2 best cores from Windows' perspective are within CCX0 even if best core based on it's properties at factory is on CCX1.
edit: fixed typos (not going to promise there isn't still some in there)

 

[ORLY]

No, different CCXs are equal to each other. Loading one CCX fully before spilling to next is better for performance. In your case, 2 best cores from Windows' perspective are within CCX0 even if best core based on it's properties at factory is on CCX1.
edit: fixed typos (not going to promise there isn't still some in there)

That's messed up then. Windows loaded it this way:
1 thread - the fastest core on CCX0
2 threads - the fastest and the second fastest core on CCX0
3 threads - all the cores on CCX0
4 threads - all the cores on CCX0 and the golden star core which is on CCX1

 

[Kaotik]

That's messed up then. Windows loaded it this way:
1 thread - the fastest core on CCX0
2 threads - the fastest and the second fastest core on CCX0
3 threads - all the cores on CCX0
4 threads - all the cores on CCX0 and the golden star core which is on CCX1

Curious, it should be fastest and 2nd fastest core on CCX0 for 1 threads already and to be honest I'm not sure on 2 threads if it's 2 cores or 3 (and cba to test, I got so much stuff running)