AMD: Windows Thread Scheduler is operating properly for Ryzen

This is why I think application profiles would be a good idea. It's no different than what GPU drivers do - they have their own special optimizations specific to each game. It may be a lot of work but I personally would argue that AMD really only needs to focus on applications made for Windows 10 and newer. And even then, only major commercial titles.

At least Ryzen CPUs aren't as complicated as GPUs. It's basically just a matter of preventing certain processes from swapping threads, or setting affinity. Not much else to it.

 

Cool, wish I would have done the same before I swapped the parts out.

 

"AMD: Windows Thread Scheduler is operating properly for Ryzen"

- The thread that just... won't ... die.

 

Oh! Mhh, why do you think a single optional variable to be possible to configure by the user is not feasible, when on the other hand endless lists of options per game to fiddle around with and optimize by the user exist already?

Spoiler

As it works with everything... if you want to tweak your out of the box performance:
-> Pick the other option in the dropdown. -> Check for improvement. -> Leave best setting forever -> Profit.

Don't know what property of Ryzen makes it unfeasible for users to optimize it in contrast to a RX 480, or a GTX 1060, or an i3/i5/i7

Spoiler

So why do you think it is unfeasible for Ryzen especially? Can you answer that?

 

i agree. especially when doing so causes cache latency thrashing/frame drops in certain games due to zen architecture. this is exactly the reason why "working as intended" doesnt mean working well...at all. but i disagree with you that theres nothing to be done about it. its 2017. we have had multicore procs for over a decade. the OS needs to be able to assign major program threads core affinity BY ITSELF automatically when its beneficial. its certainly not impossible to recognize a core cluster & prioritize keeping the thread SOMEWHERE within it, if the OS does insist upon bouncing the thread in question between certain cores for load balancing etc


again guys, from this PR statement, its clear that microsoft simply isnt cooperating. AMD was essentially forced to come out publicly & say this...& they did it as diplomatically as possible, so thats nice i guess, but id say it borders on brown-nosing.

if people keep having this very same discussion about how the OS recognizes CCX modules (or rather, how it doesnt) & if its microsofts responsibility to attempt to patch their rabid, flailing scheduler...guess wat? microsoft might feel obligated to make their own PR statement saying nothing was wrong with win10 & its literally perfect, throwing AMD under the ****ing bus. AMD head them off at the pass to avoid it.

this isnt a """"""conspiracy"""""", its not cloak & dagger. its business, & its ****ty. thats all.

 

Use your brain and think about that question.
Most users do not fiddle with settings.

Or are you trying to say us power users that post in these forums are the majority?

Answer, obviously not.

99% of the world is technically challenged, that is why companies like apple are very successful, it just works.

The average consumer is not going to want to deal with these issues.

Even average gamers are not going to know how to set up an affinity per application they use via regedit or even 3rd party programs such as process lasso to enforce affinity.

Hell i tweak so much sh1t in 10 it becomes tiresome, i certainly wouldn't want to be doing that either.

 

Exactly, and the average customer doesn't care if he gets 90 or 110 FPS in 1080p as well.
And the average customer doesn't buy a 8C/16T CPU to begin with.

So to pick up your funny 99% number here...
-> 99% of the people who buy a highly multithreaded CPU like Ryzen 7 with 8C/16 threads are not your "average Apple customers". And they do indeed fiddle with the settings.

Just as the people fiddle with their GPU Profile settings.
And if they don't then they are fine with what they get to begin with.

Nobody forces you to use 10.
Or what do you use it for? One of the four DX12 exclusives:

Fable Legends
Forza 6 Apex
Gears of War: Ultimate Edition
Quantum Break

?

Or for Cortana, Edge and the Xbox integration?

Because what else reason is there to go through the hassle as you describe it?

 

Check edited post.

And you are wrong.

Most people who buy an NVIDIA/AMD GPU just plug and play.

There's even many people who buy Intel K cpus that do not even OC.

Gamers buy within their budget and are able to understand benchmarks, but that's about it.

It's only when there is a major issue, which ryzen performance can be depending on the game, will they start to try to 'solve' the issue.
Most likely the will assume something is wrong with their PC.

If you think the average consumer who buys any decent CPU will have the technical ability of previously mentioned you clearly do not work in the tech industry or even understand the capabilities of the average consumer.

Explain why so many people complained about poor performance after buying a super fast 8 core bulldozer?

Answer: they clearly don't understand the flaws of the architecture.

These types of forums here are the minority of user base.

 

The performance is not "low". Slightly lower than the top end , more expensive, i7's? Yes, but still very good and a massive improvement over the previous generation of AMD CPU's.

 

Are you talking about slightly lower performance in games?

 

I never overclock anything and I don't "tweak" Windows past what is offered in it's own settings. Guess what? I never have problems either and I don't blame Microsoft or AMD/Intel/Nvidia for problems I have caused myself.

 

Absolutely yes. The difference is minor, not huge. Besides that, if you're exceeding 100fps, then what you have is already overkill and who cares about a slightly higher number? Especially if it comes at a greatly inflated price.

 

There is no "major " issue in any game.

 

Lol, in games the £329 Ryzen loses even to a £200 I5 mate, stop embarrassing yourself.

 

People buy K CPUs so they CAN OC, not because they intend to right away. I only started OCing my 2500k recently because BF1 was giving it trouble @ stock.

 

Good idea. If you can, use something that gets framevalues in a format that Excel can understand. I want to see variance etc :nerd:

 

Question for the more savvy people:

In another thread there was talk that the CCX latency could be helped if Windows handled CCX clusters as NUMA nodes.

Is this valid and doable via Windows update, do you think?

 

Ok, I am not a computer engineer but how can one physical core contain one physical thread and one logical? To my understanding it is one physical core with two logical threads.

 

Posted this in another thread but AMD is right, it's not a scheduler... but theres another issue cross talk between CCX modules. Windows is assigning threads to the physical cores as it does with Intel CPU's however the Ryzen doesn't like data being synced across different CCX modules as cross talk is around nearly 3x longer, 40***956;s on the same CCX module or 100***956;s across modules, games with graphics rendering, AI physicals etc cause huge latency penalties.

Download "process Lasso" run that in the background, start your game, minimise the game and go into process Lasso - go to active processes - right click on the game running - go to CPU affinity (limit CPU use) next go to always - select CPU affinity - de-select all cores on the right hand column (8 total - 8-15) go back into your game. Now enjoy i7 gaming performance. (will need process Lasso running in the background however)

Since you no longer have cross CCX module talk your performance will now increase massively, since latency has been reduced by 3X! Performance will vary but for me in armoured warfare i gone from max 110fps to 160fps.

https://www.youtube.com/watch?v=JbryPYcnscA (Ryzen 1800x - Windows scheduling threads across CCX units) this was what inspired me to try this and it works!

 

Unknown if that'd even work - likely doing so would cause a BSOD as NUMA layout is deployed differently. Each node has it's own allocation of memory accessed via a NUMA supportive memory controller & accessed across the NUMA connection - if anything I feel as if it'd be emulating NUMA more than running in NUMA since NUMA (processor + memory per node across NUMA span) is a whole lot different in nature. Yeah I say it'd BSOD...