AMD: Windows Thread Scheduler is operating properly for Ryzen
Click here to post a comment for AMD: Windows Thread Scheduler is operating properly for Ryzen on our message forum
moeppel
Is the phenomenon Ryzen specific or can be observed across all platforms?
Either way, inevitably not the best of news 😉
SirDremor
LOL!
So much for those trying to Prove how bug in Windows slows down Ryzen!
Good luck!
See ya!
Perjantai
Pcper had great investigation about this scheduler bug. It looks like windows does spread load like it would with Intel processors. However ryzen works bit differently. Cpu cores talk Latency between different ccx is over 100ns but within same ccx its lower than intels at 40ns. This could be fixed within windows.
Ryu5uzaku
Yup. Linux handles this better currently. Regarding SMT nothing wrong there pretty much.
Enizax
Sounds promising what with the Naples chip coming out soon.
...I can't wait to actually see some competition in the commercial server sector...
:cheers:
Mtom
Actually AMDs answer is stupid, and doesn't make sense. So W10 scheduler is working as intended, but W7 work better?
Also many proved that it might work as intended, but its NOT working as it should with Ryzen.
One example look at BF1 20% difference if they let the scheduler throw the workload between the CCXs
http://image.prntscr.com/image/57d46709a9eb4491aa4cd9a568a854e8.png
Other example:
Assigning threads to specific CCX: https://www.youtube.com/watch?v=BORHnYLLgyY
Windows randomly selecting CCXs: https://www.youtube.com/watch?v=JbryPYcnscA
Kaarme
That wasn't a very good answer from AMD. It hardly addressed at all the unexpectedly poor performances in some games. It's sad if they don't know their own CPU well enough to say anything about it, but it's even sadder if they in fact do know, but can't do anything about it, and thus try to dismiss the whole issue.
__hollywood|meo
kaarme, it seems from this oddly diplomatic "concession" made by AMD...that their official statement actually stems from microsoft being unwilling to promptly address the issue. you have to read between the lines here.
its still quite odd of them & rather questionable as well considering its demonstrably false
vase
So this means...?
Set core affinity profiles for gaming and basically close up to 7700k performance.
So you get your parallel workload monster & a solid gaming CPU for 330 bucks as long as you are taking care of setting up your PC properly.
-> Set affinity profile for game exes to 1 CCX is 2 clicks... seems doable considering it's a one time setup which is just a vanishingly low effort which just slips into the rest of the first setup of a game's settings...
And once games are able to fill all 16 threads to 100% you'll just profit again.
+30FPS for correct affinity (force 8 threads, and high perf plan):
https://community.amd.com/community/gaming/blog/2017/03/14/tips-for-building-a-better-amd-ryzen-system
anthos
Yeah, 2040 can't come fast enough.
vase
Even if it doesn't come fast.
Noisiv
But mostly you get a permanent new hobby. :hatty:
Like follow closely AMD, MS, developer tweets, write your own scheduler, engage in back and forth with your RAM/BIOS manufacturer
"taking care of setting up your PC properly"
LOL that too:
- Hmmm...this FPS-tanking im getting. Is it working as intended, or did I fail to set up my PC properly??
My main gripes are as follows:
Birthing issues. promises, promises..lets blame Windows 10 etc
No iGPU. Raven Ridge late 2017 (translation: 2018.)
I doubt Zen will be be able to brute-force its way through unoptimized legacy code (WoW, ARMA, DayZ, DCS)
Occasional and mysterious FPS-tanking in upcoming games is probable. Due to not being as popular and optimized for as Intel
400€ for the cheapest of the Zen bunch - supposedly the deal of the century(woot!)
Nevertheless Zen is a fine and an intriguing new product. If I had the time to tinker I would certainly get one.
8/16 sounds godlike, even if its really 2x 4/8. And it should only get better from now on, right?
Then again 400 euros for a piece of silicone, that I am not going to utilize fully, coupled with some worrying issues wtf am I thinking... Vega pls 😀
Fox2232
I was doing same thing with intel's Nehalem i7. But instead of putting threads on one CCX, I was manually assigning threads to real cores as use of HT had negative effect similar to using different CCX.
Nehalem was choked by HT in some cases and Ryzen has trouble syncing threads due to increased latency.
I wonder how will 6C/12T do. And it is apparent that 4C/8T will deliver wonderful experience within its parameters as it will not have this downside.
vase
Yes indeed, it helps on 8thread intels as well! (my 4790k for example)
Just certain people aren't interested in tuning their system it seems!
rather talk smack...
schmidtbag
This doesn't sound right to me, at all. There is evidence suggesting W7 can sometimes perform better than 10, despite not getting much attention from either AMD or MS. There is evidence suggesting disabling SMT improves performance, where the additional logical threads created by SMT are directly related to the scheduler.
There is a scheduler problem. Just because it behaves the way they intended, that doesn't mean it's behaving correctly. That's an important distinction. When a child decides to get mouthy to his teachers, the parent could say "oh that's normal - he always does that when he doesn't get his way" but that doesn't mean his behavior isn't a problem.
RealNC
I see the "Linux handles it better" argument flying around too much. It should be pointed out that Linux's scheduler prefers high throughput. It is able to maximize CPU utilization better, but when it comes to latency it does worse than Windows. However, that's by default. Windows has a scheduler mode that also maximizes throughput. Windows server editions have that enabled by default.
Athlonite
So glad I'm waiting to see what happens with Ryzen / windows 10 there'll always be teething problems with new archs so this time I'm not going rush out and buy it
tsunami231
didnt AMD say it had nothing to do with scheduleer weeks back? and even if it did it would make much difference?
Chillin
I love this, some of the previous replies here are amazing in their alternate facts© .
So here we have AMD themselves coming out and saying that there isn't going to be major performance boosts and that the system is working correctly, and people here on this forum seem to know better and are saying that AMD is wrong and that there will be magical future performance boosts (against what AMD is saying).....
People make excuses for AMD and Ryzen no matter what the reality, and it's sad. The market doesn't care for these excuses outside of some noisy fanboys***8203; on some forums, and AMDs market share reflects that.
tehort
I wonder if the CCX problem is actually a Windows/Kernel problem, or an application-level optimization?
It's pretty much impossible for Windows to guess the kind of work a certain algorithm is trying to process, it should be the application`s duty to be optimized for certain architectures according to its algorithm optimizations or lack of.
also, it`s pretty common to blame Windows, and expect miracles from AMD, so...
Besides, Ryzen doesn`t seems to have more or even less advantages against Intel on Linux.
