Windows 10 Scheduler bug causes poor RyZen performance

[zMeul]

On 3/11/2017 at 10:25 AM, Valentyn said:

I would like them to investigate why Windows 7 is apparently offering much better gaming performance compared to 10. 

why?! because W7 always offered better perf than W10

I even did my own testing back in the day

 

[xAcid9]

15 hours ago, zMeul said:

why?! because W7 always offered better perf than W10

I even did my own testing back in the day

 

Do you have something new that's not using pre-historic quad cores? Did my own test and the different is totally negligible.

http://www.3dmark.com/compare/sd/4568448/sd/4528633#

Window 10 is my daily driver with many craps installed while Window 7 is totally clean.

 

I'm know some games prefer 7 over newer windows but it's not always like you claim.

[zMeul]

9 hours ago, xAcid9 said:

Do you have something new that's not using pre-historic quad cores? Did my own test and the different is totally negligible.

http://www.3dmark.com/compare/sd/4568448/sd/4528633#

Window 10 is my daily driver with many craps installed while Window 7 is totally clean.

 

I'm know some games prefer 7 over newer windows but it's not always like you claim.

why the processor used even matters!?

the test are valid as long as the CPUs used are not SkyLake or KabyLake

[cj09beira]

On 11/03/2017 at 8:56 AM, Valentyn said:

Ignoring reddit, more and more people on overcloker forums and YouTube are popping up with better results on Windows  7.

 

I'd like a proper publication like PCper to investigate so we can see what's actually going on, or if it's just hopeful spiel. 

 

Also windows 10 being inconsistent is also an issue, and does the dual CCX on R7 RyZen play into that to some extent? 

 

Guru3d tested their 1080ti on RyZen 4ghz and 5960x 4.3ghz at 1080p. Some games Haswell was around 2 fps ahead, on others nearly 40fps, with Ti worse than normal 1080 in some on RyZen. 

 

The massive variance between games and even reviewers still annoys me. 

 

For instance in GTA V in 4K Linus 1080ti got x2 the minimum Fps compared to the 7700k, in fact better mins on all gpus than Intel. 

This is despite RyZen being rather average and below Kabylake at 720p and 1080p in GTA V. 

 

it does its the fact that it has 2 ccxs that is causing this, it seems that ryzen has low throughput between ccxs

[Valentyn]

6 minutes ago, cj09beira said:

it does its the fact that it has 2 ccxs that is causing this, it seems that ryzen has low throughput between ccxs

That shouldn't make Windows 7 give up to a 10% increase in performance, or why in the 4K GTA 5 test it giving double the minimum FPS compared to the 7700K.

There's more going than the CCX communication bandwidth.

 

 

[MoonSpot]

21 minutes ago, Valentyn said:

There's more going than the CCX communication bandwidth.

That's more or less the conclusion PCPer arrived at. https://www.pcper.com/reviews/Processors/AMD-Ryzen-and-Windows-10-Scheduler-No-Silver-Bullet

Would like to see the title of this thread get changed since it doesn't reflect the more credible view(s) on the status of these issues.

 

Even with this perf delta, which to be frank I find to be negligible in the majority of cases, I'm absolutely gunning for a ryzen build.  I don't know what the hell people were expecting to be so bitchy and feeling "betrayed" or disappointed.  I think that this release is nothing short of pants down, tits out flipping awesome.  Sure there are "things" to iron out, so what?

[Valentyn]

40 minutes ago, MoonSpot said:

That's more or less the conclusion PCPer arrived at. https://www.pcper.com/reviews/Processors/AMD-Ryzen-and-Windows-10-Scheduler-No-Silver-Bullet

Would like to see the title of this thread get changed since it doesn't reflect the more credible view(s) on the status of these issues.

 

 

Not entirely, it makes a difference yes; but it still doesn't explain the massive variance between tests of the same game; or why Windows 7 is pushing out better frames.

Still doesn't explain why at 4K GTA 5 it gives DOUBLE the minimum FPS compared to the 7700K with a GTX 1080ti, or increase minimums with all other GPUs.

 

If it was Only the CCX bandwidth why are so many reviews, and tests showing so much variation. Even when testing the exact same games.

Where the differences are motherboards as well.

 

I'd love for PCper and other publications to at least dive more into this.

Is Windows 7 not only NUMA aware, but already treating Ryzen like a 2-socket system? Why is there so much variation between gaming performance based on motherboards?

 

They even go on to disable 4 cores, and then still find Windows 10 spilling over into the second CCX causing performance drops, instead of maxing out the first CCX like it's supposed to. So despite them claiming that the Scheduler is working correctly, they just showed that it's not.

 

I just want more answers, instead of questions.

[zMeul]

4 minutes ago, Valentyn said:

I just want more answers, instead of questions.

the Zen arch design is bad

and if you treat each Zen CPU as 2 NUMAs .. well then, why do we even have this debate - it practically makes each Zen CPU a dual socket system

AMD doesn't seem to understand that core modules in CPUs is a bad idea - it was bad with Bulldozer, it's bad with Zen

[Belgarathian]

Ahhhh zMeul, I should have known you'd show up to an AMD thread. 

 

You do have valid points though, particularly about how AMD would have seen this is testing. 

[leadeater]

9 hours ago, Valentyn said:

I'd love for PCper and other publications to at least dive more into this.

Is Windows 7 not only NUMA aware, but already treating Ryzen like a 2-socket system? Why is there so much variation between gaming performance based on motherboards?

 

Window 7 isn't seeing Ryzen as a two socket system as that is not possible, Windows 7 and every other version of Windows Desktop editions only support a single socket and will disable the second CPU if one is installed.

 

Basically if this were true Ryzen in Windows 7 would only have 4 usable cores which isn't the case. 

 

Edit:

Correction, Windows Professional and higher support two sockets.

8 hours ago, zMeul said:

the Zen arch design is bad

and if you treat each Zen CPU as 2 NUMAs .. well then, why do we even have this debate - it practically makes each Zen CPU a dual socket system

AMD doesn't seem to understand that core modules in CPUs is a bad idea - it was bad with Bulldozer, it's bad with Zen

 

[Valentyn]

5 minutes ago, leadeater said:

 

Window 7 isn't seeing Ryzen as a two socket system as that is not possible, Windows 7 and every other version of Windows Desktop editions only support a single socket and will disable the second CPU if one is installed.

 

Basically if this were true Ryzen in Windows 7 would only have 4 usable cores which isn't the case.

 

I want someone to investigate why 7 is giving much better performance on some tests and games then. What's going on.

 

Also on Windows 10 the scheduler dropping drawcalls rather harshly.

 

 

[juri-han]

this would explain why disabling cores and turning it into a quad core gives 7700k levels of performance in games, linux seems to have less problems with it then windows so there is still hope for the r7 lineup for improvement in 1080p gaming other wise i hope the 1600x uses a different design and give i7s a run for their money in gaming i know the 1500x will blow the i5s out of the water since they are just a standard 4/8 and for $100 less 

[Tellos]

Did people just put maybe too much hype into those AMD benchmarks? Do people forget they do everything possible to advantage their own products? I think this may be part of the anger. I think they may have even been picky with windows versions they used.

[cj09beira]

5 hours ago, zMeul said:

the Zen arch design is bad

and if you treat each Zen CPU as 2 NUMAs .. well then, why do we even have this debate - it practically makes each Zen CPU a dual socket system

AMD doesn't seem to understand that core modules in CPUs is a bad idea - it was bad with Bulldozer, it's bad with Zen

i think they did it to have a design they can use everywhere and save a bunch of money in arranging the cores for each sku 

infinity fabric is the same, a way to save money and not have to wire every single chip (i might be wrong)

[AEG_90]

On 3/9/2017 at 5:11 AM, yian88 said:

Stop dreaming and hope threads, ryzen just isnt as good in IPC for gaming and they failed to have a smooth launch because of memory/BIOS issues.

Ryzen wont get better, because x86 cpu's are not gpu-s, they dont have drivers with millions of lines of codes and possibility to optimize drivers for each app/game.

Its either good since day one or it isnt. New bios might help with memory and windows fix with SMT a bit but thats it, the IPC will stay the same, and nothing changes the fact that the silicon from globalfoundries used in zen is complete garbage cant even OC at all.

They really need to make Zen+ release in mid 2018 and fix all the problems + better IPC for games and better silicon for OC.

You are incorrect. IPC is due to slower clock speeds. Clock for clock the Ryzen lineup R7 that is competes with the 7700k. I did tests with a 7700k and a 1700 clocked at 4.1ghz each. Of course I had to down clock the 7700k a little bit, but it works out. I was getting in margin of error of 5 to 10 fps difference in my testing. So crysis 3 hit 80 fps on average between the two cpus. The 1700 kept a higher minimum though. People are testing a higher clocked cpu the 7700k to a lower clocked ryzen at 3.6ghz or less.

[leadeater]

2 hours ago, ajkid93 said:

You are incorrect. IPC is due to slower clock speeds. Clock for clock the Ryzen lineup R7 that is competes with the 7700k. I did tests with a 7700k and a 1700 clocked at 4.1ghz each. Of course I had to down clock the 7700k a little bit, but it works out. I was getting in margin of error of 5 to 10 fps difference in my testing. So crysis 3 hit 80 fps on average between the two cpus. The 1700 kept a higher minimum though. People are testing a higher clocked cpu the 7700k to a lower clocked ryzen at 3.6ghz or less.

IPC (Instructions per Cycle) doesn't change with clock rate. Ryzen has both slightly lower IPC and much lower frequency than a 7700K. Of course that isn't the full story/picture of the situation, minimum frame rate for example and why this is the case.

[MoonSpot]

8 hours ago, Valentyn said:

--snip--

PcPer: Windows 10 Scheduling; the continuing story.

 

They clarify, expand upon, and spitball.  Ya know, chew the fat.  But, its still more questions at the end of the day then answers.

[Tellos]

@leadeater  Only know of a brief period where multi socketed CPUs boards were made best I understand multi cores ended that.The OS's required were not consumer available and required I believe specialized timing systems to ensure the CPUs worked right. It became redundant and now I doubt any OS even knows how to work with those types of systems as multi core CPUs are the norm. So I believe you are correct in that there is no way windows 7 8 8.1 or 10 are treating anything as a multi socketed CPU pair.

[leadeater]

1 minute ago, Tellos said:

@leadeater  Only know of a brief period where multi socketed CPUs boards were made best I understand multi cores ended that.The OS's required were not consumer available and required I believe specialized timing systems to ensure the CPUs worked right. It became redundant and now I doubt any OS even knows how to work with those types of systems as multi core CPUs are the norm. So I believe you are correct in that there is no way windows 7 8 8.1 or 10 are treating anything as a multi socketed CPU pair.

Multi socket systems are reserved to Windows Server editions, Windows is actually perfectly capable of using more than 1 socket but Microsoft does not allow it.

[Tellos]

@leadeater I stand corrected then. I was under the impression those server editions were not common now due to multi cores proliferation. Either way I stand by my stance your orgianol statement was correct so no harm.

[Taf the Ghost]

5 hours ago, Valentyn said:

 

I want someone to investigate why 7 is giving much better performance on some tests and games then. What's going on.

 

Also on Windows 10 the scheduler dropping drawcalls rather harshly.

 

Considering what we've seen (and the major reviewers not noticing something was amiss), my first guess is that modern controls are probably "too clever by half" for running Ryzen like an Intel CPU.  I've also seen discussion about Core Parking crop up, with Win7 and Win10 using different algos for it.  It could just be that the performance boosts we see in a lot of Win10 situations are actually causing the Ryzen bottleneck.

 

It should be noted that there's a fairly advanced linux scheduler already, and Ryzen is an utter beast at Compile, Encrypt/Decrypt and Render tasks.  AMD has a monster on their hands, it's just a matter of sorting out some of the Windows 10 issues for gaming.  The unique demands for gaming run into some sort of bottlenecks that have little to do with the practical power of the CPU and something to do with the process path.  (It's a New Architecture on a new Chipset with all brand new controllers. This has actually be a pretty solid launch so far.)

[2FA]

2 minutes ago, Tellos said:

@leadeater I stand corrected then. I was under the impression those server editions were not common now due to multi cores proliferation. Either way I stand by my stance your orgianol statement was correct so no harm.

His post wasn't entirely correct. While true Server editions support more than one socket, Pro and up versions also support dual socket while Home/Basic versions do not.

[leadeater]

13 minutes ago, DeadEyePsycho said:

His post wasn't entirely correct. While true Server editions support more than one socket, Pro and up versions also support dual socket while Home/Basic versions do not.

Thanks, didn't think the later versions of Windows desktop editions still supported more than 1 socket.

[juri-han]

2 hours ago, MoonSpot said:

PcPer: Windows 10 Scheduling; the continuing story.

 

They clarify, expand upon, and spitball.  Ya know, chew the fat.  But, its still more questions at the end of the day then answers.

really interesting video so a fix will happen it just wont happen overnight, some good news out of that video they said "it would be better if it was 6 cores" cant wait to see the 1600x benchmarks since its a 6 core has he same clock speed at the 1800x 3.6 turbo 4.0 hopefully it will come out of the gate swinging 

[xAcid9]

5 minutes ago, juri-han said:

really interesting video so a fix will happen it just wont happen overnight, some good news out of that video they said "it would be better if it was 6 cores" cant wait to see the 1600x benchmarks since its a 6 core has he same clock speed at the 1800x 3.6 turbo 4.0 hopefully it will come out of the gate swinging 

Why it would better with 6 cores? It still have 2 CCX iinm.

 

Btw does 4C/8T Ryzen 1500/1500x only have 1 CCX or it come with 2 cutdown CCX?