Windows 10 Scheduler bug causes poor RyZen performance

[leadeater]

3 minutes ago, xAcid9 said:

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?

I would do 1 CCX since it should be cheaper and no cross CCX issues but then it could all depend on how many partial faulty CCX's there are and how well 6 core CPUs sell to use those faulty CCXs with disabled cores.

[Techstorm970]

On 3/9/2017 at 3:32 AM, Space Reptile said:

running a game on a proper core instead of a thread wont give a performance increase , shure bud          (ding)

 

>comparing bulldozer w/ zen          (ding)

 

unless games use more cores/ threads the better IPC will always win.....it wasn't meant to compete [with] the 77k to begin [with]          (ding)

You would be great at CinemaSins! 

Or should I say...TechnologySins. 

[JoeCoke]

Nice to know M$ is on the issue. I wonder how much of a performance increase we'll get, most likely really small, but still.

[xAcid9]

24 minutes ago, leadeater said:

I would do 1 CCX since it should be cheaper and no cross CCX issues but then it could all depend on how many partial faulty CCX's there are and how well 6 core CPUs sell to use those faulty CCXs with disabled cores.

I'll be piss if they release 2 version of 4C/8T proc. One with 1 CCX and another one with 2 CCX.

[juri-han]

49 minutes ago, xAcid9 said:

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?

not sure but they literally said  in that video "it would be better if it was only 6 cores" my guess each each time data is sent around the cpu it adds up so less cores less delay.... i hope this issue is resolved by the time the 1600x comes out but if not it should have better performance... im in dire need of a new CPU and was holding off for ryzen 

[LAwLz]

1 hour ago, techstorm970 said:

You would be great at CinemaSins! 

Or should I say...TechnologySins. 

Except he was wrong... Actually, that would be similar to CinemaSins!

 

1 hour ago, juri-han said:

not sure but they literally said  in that video "it would be better if it was only 6 cores" my guess each each time data is sent around the cpu it adds up so less cores less delay.... i hope this issue is resolved by the time the 1600x comes out but if not it should have better performance... im in dire need of a new CPU and was holding off for ryzen 

I haven't watched the video so I don't know the context of the "it would be better if it was only 6 cores", but the 6 core Ryzen will still have the issue of high latency when one CCX needs to talk to another CCX. They might be able to do a revision to the architecture which improves it, or make software try and avoid it as much as possible (which judging by the first round of test results is already happening). But in the end, it is a hardware issue caused by AMD trying to save money.

 

 

I am surprised Nvidia fanboys haven't already started calling Ryzen a "quad core" just like the 970 "3.5GB" meme.

[leadeater]

5 minutes ago, LAwLz said:

I am surprised Nvidia fanboys haven't already started calling Ryzen a "quad core" just like the 970 "3.5GB" meme.

Ryzen 4.4 cores or 4 core Crossfire .

 

I'll be interested to see what changes are made in Zen+ in regards to CCX interconnection.

[LAwLz]

2 hours ago, leadeater said:

Ryzen 4.4 cores or 4 core Crossfire .

 

I'll be interested to see what changes are made in Zen+ in regards to CCX interconnection.

Pretty sure AMD said that they know about several areas where they have "easy gains" for Zen+. I wouldn't be surprised if the CCX interconnector is one of those areas.

[juri-han]

6 hours ago, leadeater said:

Ryzen 4.4 cores or 4 core Crossfire .

 

I'll be interested to see what changes are made in Zen+ in regards to CCX interconnection.

haha i have to admit the first thing i thought of when i heard this was CF CPU...... i hope this is just early adopter problems intel has been way to complacent over the last decade, rumors of the i7 going from a quad core to a core to 6 and HT coming to i5 just shows they need a good kick in the ass  (also rumors of skylake-x having a 12 core but id settle for a more reasonable launch price) you want to talk about adopter problems take a look back at the x99 .... i hope the x299 fairs much better

 

i hope AMD can have these issues worked out before the 6 and 4 core launch let alone zen+ because those are going to be the determining factor if ryzen was a hit or a flop but im afraid most of this rest of MS shoulders 

[4960X]

why wasn't this issue addressed before Ryzen was released?

[sof006]

44 minutes ago, 4960X said:

why wasn't this issue addressed before Ryzen was released?

AMD tight lipped on blueprints, specs etc maybe? Didn't want Microsoft to "accidentally" leak info to the market. I'm not sure, highly speculative. 

 

I'm sure its something to do with AMD also wanting to focus on making the hardware the best it is and possibly didn't have time to fiddle about sorting out getting it optimal with Windows etc

[LAwLz]

1 hour ago, 4960X said:

why wasn't this issue addressed before Ryzen was released?

45 minutes ago, sof006 said:

AMD tight lipped on blueprints, specs etc maybe? Didn't want Microsoft to "accidentally" leak info to the market. I'm not sure, highly speculative.

Fix what?

Making the scheduler SMT aware (which is the what original post was about)? It already is. There is nothing to fix in that regard. Windows 10's scheduler is optimized for Ryzen's SMT implementation already. (I have no idea why this thread hasn't been renamed, and the OP changed).

 

The high latency between the CCX? That is a hardware issue. Can't be fixed without reworking the physical CPU. Hopefully it will get better (or maybe even fixed) with Zen+ next year.

 

Windows possibly not being aware of the hardware limitations imposed by AMD's design, and trying to avoid the pitfalls? Because it would require a significant rework of how the scheduler works and that's not a small thing to do. If Microsoft was even aware of what AMD was planning (which I would doubt if the rumors about motherboard manufacturer being kept in the dark for so long), they most likely still haven't had time with hardware on hand to test things.

 

 

1 hour ago, sof006 said:

I'm sure its something to do with AMD also wanting to focus on making the hardware the best it is and possibly didn't have time to fiddle about sorting out getting it optimal with Windows etc

That's not how it works. It's not like they have a team of 5 people who does everything. They have people dedicated to different things. The people who designed the interconnector are most likely not the same people who would be helping Microsoft write code for the scheduler, just like it's not the doctors at a hospital who cook the food in the cafeteria. You can't blame the bad food on "we were too busy operating on our patients".

[juri-han]

37 minutes ago, LAwLz said:

 

The high latency between the CCX? That is a hardware issue. Can't be fixed without reworking the physical CPU. Hopefully it will get better (or maybe even fixed) with Zen+ next year.

 

again watch the video, they said it could be improved on the windows side but wont be easy.... im not sure how the 6 core will work if its a quad core paired with a duo core it will still have the same CCX lag but to less a degree, but since there is no duo core maybe its unique? the 1500x should be a single quad core and wont have any issues 

[LAwLz]

6 minutes ago, juri-han said:

again watch the video, they said it could be improved on the windows side but wont be easy.

Read the next paragraph in my post. It addresses that point specifically.

The high latency can not be fixed. What can be fixed is that Windows can try to avoid needing to data between the CCXs, but that's no easy thing to do.

 

8 minutes ago, juri-han said:

im not sure how the 6 core will work if its a quad core paired with a duo core it will still have the same CCX lag but to less a degree, but since there is no duo core maybe its unique? the 1500x should be a single quad core and wont have any issues 

I just watched the video and I did not hear them say the 6 core would not have this issue. Do you have a timestamp?

As for how the 6 core will work, my guess is that they will just take a defect CCX, disable two cores and then use that along with a fully functioning one. So you have a chip with some broken cores, but 6 fully working ones.

[Mira Yurizaki]

I don't want to bring up a sore point or a beaten horse, but I did see that part about how "Windows 7 performs better than Windows 10"

 

Windows XP performs better than Windows 7. Windows 98 performs better than Windows XP. MS-DOS outperforms them all.

 

ergo, we should be using MS-DOS.

[79wjd]

51 minutes ago, LAwLz said:

Read the next paragraph in my post. It addresses that point specifically.

The high latency can not be fixed. What can be fixed is that Windows can try to avoid needing to data between the CCXs, but that's no easy thing to do.

 

I just watched the video and I did not hear them say the 6 core would not have this issue. Do you have a timestamp?

As for how the 6 core will work, my guess is that they will just take a defect CCX, disable two cores and then use that along with a fully functioning one. So you have a chip with some broken cores, but 6 fully working ones.

Wasn't there an update for some Linux distroy that drastically improved on the ccx issue? 

 

As for the six core...I think it's also likely it could be two modules with one core disabled on each. For that matter, the quad core could be the same way depending on yields. 

[AnonymousGuy]

1 hour ago, LAwLz said:

Windows possibly not being aware of the hardware limitations imposed by AMD's design, and trying to avoid the pitfalls? Because it would require a significant rework of how the scheduler works and that's not a small thing to do. If Microsoft was even aware of what AMD was planning (which I would doubt if the rumors about motherboard manufacturer being kept in the dark for so long), they most likely still haven't had time with hardware on hand to test things

I wouldn't hold my breath waiting for MSFT to dedicate the resources to do this.  Why should they when this series of CPUs probably will be single digit % of the Windows market, and an even smaller percent will even notice anyways.  If it were me I'd have it in the backlog of shit to do which basically means it's never happening but I can claim it's being "worked on".

[AlwaysFSX]

15 hours ago, leadeater said:

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.

Except Windows wouldn't detect this as a multi-socket system because all it's seeing is the 8 cores it's given. The "socket" is handled by the CPU, not the motherboard and therefore not Windows.

 

Someone can correct me if I'm wrong but that would explain why all the cores + threads show up in Win7.

[LAwLz]

41 minutes ago, djdwosk97 said:

Wasn't there an update for some Linux distroy that drastically improved on the ccx issue? 

Pretty sure you're thinking of the Linux patch that fixed an issue where the scheduler could not differentiate between a real core and an SMT core. That was never an issue on Windows though (despite what Reddit said and users on this forum repeating it over, and over and over again).

 

43 minutes ago, djdwosk97 said:

As for the six core...I think it's also likely it could be two modules with one core disabled on each. For that matter, the quad core could be the same way depending on yields. 

Oh right. One core disabled on each would make more sense. Or maybe it will be a mix.

[Prysin]

11 hours ago, LAwLz said:

Except he was wrong... Actually, that would be similar to CinemaSins!

 

I haven't watched the video so I don't know the context of the "it would be better if it was only 6 cores", but the 6 core Ryzen will still have the issue of high latency when one CCX needs to talk to another CCX. They might be able to do a revision to the architecture which improves it, or make software try and avoid it as much as possible (which judging by the first round of test results is already happening). But in the end, it is a hardware issue caused by AMD trying to save money.

 

 

I am surprised Nvidia fanboys haven't already started calling Ryzen a "quad core" just like the 970 "3.5GB" meme.

Ryzen is a dual core. Obviously. 2 big clusters with "AMD cores" inside.... come on man, get the facts straight!!!

[leadeater]

42 minutes ago, AlwaysFSX said:

Except Windows wouldn't detect this as a multi-socket system because all it's seeing is the 8 cores it's given. The "socket" is handled by the CPU, not the motherboard and therefore not Windows.

 

Someone can correct me if I'm wrong but that would explain why all the cores + threads show up in Win7.

No that is correct, that was basically the point. If AMD somehow designed a CPU that showed up to Windows as a two socket processor anyone with Windows Home edition could only use half the processor, this isn't the case and AMD wouldn't do that.

 

However the motherboard does handle the socket reporting to Windows, just like it tells Windows which PCIe slot a device is in etc.

[Tomsen]

13 minutes ago, Prysin said:

Ryzen is a dual core. Obviously. 2 big clusters with "AMD cores" inside.... come on man, get the facts straight!!!

AMD IPC almost 4x Intels! Buy stock now!!!

[Mr.Meerkat]

On 09/03/2017 at 10:11 AM, yian88 said:

the silicon from globalfoundries used in zen is complete garbage cant even OC at all.

4.3GHz with the 6900K...ermmm...I think Intel is letting you down a little bit with that statement...

[Prysin]

27 minutes ago, Tomsen said:

AMD IPC almost 4x Intels! Buy stock now!!!

Intel 4x more cores then AMD. 4x worse IPC. "MOAR CORES" lolz

[SaladFingers]

So, If Windows goes to sleep, I will be able to run Crysis afterwards?

 

meh, wrong thread sorry