Microsoft Confirms Windows Issue is Holding back AMD's Ryzen?

[Zarathustra[H]]

Well, not so fast. Guru3D has a story to this effect running today, and while it does look like Microsoft's statements suggest that they are looking to patch things for Ryzen, I have never seen anyone put so many weasel words into a single tweet before. That's impressive considering the 140 character limit.

They do state that there seem to be some puzzling issues with Windows detecting an incorrect cache size on Ryzen though, as well as some issues with 0.25x multipliers, so it does seem like there is a legitimate need for at least some patching on Microsoft's part, but it is unclear to us what real performance impact this might have. We have asked Microsoft for clarification, and will update you with what we hear.

"Back in the days when Intel introduced Hyper-threading many similar issues have been addressed. From here on-wards it will be a waiting game as we'll have to wait and see how Microsoft will address this matter and see what effect these two factors will have on the slightly disappointing game performance results. But sure, it is terrific news that Microsoft found an issue and is correcting it, let us hope that Microsoft sorts out the scheduler and cache issues as soon as possible. Maybe the upcoming Patch Tuesday update scheduled for March 14 will bring this coveted patch."

 

[Trimlock]

My guess its .01% on everything and .5% on cache starved programs.

 

[naib]

"windows10 only"

 

[ole-m]

My guess its .01% on everything and .5% on cache starved programs.

gained 10% disabling SMT... Intel lost 7% on their 8 core, hmm, I think there's bigger gains to be done.
I doubt the ryzen will be a better CPU than Intel regardless.

 

[Shmee]

¯\_(ツ)_/¯ I am sure there are things to work on, but I doubt there will be any major gains. Guru3D is looking for some clicks, and making things look worse than they are is a good way to do that. At least Windows is stable with Ryzen, unlike Linux. Slightly buggy stuff is one of the problems with being an early adopter. If you need things to work perfectly, you wait a couple months.

 

[Trimlock]

gained 10% disabling SMT... Intel lost 7% on their 8 core, hmm, I think there's bigger gains to be done.
I doubt the ryzen will be a better CPU than Intel regardless.

You are talking about individual applications that were ran through the Intel compiler. This is a windows "scheduler" patch, only instead of correctly assigning threads to cores to not jam up the FPU's. I'm assuming they are correcting cache recognition since its broken into two separate blocks for AMD and this won't have much if anything to do with the 1080p gaming benchmarks.

 

[Zarathustra[H]]

¯\_(ツ)_/¯ I am sure there are things to work on, but I doubt there will be any major gains. Guru3D is looking for some clicks, and making things look worse than they are is a good way to do that. At least Windows is stable with Ryzen, unlike Linux. Slightly buggy stuff is one of the problems with being an early adopter. If you need things to work perfectly, you wait a couple months.

Everything I read is that Linux is fine if you have a recent enough kernel. If you just install Ubuntu with the latest kernel in their repository though, yes, things will be unstable and crash.

Right now Linux requires installing a 4.10 kernel for stability with Ryzen, and for most distributions that means a manual install. (yes, I know, an handful already have 4.10, but not the most used ones)

Once Ubuntu 17.04 comes out, no manual install will be needed anymore

 

[naib]

Everything I read is that Linux is fine if you have a recent enough kernel. If you just install Ubuntu with the latest kernel in their repository though, yes, things will be unstable and crash.

Right now Linux requires installing a 4.10 kernel for stability with Ryzen, and for most distributions that means a manual install. (yes, I know, an handful already have 4.10, but not the most used ones)

Once Ubuntu 17.04 comes out, no manual install will be needed anymore

4.10 is only part of the story.
https://git.kernel.org/cgit/linux/k.../?id=08b259631b5a1d912af4832847b5642f377d9101

a33d331761bc ("x86/CPU/AMD: Fix Bulldozer topology") our SMT scheduling topology for Fam17h systems is broken, because the ThreadId is included in the ApicId when SMT is enabled. So, without further decoding cpu_core_id is unique for each thread rather than the same for threads on the same core. This didn't affect systems with SMT disabled. Make cpu_core_id be what it is defined to be.

4.11 will be needed for the ALC1200 audio support (the chip on the newer kaby and ryzen mobo's )

 

[fuzzylogik]

other rumors were that the last skipped Patch Tuesday was specifically to get this fixed. Lol. we'll see. Having the OS think that the computer has ~6x (? what was it? 130ish MB reported?) more cache over what it actually has can't be good. It'll be nice to have it working as it should though and then go from there. Don't expect much then you shouldn't be disappointed lol.

 

[ole-m]

Everything I read is that Linux is fine if you have a recent enough kernel. If you just install Ubuntu with the latest kernel in their repository though, yes, things will be unstable and crash.

Right now Linux requires installing a 4.10 kernel for stability with Ryzen, and for most distributions that means a manual install. (yes, I know, an handful already have 4.10, but not the most used ones)

Once Ubuntu 17.04 comes out, no manual install will be needed anymore

It seems a lot better to me, but I think CCX walking is still an issue and not thought of yet, and increasing latencies to the L3 cache as L3 cache is split per CCX.
Core 0-3 shares L3 cache of 8MB.
Core 4-7 shares L3 cache of 8MB.

If Core 0 load in single threaded game is moved to core 6 then the core 6 would utilize L3 cache in CCX1 instead of CCX2 which is nearer.
CPU probably identifies this issue and moves cache to CCX2, OS then moves load to Core 3 - repeat.

This is why allcore load is off the charts, single thread is pretty bad..

 

[naib]

It seems a lot better to me, but I think CCX walking is still an issue and not thought of yet, and increasing latencies to the L3 cache as L3 cache is split per CCX.
Core 0-3 shares L3 cache of 8MB.
Core 4-7 shares L3 cache of 8MB.

If Core 0 load in single threaded game is moved to core 6 then the core 6 would utilize L3 cache in CCX1 instead of CCX2 which is nearer.
CPU probably identifies this issue and moves cache to CCX2, OS then moves load to Core 3 - repeat.

This is why allcore load is off the charts, single thread is pretty bad..

do you have aggressive load balancer set, ie for energy saving? this will move processes around ALOT to make use of the lowest number of cores (so the remaining can go into deep sleep). Couple this with an aggressive cpu govenor (ie onDemand... ) and this could explain that

 

[cyclone3d]

It seems a lot better to me, but I think CCX walking is still an issue and not thought of yet, and increasing latencies to the L3 cache as L3 cache is split per CCX.
Core 0-3 shares L3 cache of 8MB.
Core 4-7 shares L3 cache of 8MB.

If Core 0 load in single threaded game is moved to core 6 then the core 6 would utilize L3 cache in CCX1 instead of CCX2 which is nearer.
CPU probably identifies this issue and moves cache to CCX2, OS then moves load to Core 3 - repeat.

This is why allcore load is off the charts, single thread is pretty bad..

So enabling "core parking" should alleviate that, right?

 

[Nickbaldwin86]

these issues couldn't have been resolved pre-release?

What OS did they develop for if not Windows, the OS with the most market share hands down, even if it is only Windows X or Y they shouldn't have these kinds of issues with such a huge platform. Very concerning for such a promising CPU generation for AMD.

 

[Pusher of Buttons]

these issues couldn't have been resolved pre-release?

What OS did they develop for if not Windows, the OS with the most market share hands down, even if it is only Windows X or Y they shouldn't have these kinds of issues with such a huge platform. Very concerning for such a promising CPU generation for AMD.

I'm talking out of my ass here....but....

My guess is they worked with MS by giving them the chip design and feature-set and MS said "Sure, we can work with that"

AMD can't just straight copy Intel's design (anymore) so there HAS to be some difference in the chip, and any small difference there is going to lead to optimization issues potentially at the OS level. It's not like the chip can't handle x86 instructions and run Windows, it's just missing some optimizations.

 

[SvenBent]

So enabling "core parking" should alleviate that, right?

Anything using less than 25% of the total CPU can be affinity locked to one CCX and disabled SMT if needed bee.

my software already "disables" smt on any forgrodun applications which have helpes older games on a SMT CPU.
im working on a more smart method to due it that will place threads of process in a more optimal way to avoid CCX cache issues as well as SMT.

SMT is not just a performance penalty on AMD. There ae situations it is on even the newest Intel CPU's in windows10 on lightly multithreaded load.
Its the nature of SMT to have this penalty in this specific situation.

 

[Shmee]

Everything I read is that Linux is fine if you have a recent enough kernel. If you just install Ubuntu with the latest kernel in their repository though, yes, things will be unstable and crash.

Right now Linux requires installing a 4.10 kernel for stability with Ryzen, and for most distributions that means a manual install. (yes, I know, an handful already have 4.10, but not the most used ones)

Once Ubuntu 17.04 comes out, no manual install will be needed anymore

I wasn't bagging on Linux. I was just using it as an example to prove my point that early adopters always have to deal with issues. Yes if you grab the new kernel things seem to be running pretty well, but even then it will take a while for the processer drivers to reach their full potential.

 

[Stimpy88]

You cannot tell me that AMD have not seen any of these issues since they had working silicon back from the fabs almost a year ago?

So MS has only just now been informed of the compatibility issues Windows has with Ryzen? Bullshit... But IF this is true, then AMD deserves all the trouble it gets.

However, when Windows gets fixed, and the awful motherboard BIOS issues get properly fixed, then Ryzen will be in my next build.

 

[buzzbomb]

I'm talking out of my ass here....but....

My guess is they worked with MS by giving them the chip design and feature-set and MS said "Sure, we can work with that"

AMD can't just straight copy Intel's design (anymore) so there HAS to be some difference in the chip, and any small difference there is going to lead to optimization issues potentially at the OS level. It's not like the chip can't handle x86 instructions and run Windows, it's just missing some optimizations.

You're right - you are talking out your ass.

 

[Stimpy88]

I'm talking out of my ass here....but....

My guess is they worked with MS by giving them the chip design and feature-set and MS said "Sure, we can work with that"

AMD can't just straight copy Intel's design (anymore) so there HAS to be some difference in the chip, and any small difference there is going to lead to optimization issues potentially at the OS level. It's not like the chip can't handle x86 instructions and run Windows, it's just missing some optimizations.

Hmmm, knowing AMD, they have probably not got round to talking to MS about the Windows issues until a few days or weeks ago.

And I don't think I would ever say that AMD just copies Intel CPU's. Actually, if they did copy Intel (not that Intels lazy old designs are any kind of gold standard, other than we were forced to buy them due to typical AMD incompetence), they would have not almost gone bankrupt over the last few years, would they?

 

[Pusher of Buttons]

You're right - you are talking out your ass.

At least I'm honest...if I talked out of my ass and didn't admit it than that would make me a YouTube reviewer

Hmmm, knowing AMD, they have probably not got round to talking to MS about the Windows issues until a few days or weeks ago.

And I don't think I would ever say that AMD just copies Intel CPU's. Actually, if they did copy Intel (not that their lazy old designs are any kind of gold standard, other than we were forced to buy them due to typical AMD incompetence), they would have not almost gone bankrupt over the last few years, would they?

And AMDs biggest problem was terrible business decisions less so than bad tech...until Bulldozer...and that was just a swing and a miss. Intel did that with the P4 but had the marketing clout and dirty tricks department to recover from it....AMD didn't have nearly the resources to have a bad chip. At some point Intel has effectively been using the Pentium Pro architecture for 20 years....they got really lucky and designed an incredibly scalable and flexible design that's carried them. But even they don't seem to have any What's Next? products. The fact that AMD can compete at all at this level is shocking to me. Ryzen even with some teething troubles is a monumental achievement considering where they were a few years ago. No doubt in the process, though, a lot of the rough edges get missed and optimizations are obviously one of them.

 

[DukenukemX]

Windows is holding back computers systems in general, but that's a discussion for another topic.

 

[geok1ng]

they would have not almost gone bankrupt over the last few years, would they?

i kindly remind you that AMD went (almost) bankrupt in a time when they had a better chip than Intel, things would be much worse if they played copycat game. AMD can not win, AMD can not draw the game and AMD can (should not) not forfeit the game.

 

[Stimpy88]

i kindly remind you that AMD went (almost) bankrupt in a time when they had a better chip than Intel, things would be much worse if they played copycat game. AMD can not win, AMD can not draw the game and AMD can (should not) not forfeit the game.

Very true, AMD had a Commodore level of management incompetence. AMD still reeks of a company that does not quite know what it's doing, and knows even less about what it's upper middle management are up to.

To be honest I'm amazed that they are still in business, but I'm damn glad they are, because Intel and Nvidia are pretty much the devil as it is, and really would be the devil if AMD ceased trading.

However, to be honest, I'm still expecting AMD to throw it all away over the next few years. Zen 2.0 needs to totally kick ass, not just in some situations but in all, and I'm not sure they have the intelligence (management, not engineering) to do it.

 

[Pusher of Buttons]

Very true, AMD had a Commodore level of management incompetence. AMD still reeks of a company that does not quite know what it's doing, and knows even less about what it's upper middle management are up to.

To be honest I'm amazed that they are still in business, but I'm damn glad they are, because Intel and Nvidia are pretty much the devil as it is, and really would be the devil if AMD ceased trading.

You gotta remember...there was a handfull of little companies orbiting the x86 behemoth that was Intel in the 90s...AMD, Cyrix, NexGen...I don't think AMD was really prepared for the comparative success that hit them with Athlon. They'd been around for decades as a small fry and suddenly within a couple years became the sole competition to arguably one of the most landscape altering companies of the last 30 years. I don't think the company knew how to properly manage the new scale they were producing at. They still had good brains in the tech department, but I don't ever think the top management knew how to capitalize on it. Don't forget the dirty tricks department at Intel was disastrous behind the scenes....the $1.5 billion fine they paid in all likelihood cost AMD tens if not hundreds of billions in revenue long term. It was a bargain....and I wouldn't put it past them to do it again.

 

[Stimpy88]

You gotta remember...there was a handfull of little companies orbiting the x86 behemoth that was Intel in the 90s...AMD, Cyrix, NexGen...I don't think AMD was really prepared for the comparative success that hit them with Athlon. They'd been around for decades as a small fry and suddenly within a couple years became the sole competition to arguably one of the most landscape altering companies of the last 30 years. I don't think the company knew how to properly manage the new scale they were producing at. They still had good brains in the tech department, but I don't ever think the top management knew how to capitalize on it. Don't forget the dirty tricks department at Intel was disastrous behind the scenes....the $1.5 billion fine they paid in all likelihood cost AMD tens if not hundreds of billions in revenue long term. It was a bargain....and I wouldn't put it past them to do it again.

Brings back good memories of the choices we all had in those days! AMD had a great product with the Athlon series, and it was so sad to see them throw it all away. The stars really aligned in their favour when Intel was trying to trick us with how amazing their Netburst architecture was, and how GHz was king. I loved seeing Intel getting its ass handed to them, also because we only got the Core architecture because of AMD, and Intel have been reaping the benefits of that ever since.

I remember all too well Intel's business tactics of the time. I was subjected to many "training" days with Intel sales and marketing, and pressured to shit on AMD based products, not stock them, and if we did, there were rules regarding how AMD systems could be displayed, and their physical proximity to Intel based systems, and even the specifications of them, so as to not embarrass Intels products. I'm sure those days never ended in many board rooms of many companies, but they did mostly end in retail.

However, you can be sure Intel are already opening boxes of new tricks to unleash on Microsoft, motherboard manufacturers, chipset manufacturers, software companies etc...

 

[geok1ng]

Intel and Nvidia are pretty much the devil as it is, and really would be the devil if AMD ceased trading.

the age of my signature rig displays how evil Intel is. NVIDIA on the other hand kept offering costumers faster and cheaper products over time not the "5% faster every 2 years" crap Intel has been pulling since the 2600K.

 

[bigdogchris]

"windows10 only"

I brought up before it would be interesting to see Windows 7 vs Windows 10 benchmarks with Ryzen (because of the Windows 7 official compatibility issue). That would of exposed any issues like this.

 

[Caned]

All I want is ryzen to beat my 4790k by even as little as 2% (in games) and I will jump ship. I used to love the old t-birds and original AXP, I am a bit of a sucker for AMD hardware, always hoping they beat big bad Intel and nvidia lol. (this gives me an excuse to "upgrade" and give my 12 year old a decent gaming pc lol)

So hopefully software patches and better revisions of cpu and mobos in the near future fix these teething problems and I get to feel better about supporting the underdog. One can dream.

(damn autocorrect, edit, edit, edit)

 

[polonyc2]

other rumors were that the last skipped Patch Tuesday was specifically to get this fixed. Lol. we'll see...

conspiracy theories!!...I love it

 

[fuzzylogik]

conspiracy theories!!...I love it

Fairly amazing how many ideas get passed around on stuff like this lol

 

[Stimpy88]

the age of my signature rig displays how evil Intel is. NVIDIA on the other hand kept offering costumers faster and cheaper products over time not the "5% faster every 2 years" crap Intel has been pulling since the 2600K.

It sometimes pays to look at the bigger picture sometimes...

 

[Schtask]

I'm talking out of my ass here....but....

My guess is they worked with MS by giving them the chip design and feature-set and MS said "Sure, we can work with that"

AMD can't just straight copy Intel's design (anymore) so there HAS to be some difference in the chip, and any small difference there is going to lead to optimization issues potentially at the OS level. It's not like the chip can't handle x86 instructions and run Windows, it's just missing some optimizations.

I'm not tracking what you're putting down....

If AMD ever "straight copied Intel's designs" how would we have ever had the glorious Athalon 64 or the inglorious Bulldozer? I don't think the architecture of any AMD CPU ever copied an Intel design.

Just from a logical perspective, the performance envelopes of either companies products are drastically different compared to one another in a timeline... Until now anyway.

If they were copies, performance measurements would be comparable more often.

 

[elzeus]

Microsoft will be all over this fix they need it to work at maximum efficiency for when they drop Xbox Scorpio which may be using a (low end) Ryzen cpu.

 

[ChadD]

I'm not tracking what you're putting down....

If AMD ever "straight copied Intel's designs" how would we have ever had the glorious Athalon 64 or the inglorious Bulldozer? I don't think the architecture of any AMD CPU ever copied an Intel design.

Just from a logical perspective, the performance envelopes of either companies products are drastically different compared to one another in a timeline... Until now anyway.

If they were copies, performance measurements would be comparable more often.

Well ummm X86 was sort of not invented by AMD.

Yes AMD started as one of demm er cloners.

In Feb 1982 AMD signed a deal with Intel to produce 8086 and 8088 chips. IBM had a policy of requiring 2 sources for chips. Later in 86 I think it was Intel tried to cancel their deal around the time of the 386... AMD sued and won, Intel challanged and it went all the way to the supreme court and wasn't settled until 92. After that AMD had to jump through hoops to prove they had the rights to use their X86 microcode, and had to clean room the code. This lead to Intel pushing out 486 (as AMD was making 386s)... and when Intel couldn't patent "486" and other companies started putting out chips with 4s in them the Pentium™ followed.

So yes AMD started as a second source chip manufacturer so Intel could sell to IBM. After years of legal wrangling back engineered their version of x86 microcode. Intel should thank AMD on many levels. If it wasn't for AMD we would likely have never had dancing Chip engineer Pentium commercials, never mind 64bit chips, mmx... and mmx 2 3 4 5 sse sse 2 3 4 ect ect. If they had dropped out after the Athlon we would all likely be running 8ghz Pentium 6s that would be doing less work then our ARM based phones. AMD has always kept Intel from just sitting back and manufacturing the same thing over and over. On a personal note I would have had to live with out Weird Al singing about pentium chips, my life would have been lesser.

So yes up until 386 they where clone makers. After that they tried marketing their own names and jazzed up instruction sets like K6 with 3Dnow (vs pentium and pentium mmx)... and then the faithful Athlon which put the beat down on the PII. The entire thing is doubly funny when you consider that Intel switched the PII to a stupid slot style socket design they could patent to keep AMD chips from ending up in motherboards designed for Intel chips. (the K6 ran in Pentium socket 7 boards) So AMD being forced to create their own chipset brought us the Athlon. If Intel hadn't built such a massive War Che$t and been willing to play dirty things could have went very different.

 

[DPI]

Microsoft will be all over this fix they need it to work at maximum efficiency for when they drop Xbox Scorpio which may be using a (low end) Ryzen cpu.

HUH?

 

[BulletDust]

Well ummm X86 was sort of not invented by AMD.

Yes AMD started as one of demm er cloners.

In Feb 1982 AMD signed a deal with Intel to produce 8086 and 8088 chips. IBM had a policy of requiring 2 sources for chips. Later in 86 I think it was Intel tried to cancel their deal around the time of the 386... AMD sued and won, Intel challanged and it went all the way to the supreme court and wasn't settled until 92. After that AMD had to jump through hoops to prove they had the rights to use their X86 microcode, and had to clean room the code. This lead to Intel pushing out 486 (as AMD was making 386s)... and when Intel couldn't patent "486" and other companies started putting out chips with 4s in them the Pentium™ followed.

So yes AMD started as a second source chip manufacturer so Intel could sell to IBM. After years of legal wrangling back engineered their version of x86 microcode. Intel should thank AMD on many levels. If it wasn't for AMD we would likely have never had dancing Chip engineer Pentium commercials, never mind 64bit chips, mmx... and mmx 2 3 4 5 sse sse 2 3 4 ect ect. If they had dropped out after the Athlon we would all likely be running 8ghz Pentium 6s that would be doing less work then our ARM based phones. AMD has always kept Intel from just sitting back and manufacturing the same thing over and over. On a personal note I would have had to live with out Weird Al singing about pentium chips, my life would have been lesser.

So yes up until 386 they where clone makers. After that they tried marketing their own names and jazzed up instruction sets like K6 with 3Dnow (vs pentium and pentium mmx)... and then the faithful Athlon which put the beat down on the PII. The entire thing is doubly funny when you consider that Intel switched the PII to a stupid slot style socket design they could patent to keep AMD chips from ending up in motherboards designed for Intel chips. (the K6 ran in Pentium socket 7 boards) So AMD being forced to create their own chipset brought us the Athlon. If Intel hadn't built such a massive War Che$t and been willing to play dirty things could have went very different.

The whole damn PC platform was reverse engineered! it's not like ASUS, Gigabyte and MSI suddenly just decided on the common x86 platform.

AMD processors are x86/64 compatible, that doesn't however indicate they're exact clones.

 

[Uvaman2]

The whole damn PC platform was reverse engineered! it's not like ASUS, Gigabyte and MSI suddenly just decided on the common x86 platform.

AMD processors are x86/64 compatible, that doesn't however indicate they're exact clones.

? (in general)

and also
AMD created x86-64.

 

[SvenBent]

I'm not tracking what you're putting down....

If AMD ever "straight copied Intel's designs" how would we have ever had the glorious Athalon 64 or the inglorious Bulldozer? I don't think the architecture of any AMD CPU ever copied an Intel design.

Just from a logical perspective, the performance envelopes of either companies products are drastically different compared to one another in a timeline... Until now anyway.

If they were copies, performance measurements would be comparable more often.

Kinda forgettinh that Intel had to share CPU design with AMD due to their IBM contract. and there was plenty of of suing from AMD side because Intel on purpose "Forgot" some part of the blueprints.
So yes there has been a time where AMD straight up copied Intels designs.

Your timeline you are referring too seems to consist of a very short time of the real timeline. and totally neglectin the 386 468 era

 

[ChadD]

The whole damn PC platform was reverse engineered! it's not like ASUS, Gigabyte and MSI suddenly just decided on the common x86 platform.

AMD processors are x86/64 compatible, that doesn't however indicate they're exact clones.

No not at all. I was simply stating their history. In their 8086/88 days yes they where exact clones. They are responsible for all the innovation since after their fallout with Intel. After 92 and the intel v amd supreme court AMD clean roomed x86 and from their between themselves and the push they put on intel they drove every bit of x86 tech we have today.

 

[N4CR]

the age of my signature rig displays how evil Intel is. NVIDIA on the other hand kept offering costumers faster and cheaper products over time not the "5% faster every 2 years" crap Intel has been pulling since the 2600K.

Always wondered if the 2500/2600k got someone a major pay cut and or firing.... I should ask my contacts there and see.

Seems since then they have not really released anything 'UBERCONCMEGAWOW!'.

Ryzen is far more exciting but I'm wondering about picking up an Asrock + 1700 to tie me over until naples/opteron WS are out. I hope the opterons are binned better/better sillicon and can be OC'd per core. I'd run the 16 core at 4-8 cores and see where it goes. My old opteron 164 went harder overclocking than most 2600ks... around 60% OC ON AIR.