Ryzen: Strictly technical

[Justinbaileyman]

This might be believable, if linux hadn't fixed theirs in less than a week. But that fact that they did makes me think this should not take alot of man hours for MS to fix. Something else is going on here.

Something else is going on here? Um Yeah... MS said they aren't gonna fix diddly squat. The reason is cause it would brick the code they have written for Intel cpu's is my best guess.All I need to know is when is Ubuntu updated with the proper drivers to run my Ryzen. I just can't see my self supporting a company like MS any longer if all they are gonna do is cater to the rich guy.

 

[Justinbaileyman]

They don't have the engineers to spare to microsoft, but they will send some to every triple a dev team?

That's the same thing I'm saying... Hits the nail flat on the head!!

 

[Snarf Snarf]

They don't have the engineers to spare to microsoft, but they will send some to every triple a dev team?

More likely have a small team that is e-mailing the devs back and forth all day working on all of the fixes at once. Granted that lots of AAA games are on similar or the exact same engine, they don't need to optimize every single title, just the most popular engines that have issues.

 

[Snarf Snarf]

This might be believable, if linux hadn't fixed theirs in less than a week. But that fact that they did makes me think this should not take alot of man hours for MS to fix. Something else is going on here.

I'm not sure if it was this thread or the main thread where it was discussed that this type of patch wouldn't be included in a small update. Considering the level of work it takes it would likely come in a major build update which would take a few months of quality testing due to how MS does things. If I recall correctly it took a long time to get Bulldozer optimizations updated into Windows 7, AMD is being proactive here and trying to fix the issues head on rather than waiting for MS to go through their incredibly slow process.

 

[Jan Olšan]

basically you are actually BETA testing.

Does Ryzen IPC look to you as being impaired by some bug/issue/erratum? It certainly doesn't look that way to me. The CPU's performance landed higher than anybody dared to hope 3 months back. Not just IPC, the clock too. In autumn, we thought we will thank God for every 100 MHz it will manage to do over 3 GHz in max boost state for single core. And we got 4.1 actually. That's actually overwhelming.

Those 2nd and 3rd revisions the mysterious insider talks about are merely the next architectural versions that will come out as regular roadmap updates with the usual yearly or longer cycles. No "magic fixed Ryzen revision". The current revision will keep selling for the year or so it was meant to. That is just how it is done, we might see 100MHz speedbump if we are lucky and the manufacturing improves.

Intel Kaby Lakes achieve higher max FPS in games - so what? That in no way suggest buggy chip. It's more or less the same with Broadwell-E. Kaby Lake manages higher memory bandwdith with fast DDR4, it has non-trivially higher IPC, and significantly higher clock to boot. Those are the things most current or past games respond to, so what else do you expect? Application software using multiple cores shows that Ryzen performance is real deal, the speed is there. I'd say at the moment there is no indication Ryzen performs bellow the plan in any area.

 

[looncraz]

They don't have the engineers to spare to microsoft, but they will send some to every triple a dev team?

Different skills needed. I doubt AMD has more than a handful of developers capable of working on a kernel scheduler - and probably none that are familiar with Windows' scheduler.

 

[DisEnchantment]

Right now, AMD already has 2nd and 3rd revision CPUs without all the nonsense and the unfortunate thing is if you bought this round, you'll likely be stuck with with performance.

Ok, for me as a BETA Tester for AMD... Which is this new silicon, Is it the Zen Ver2 or simply a newer stepping/revision of the current Zen Silicon?
I mean the Zen Ver2 lauches early next year when is this new silicon lauching? Would R5/R3 have this new revision?
I had this processor for 4 days now it is apparently outdated.
No problem though it is a great performer for me, I hardly play any game. But had I known this I would have waited.
I am a little ticked off since I hardly upgrade my PC. My PC pays for my food.

 

[Magic Hate Ball]

For Ryzen owners.
Please test Ryzen with WinRAR benchmark in single-channel mode and dual-channel mode.
We need WinRAR benchmark results with both modes: Multithreading option off / on.

With dual-channel mode:
Test 1 - WinRAR
Test 2 - WinRAR with Affinity to cores 0-7 (CCX0)
Test 3 - WinRAR with Affinity to cores 8-15 (CCX1)

Then run benchmark with single-channel mode.
You must remove one RAM module. So system will use only one channel.
Test 4 - single-channel - WinRAR
Test 5 - single-channel - WinRAR with Affinity to cores 0-7 (CCX0)
Test 6 - single-channel - WinRAR with Affinity to cores 8-15 (CCX1)

If results of single-channel tests are better, then Ryzen uses separated memory controllers with one memory controller per one CCX.

Performed @ 4.0ghz SMT enabled 3200mhz CL14 RAM dual channel.

Test 1 - 12,900-13,000
Test 2 - 7,950-8,000
Test 2.5 - 7,750 (affinity set to cores 04-11 spanning CCX)
Test 3 - 7950-8,000

If I get time later (going to go see Logan in a minute) I'll swing back and try the single channel stuff.

 

[DisEnchantment]

From here

1. Finally, as part of AMDs ongoing development of the new AM4 platform, AMD will increase support for overclocked memory configurations with higher memory multipliers. We intend to issue updates to motherboard partners in May that will enable them, on whatever products they choose, to support speeds higher than the current DDR4-3200 limit without refclk adjustments. AMD Ryzen™ processors already deliver great performance in prosumer, workstation, and gaming workloads, and this update will permit even more value and performance for enthusiasts who chose to run overclocked memory.

By the first week of April, AMD intends to provide an update for AMD Ryzen™ processors that optimizes the power policy parameters of the Balanced plan to favor performance more consistent with the typical usage models of a desktop PC.

Which kind of updates? Software patches/BIOS which update?

 

[looncraz]

I'm not sure if it was this thread or the main thread where it was discussed that this type of patch wouldn't be included in a small update. Considering the level of work it takes it would likely come in a major build update which would take a few months of quality testing due to how MS does things. If I recall correctly it took a long time to get Bulldozer optimizations updated into Windows 7, AMD is being proactive here and trying to fix the issues head on rather than waiting for MS to go through their incredibly slow process.

Microsoft wouldn't need to make major changes, really, just many small ones:

1. Change default core parking behavior to park a full CCX and half the logical cores on one CCX.
2. Reduce the load required for unparking logical cores on the first CCX.
3. Increase the load required to unpark cores on the second CCX.
4. Use only two power states: full-on, full idle.
5. Do not load balance across CCXes.

And two primary fixes (well, one fix, one work-around):

1. Fix process affinity mask restricting threads to only two cores when selecting logical cores on Ryzen.
2. Completely hide the L3 cache from programs - it's a bifurcated victim cache and should not be considered by programs as a real cache when using cache-aware algorithms.

That is enough from the OS side of things - as the OS can't be aware of each application's needs for cache locality.

AMD needs to reduce the inter-CCX penalty and the memory latency. That, alone, is worth 5~10% performance in scenarios which find those to be problematic (namely certain games).

Microsoft will never make these changes, but I do suspect they will fix the glaring affinity bug. We can emulate some of these behaviors by tweaking power management settings ourselves - and likely someone will create a driver to park the cores in an intelligent manner (it won't be me, I detest Windows programming).

In six months time, this will all be a non-issue.

 

[Elixer]

But they are tied to this for years. A few million shouldn't stop them from fixing it.

That depends, if they have something in the pipeline that mitigates the issue(s) in the first revision of Ryzen, why would you spend millions when instead, you can use that for R&D?
AMD don't exactly have cash to throw away.

That is what I basically get from AMD's PR response of, "nothing is wrong" is code for, "we don't have the funds to give to MS to fix the problem".

If they can get the "infinity fabric" to run much faster, that is a way of "hiding" one of the main issues.

 

[formulav8]

All I need to know is when is Ubuntu updated with the proper drivers to run my Ryzen. I just can't see my self supporting a company like MS any longer if all they are gonna do is cater to the rich guy.

Apparently things are getting even better for Linux on Ryzen if I'm understanding correctly.

LLVM 5.0 Should Offer Better AMD Ryzen Performance

 

[Kromaatikse]

First things first - the quad-core versions of Ryzen won't have any of this inter-CCX business, because they only have one CCX. So in that sense, R5 and R3 will be "fixed", to the extent that a "fix" is needed. And they'll be 4C/8T (or 4C/4T) with performance reliably exceeding the ever-popular Haswell i7s (or i5s) for half the price.

As for Ubuntu, there are preview editions of 17.04 available already. There's a clue in that version number - full release next month, right on their usual 6-monthly schedule.

 

[Blake_86]

First things first - the quad-core versions of Ryzen won't have any of this inter-CCX business, because they only have one CCX. So in that sense, R5 and R3 will be "fixed", to the extent that a "fix" is needed. And they'll be 4C/8T (or 4C/4T) with performance reliably exceeding the ever-popular Haswell i7s (or i5s) for half the price.

As for Ubuntu, there are preview editions of 17.04 available already. There's a clue in that version number - full release next month, right on their usual 6-monthly schedule.

i'm curious to see how the 6 core r5 will be affected by the inter CCX issue.
So, in practice 2 core should have the same L3 cache as the other 4, resulting in double the l3 cache per core. This should happen in the case of a 4+2 configuration. In a 3+3 scenario L3 cache is balanced whit 33% more L3 cache per core against the 8 core version but the cpu will suffer greatly of the inter ccx connection issue.
Which of those 2 scenarios is the more likely?

 

[Kromaatikse]

It's going to be 3+3. That's the only 6-core configuration for a downcored 8C die.

 

[innociv]

This might be believable, if linux hadn't fixed theirs in less than a week. But that fact that they did makes me think this should not take alot of man hours for MS to fix. Something else is going on here.

Linux didn't adjust their scheduler to account for the "2x4 core" design yet, dude.

It's expected to happen in the upcoming months, and programmers have said it's on their list, but it was not "fixed in less than a week".

 

[looncraz]

Linux didn't adjust their scheduler to account for the "2x4 core" design yet, dude.

It's expected to happen in the upcoming months, and programmers have said it's on their list, but it was not "fixed in less than a week".

Are you saying there are plans for even more Ryzen related improvements in Linux? Because last I saw, the CCXes were accounted for already...

 

[looncraz]

Okay, for those of you interested: Ryzen scaling with frequency is all over the place.

Some apps don't really scale at all, despite scaling linearly with Sandy Bridge (Cinebench king amongst these). Interestingly, the scaling issues are all related to multi-threaded apps. I've seen some weirdness in Windows 7 with multi-threaded scaling, so this may be a Windows 7 issue - changing to Windows 10 to find out.

AIDA64 skyrockets with higher clocks - showing ~25%+ higher scaling than anticipated... as in, they're showing that AIDA64 is useless for benchmarking Ryzen... and, yes, I'm using the latest beta.

One thing that does scale, in the real world, with frequency that is unexpected: memory performance. Usually changing the core frequency doesn't have much of an impact on memory reads and writes - maybe 500MB/s or so. I'm seeing 35GB/s changing to 43GB/s going from 3GHz to 3.8GHz - and Geekbench memory scores jumping from 3500 @ 3Ghz to 4000 at 3.8GHz.

All my results are with no SMT and only one CCX enabled.

Frankly, the results are bloody confusing.

EDIT: Results make sense on Windows 10 - positive scaling is still seen, but only in memory-sensitive apps. Further, multi-threaded scaling is as predicted. Don't use Ryzen + Windows 7 - the grass is not greener.

 

[unseenmorbidity]

cpu frequency increases memory performance?

 

[looncraz]

cpu frequency increases memory performance?

Yes. Reliably... and by large amounts.

Doesn't on my i7-2600k hardly at all with the same tests - which is what you'd expect.

I'm imaging my Windows 7 install now so I can return to Windows 10 and try the scaling tests all over again (so a few more hours of testing...).

 

[innociv]

Are you saying there are plans for even more Ryzen related improvements in Linux? Because last I saw, the CCXes were accounted for already...

I think you misread/misheard.

The only update for Linux, which happened right away, was to correctly assign SMT threads.

-znver1 is relying upon the btver1 scheduler model. Btver1 is for AMD's Bobcat.
People are finding using the Haswell scheduler model improves performance 5-10% in Linux, but they are working on a proper Zen scheduling model that should be more like 10-20% improvements in some cases.

And this is exactly what Windows needs, its own scheduler model for Ryzen...

 

[looncraz]

I think you misread/misheard.

The only update for Linux, which happened right away, was to correctly assign SMT threads.

-znver1 is relying upon the btver1 scheduler model. Btver1 is for AMD's Bobcat.
People are finding using the Haswell scheduler model improves performance 5-10% in Linux, but they are working on a proper Zen scheduling model that should be more like 10-20% improvements in some cases.

And this is exactly what Windows needs, its own scheduler model for Ryzen...

So, more than just this fix:

http://www.pcworld.com/article/3176...ves-linux-support-for-zen-multithreading.html

 

[innociv]

So, more than just this fix:

http://www.pcworld.com/article/3176...ves-linux-support-for-zen-multithreading.html

Yes... that "new" article is talking about a fix made over a week ago, and it details what it does in the article...

 

[looncraz]

Okay, so Windows 7 (or maybe just my install?) really robs Ryzen of multi-threaded performance. Windows 10 scaling is as expected.

My recommendation of not using Windows 7 with Ryzen is now solidified - you're just throwing away performance.

No game and no benchmark I tested performed better on Windows 7... while using AMD graphics.

If someone is willing to send me an nVidia graphics card that is fast enough to potentially be bottlenecked by a CPU in Battlefield 4 or Battlefield 1, then I am more than willing to do some testing. A GTX 980ti or above would be needed.

 

[OrangeKhrush]

Does Ryzen IPC look to you as being impaired by some bug/issue/erratum? It certainly doesn't look that way to me. The CPU's performance landed higher than anybody dared to hope 3 months back. Not just IPC, the clock too. In autumn, we thought we will thank God for every 100 MHz it will manage to do over 3 GHz in max boost state for single core. And we got 4.1 actually. That's actually overwhelming.

Those 2nd and 3rd revisions the mysterious insider talks about are merely the next architectural versions that will come out as regular roadmap updates with the usual yearly or longer cycles. No "magic fixed Ryzen revision". The current revision will keep selling for the year or so it was meant to. That is just how it is done, we might see 100MHz speedbump if we are lucky and the manufacturing improves.

Intel Kaby Lakes achieve higher max FPS in games - so what? That in no way suggest buggy chip. It's more or less the same with Broadwell-E. Kaby Lake manages higher memory bandwdith with fast DDR4, it has non-trivially higher IPC, and significantly higher clock to boot. Those are the things most current or past games respond to, so what else do you expect? Application software using multiple cores shows that Ryzen performance is real deal, the speed is there. I'd say at the moment there is no indication Ryzen performs bellow the plan in any area.

It looks right about on par, possibly just stability and power state issues at present.