AMD Zen/Ryzen thread

Message

Tony0945 · Post by **Tony0945** » Sat Mar 11, 2017 2:51 pm

wrc, I tried the patch on 4.10.1 and got

patch -p1 < zen.patch
patching file arch/x86/include/asm/module.h
Reversed (or previously applied) patch detected!

Except excavator and Zen are NOT in the menu.

I'll try manual patching.

wrc1944 · Post by **wrc1944** » Sat Mar 11, 2017 4:07 pm

Hmmm. Are you using the 4.10.1-gentoo-sources with the experimental USE flag enabled? IIRC, that would already have the 5010 gcc opts patch which is virtually identical to the graysky2 gcc patch, but there is some renaming in the newer graysky2 version. I PM'ed mpagano, the gentoo-sources patch/kernel maintainer and asked him to please update the gentoo-sources 5010 gcc opts patch.

Maybe there is a conflict if the old patch version is already applied. The git-sources 4.11-rc1 i did is just a plain vanilla kernel.

I'll try on a new 4.10.1 gentoo sources, and see if what you got also happens to me.

Are we sure we have the same patch? mine is 20.8kb (21,260) in size, and i downloaded the zip file from the button on the page I linked to.

EDIT OK- downloaded a new 4.10.1-gentoo-sources, but disabled the experimental USE flag which pulls in the older 5010 gentoo gcc opts version.

I then applied the graysky new zen enabled gcc opts version, and worked fine, as expected.

Code: Select all

gentoo-amd64 wrc # cd /usr/src/linux-4.10.1-gentoo
gentoo-amd64 linux-4.10.1-gentoo # patch -p1 < enable_additional_cpu_optimizations_for_gcc_v4.9+_kernel_v3.15+.patch
patching file arch/x86/include/asm/module.h
patching file arch/x86/Kconfig.cpu
patching file arch/x86/Makefile
patching file arch/x86/Makefile_32.cpu
gentoo-amd64 linux-4.10.1-gentoo #

When I have some more time later on, I'll try it with gentoo-sources experimental flag enabled, just to confirm that was causing the problem you had. I would have thought that just applying the new graysky2 over the gentoo 5010 wouldn't be a problem, but I could be wrong.

EDIT2: OK I tried applying it to a 4.9.11-gentoo which already was using experimental gcc opts, and it did the same thing you had, so maybe it is a conflict. I guess until mpagano updates the gcc opts 5010 patch for zen, we can either disable the experimental USE flag on gentoo sources and just use the newer graysky2 patch, or possibly edit the genttoo 5010 patch ourselves.

Code: Select all

gentoo-amd64 linux-4.9.11-gentoo # patch -p1 < enable_additional_cpu_optimizations_for_gcc_v4.9+_kernel_v3.15+.patch
patching file arch/x86/include/asm/module.h
Reversed (or previously applied) patch detected!  Assume -R? [n]

Website · Post by **Zucca** » Sat Mar 11, 2017 5:31 pm

NeddySeagoon wrote:That makes perfect sense.
Ryzen is like a NUMA system from the L3 cache up

Interesting. Does this have any advantage when compared to "nomal" CPU?

Tony0945 · Post by **Tony0945** » Mon Mar 13, 2017 12:54 pm

I just realized something and looked it up.

Phoronix's tests are using Ubuntu 16.10 which is based on the 4.8 kernel and Ubuntu 17.04 which is based on 4.9

AFAIK, neither kernel has Ryzen support as such (i.e. just generic AMD) but does have Intel features.

A truer test would be, for example, the B350 Tomahawk board that he used with Gentoo with the 4.10.2 kernel, compiled with GCC 6.3 and march=native, compared to the Z170 Tomahawk running the same kernel, all compiled with march=native with the same compiler.

Naib · Post by **Naib** » Mon Mar 13, 2017 1:37 pm

Tony0945 wrote:I just realized something and looked it up.

Phoronix's tests are using Ubuntu 16.10 which is based on the 4.8 kernel and Ubuntu 17.04 which is based on 4.9

AFAIK, neither kernel has Ryzen support as such (i.e. just generic AMD) but does have Intel features.

A truer test would be, for example, the B350 Tomahawk board that he used with Gentoo with the 4.10.2 kernel, compiled with GCC 6.3 and march=native, compared to the Z170 Tomahawk running the same kernel, all compiled with march=native with the same compiler.

the kernel is updated from ppa

Naib · Post by **Naib** » Mon Mar 13, 2017 3:13 pm

https://www.pcper.com/reviews/Processor ... ver-Bullet

What began as a simple internal discussion about the validity of claims that Windows 10 scheduling might be to blame for some of Ryzen's performance oddities, and that an update from Microsoft and AMD might magically save us all, has turned into a full day with many people chipping in to help put together a great story. The team at PC Perspective believes strongly that the Windows 10 scheduler is not improperly assigning workloads to Ryzen processors because of a lack of architecture knowledge on the structure of the CPU.

In fact, though we are waiting for official comments we can attribute from AMD on the matter, I have been told from high knowledge individuals inside the company that even AMD does not believe the Windows 10 scheduler has anything at all to do with the problems they are investigating on gaming performance.

In the process, we did find a new source of information in our latency testing tool that clearly shows differentiation between Intel's architecture and AMD's Zen architecture for core to core communications. In this way at least, the CCX design of 8-core Ryzen CPUs appears to more closely emulate a 2-socket system. With that, it is possible for Windows to logically split the CCX modules via the Non-Uniform Memory Access (NUMA), but that would force everything not specifically coded to span NUMA nodes (all games, some media encoders, etc) to use only half of Ryzen. How does this new information affect our expectation of something like Naples that will depend on Infinity Fabric even more directly for AMD's enterprise play?

There is still much to learn and more to investigate as we find the secrets that this new AMD architecture has in store for us. We welcome your discussion, comments, and questions below!

Naib · Post by **Naib** » Mon Mar 13, 2017 4:26 pm

Intel's optimised distro still beats Ubuntu on Ryzen https://www.phoronix.com/scan.php?page= ... -AMD-Ryzen

wrc1944 · Post by **wrc1944** » Mon Mar 13, 2017 9:54 pm

I just noticed this in make xconfig- haven't ever enabled it on AMD cpus up to FX-8320.

Will it possibly be needed with AM4/Ryzen systems? (as in "AMD Carrizo and later chipsets" mentioned in the kernel config help)

CONFIG_X86_AMD_PLATFORM_DEVICE:

Select to interpret AMD specific ACPI device to platform device
such as I2C, UART, GPIO found on AMD Carrizo and later chipsets.
I2C and UART depend on COMMON_CLK to set clock. GPIO driver is
implemented under PINCTRL subsystem.

Symbol: X86_AMD_PLATFORM_DEVICE [=n]
Type : boolean
Prompt: AMD ACPI2Platform devices support
Location:
-> Processor type and features
Defined at arch/x86/Kconfig:585
Depends on: ACPI [=y]
Selects: COMMON_CLK [=y] && PINCTRL [=n]

wrc1944 · Post by **wrc1944** » Tue Mar 14, 2017 12:39 pm

AMD says Windows 10 scheduler operates correctly with Ryzen

More technical discussion and comments on the Ryzen architecture and win7 and 10, but still leaves me unclear on how (or even if) all this relates to Linux, and what can (or should) be done to address any problems.

http://www.techspot.com/news/68501-amd- ... ryzen.html

ct85711 · Post by **ct85711** » Tue Mar 14, 2017 2:54 pm

Honestly, I am feeling even less sure with getting an Ryzen with the latest reports. While the article is trying to be reassuring, it leaves a lot left unanswered or even redirecting the blame to software developers instead. If anything, the article feels like it is from a paper pusher up high in AMD and didn't bother talking with the techs and engineers.

The main part I want to pick apart is this,

AMD has also taken the time to address reports suggesting simultaneous multi-threading (SMT) was having a negative performance impact in some games. The company says they expect SMT to provide a "neutral/positive benefit", and quote a wide range of games where this behavior is seen.

AMD goes on to suggest that some game optimizations for Ryzen may be possible.

This sounds to me that their SMT is working properly, but they haven't bothered explaining why people have been getting a NEGATIVE benefit with it turned on. The second paragraph is frankly a given item, it's just like saying there's clouds in the sky.

As far as the scheduler in Windows 10 being an issue, I am uncertain of where to go on this, so I have been sitting on the sidelines. However, if my understanding on this is correct (from old reports), that the Windows 10 scheduler may have been causing performance issues. Combined with the current report saying Windows 10 isn't at fault, than it sounds to me Ryzen should be getting even better performance than they are right now and it's a on fault AMD for the missed extra performance. Either way, I don't know if I am correct on following this or not, hence why I am not holding this.

Now that I think about it more about my earlier statement, then it sounds like AMD intentionally hobbled the cpu just so that the next one has an easy performance gain for little/no work involved.

Tony0945 · Post by **Tony0945** » Tue Mar 14, 2017 5:20 pm

ct85711 wrote:As far as the scheduler in Windows 10 being an issue, I am uncertain of where to go on this, so I have been sitting on the sidelines. However, if my understanding on this is correct (from old reports), that the Windows 10 scheduler may have been causing performance issues. Combined with the current report saying Windows 10 isn't at fault, than it sounds to me Ryzen should be getting even better performance than they are right now and it's a on fault AMD for the missed extra performance. Either way, I don't know if I am correct on following this or not, hence why I am not holding this.

Sounds like corporate finger pointing to me.

Now that I think about it more about my earlier statement, then it sounds like AMD intentionally hobbled the cpu just so that the next one has an easy performance gain for little/no work involved.

That would be very stupid. If Ryzen fails there will be no Zen+ there will be no AMD.
Of course, CEO's are arrogant and stupid. Look at Sears and JC Penney.

Naib · Post by **Naib** » Tue Mar 14, 2017 5:36 pm

I have read similar things... Everything is pointing towards the CCX & how aggressively windows moves processes... Ryzen is more like two quads.
Linux equally moves processes around quite liberally IF it is attempting to save energy (hence my criticism of onDemand as it is very reactive rather than applying say a kalman...)

wrc1944 · Post by **wrc1944** » Wed Mar 15, 2017 12:38 am

LLVM 5.0 Should Offer Better AMD Ryzen Performance

http://www.phoronix.com/scan.php?page=n ... eduler-WIP

LLVM 5.0 won't be released for about six months, but hopefully the scheduler will soon be introduced in mainline LLVM so it will be easier for interested individuals to try out and see how it impacts the AMD Ryzen performance. I'll certainly run some fresh AMD Ryzen compiler benchmarks when the code is updated -- for now see my current Ryzen GCC compiler tuning and GCC vs. Clang Ryzen benchmarks.

The Impact Of GCC Zen Compiler Tuning On AMD Ryzen Performance
Written by Michael Larabel in Software on 3 March 2017.
http://www.phoronix.com/scan.php?page=a ... ver1&num=1

From the comments section:

Originally posted by Michael View Post

This LLVM article isn't about the kernel scheduler, rather the compiler scheduler model.

True but the problem here is people not recognizing that Ryzen as very new architecture. I would expect both operaing systems and compilers to show improved support over the next year. It is the way of the inustry since basically forever. Eventually support matures and we dont think About it any more.

What perplexes me though is who gets wound up over poor gaming performance. Especially on a processor that finally livesup to is billing. It seems like the negativity of the political world is spilling over into the world of AMD processors. Really guys we should be happy that Ryzen is doing so well against Intels crap. We need AMD!

Post by Hu » Wed Mar 15, 2017 1:34 am

ct85711 wrote:The main part I want to pick apart is this,
AMD has also taken the time to address reports suggesting simultaneous multi-threading (SMT) was having a negative performance impact in some games. The company says they expect SMT to provide a "neutral/positive benefit", and quote a wide range of games where this behavior is seen.
This sounds to me that their SMT is working properly, but they haven't bothered explaining why people have been getting a NEGATIVE benefit with it turned on.

As I understand hyperthreading, that result was guaranteed to happen somewhere. Some workloads benefit from the increased parallelism of SMT. Some workloads suffer because (at least in all the processors I have read about, which does not include Ryzen) splitting a core's processor into two hyperthreads does not give the same level of functionality as having two full independent cores. If the workload does not rely on the resources that are necessarily cannibalized to create that second virtual core, you get the benefit of a second thread with little or no penalty. If the workload needs those resources, you lose more than you gain back from the parallelism. If people are seeing performance drops in some games, that is unfortunate, but not unreasonable and not necessarily indicative of a problem in the hardware.

truekaiser · Post by **truekaiser** » Wed Mar 15, 2017 2:39 am

NeddySeagoon wrote:That makes perfect sense.
Ryzen is like a NUMA system from the L3 cache up

So would setting the numa options in kernel help with performance?

wrc1944 · Post by **wrc1944** » Wed Mar 15, 2017 1:35 pm

Some new info:

AMD Ryzen 5 Mainstream Processors Confirmed For April 11th Launch – 6 Core Ryzen 5 1600X To Cost $249, 4 Core Ryzen 5 1500X To Cost $189

http://wccftech.com/amd-ryzen-5-1600x-1 ... 11-launch/

[Video] Ryzen From FX: Performance Gains Over Vishera

http://wccftech.com/ryzen-fx-performanc ... um=related

stephan-t · Post by **stephan-t** » Wed Mar 15, 2017 3:43 pm

wrc1944 wrote:Some new info:
AMD Ryzen 5 Mainstream Processors Confirmed For April 11th Launch – 6 Core Ryzen 5 1600X To Cost $249, 4 Core Ryzen 5 1500X To Cost $189
http://wccftech.com/amd-ryzen-5-1600x-1 ... 11-launch/

[Video] Ryzen From FX: Performance Gains Over Vishera
http://wccftech.com/ryzen-fx-performanc ... um=related

R5 1600X quite well. I will plan this upgrade.

steveL · Post by **steveL** » Wed Mar 15, 2017 5:06 pm

NeddySeagoon wrote:That makes perfect sense.
Ryzen is like a NUMA system from the L3 cache up

truekaiser wrote:So would setting the numa options in kernel help with performance?

Theoretically, but you need a (device-tree?) setting to tell it how the nodes are configured.

In this case each of the cores on a CCX would have a distance of 1 from each other, and a distance (say) of 10 to the nodes on the other core.
With SMT that might change a bit (1 to other SMT core; 2, 3 or whatever to same CCX; 10 to other CCX.)

The point being that it needs some sort of setup in kernel.
Sorry for not being more specific about how that's done, I'm just talking from memory.

Post by **NeddySeagoon** » Wed Mar 15, 2017 5:15 pm

stephan-t,

That's all the Ryzen 7 silicon that didn't make the grade for one reason or another.

tinea_pedis · Post by **tinea_pedis** » Sat Mar 18, 2017 5:52 pm

Not linux related, but

"Your PC uses a processor that isn’t supported on this version of Windows" error when you scan or download Windows updates

Symptoms

When you try to scan or download updates through Windows Update, you receive the following error message:
Unsupported Hardware
Your PC uses a processor that isn’t supported on this version of Windows and you won’t receive updates.
Additionally, you may see an error message on the Windows Update window that resembles the following:
Windows could not search for new updates
An error occurred while checking for new updates for your computer.
Error(s) found:
Code 80240037 Windows Update encountered an unknown error.
Cause

This error occurs because new processor generations require the latest Windows version for support. For example, Windows 10 is the only Windows version that is supported on the following processor generations:

Intel seventh (7th)-generation processors
AMD “Bristol Ridge”
Qualcomm “8996"

Because of how this support policy is implemented, Windows 8.1 and Windows 7 devices that have a seventh generation or a later generation processor may no longer be able to scan or download updates through Windows Update or Microsoft Update.

Resolution

We recommend that you upgrade Windows 8.1-based and Window 7-based computers to Windows 10 if those computers have a processor that is from any of the following generations:

Intel seventh (7th)-generation "Intel Core" processor or a later generation
AMD seventh (7th)-generation (“Bristol Ridge") processor or a later generation
Qualcomm “8996" processor or a later generation

https://support.microsoft.com/en-us/hel ... cessor-tha

That is a real dick move from MS, but of course nothing new: the old paywall game.

Also: The IT87* IO chips on AM4 boards do not work with the 'IT87* and combatibles' kernel thermal driver. ALC1220 (and ALC1150 too) codec do not work with oss-4.2.2011 (do not know about 2017, but at least it doesn't list any changes to hd_audio driver) either.

truekaiser · Post by **truekaiser** » Sat Mar 18, 2017 6:34 pm

Still a bit on topic.
I have everything for my ryzen build except for the coolmaster adapter for their liquid pro cpu cooler and the asrock taichi motherboard.
I am wondering, can I build the linux install on the nvme pciex4 root drive from my current system running a fx8350 to the point that all i have to do is stick it in an boot it? Not to the point of having a full desktop, just to the reboot point of the handbook install?
Basically can -march=native on the host machine like this via a chroot run on a ryzen cpu enough that i can boot it then recompile everything for -march=znver1?

I'm so far on day 15 of the motherboard back order and no eta on the adapter kit. Not going to spend another 100 on another closed loop water cooler..

wrc1944 · Post by **wrc1944** » Sun Mar 19, 2017 1:03 am

truekaiser,
From my long-time amd systems experiences, plus a few things I've recently read on Ryzen, I'm virtually certain that any bulldozer and up to excavator am3+ system should at the very least boot right up on any of the new AM4/Ryzen systems hardware.

I also read somewhere that even even amd k8 systems would probably work. So, I'd think that what you're proposing would work, and I'd go even further to say that you might even consider just cloning over a full system partition image of your current fx8350 to the new nvme pciex4 root drive, and try booting into the new hardware with that. If you had boot and system on the same partition, it would seem to be pretty easy. However, I've never worked with SSD drives before, so maybe I'm missing something.

truekaiser · Post by **truekaiser** » Sun Mar 19, 2017 2:01 am

wrc1944 wrote:truekaiser,
From my long-time amd systems experiences, plus a few things I've recently read on Ryzen, I'm virtually certain that any bulldozer and up to excavator am3+ system should at the very least boot right up on any of the new AM4/Ryzen systems hardware.

I also read somewhere that even even amd k8 systems would probably work. So, I'd think that what you're proposing would work, and I'd go even further to say that you might even consider just cloning over a full system partition image of your current fx8350 to the new nvme pciex4 root drive, and try booting into the new hardware with that. If you had boot and system on the same partition, it would seem to be pretty easy. However, I've never worked with SSD drives before, so maybe I'm missing something.

I plan on keeping some settings and configs. But I am mostly aiming for a clean install as the fx8350 system was used for stuff i no longer do. Geared to vm's and emulation.

tinea_pedis · Post by **tinea_pedis** » Sun Mar 19, 2017 6:22 pm

3dnow instruction set is the only one lacking from Zen compared to Piledriver, but hardly any software can be compiled with it or will actually use those instructions even if compiled with it.

Personal experience: threw disk from my fx-8350 setup to 1700x box (which someone asked me to put together) and no kernel panic, every software (everything compiled with march=native - i.e. bdver2 - on the fx-8350) worked without a hitch, except for unsupported io chip, cpu thermal driver and alc1220 not working with oss4.
Going backward and using that disk with Thuban (1090t) setup or Intel Sandy Bridge (2500k) on the other hand, it's a kernel panic and even grub hanging if it was compiled with march=native, since those two archs lack some sse or avx instruction's compared to Piledriver (fx-8350).

wrc1944 · Post by **wrc1944** » Sun Mar 19, 2017 9:38 pm

tinea_pedis,
Thanks for confirming my conclusion- seems like any fam15h system should boot on the fam17h hardware, and then can be recompiled with some kernel and make.conf fam17h tweaks with an

Code: Select all

emerge -e @system --keep-going

and then if you really want to make sure all is OK, an

Code: Select all

emerge -e @world

Or, maybe just do an

Code: Select all

emerge -e @world --keep-going

if you feel lucky....

For reference, here's the fam15h fx-8320 /proc/cpuinfo flags:

flags : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ht syscall nx mmxext fxsr_opt pdpe1gb rdtscp lm constant_tsc rep_good nopl nonstop_tsc extd_apicid aperfmperf pni pclmulqdq monitor ssse3 fma cx16 sse4_1 sse4_2 popcnt aes xsave avx f16c lahf_lm cmp_legacy svm extapic cr8_legacy abm sse4a misalignsse 3dnowprefetch osvw ibs xop skinit wdt lwp fma4 tce nodeid_msr tbm topoext perfctr_core perfctr_nb cpb hw_pstate vmmcall bmi1 arat npt lbrv svm_lock nrip_save tsc_scale vmcb_clean flushbyasid decodeassists pausefilter pfthreshold