PDA

View Full Version : F26 and AMD Ryzen ... Works!



rclark
9th August 2017, 06:01 AM
Today, I upgraded my R&D system. New Motherboard, New CPU, and New Memory.
Ryzen 5 1600, GigaByte AB350-Gaming 3, and 32G of DD4 RAM. Booted right out of the gate. Can't ask for a nicer build. I had Ubuntu 17.04 on the SSD and the new system would not run it. Tried a fresh install of Ubuntu and got a irq 7 blah blah blah error over and over (other people got this too). So burned a F26 Workstation DVD and it ran just fine. Installed F26-LXDE on the SSD. Came right up after reboot. Fast boot. Then came the problem child... For some reason I couldn't get the Nvidia driver to work.... So reinstalled and got nouveau back. This time I used the latest driver from Nivida and followed instructions found on the web. Worked fine. Up and running. GPu benchmarks say 60fps for furmark and such for the 750 ti.... Good! This is one reason I stay away from Fedora as I have 'hours' of work to get every thing working again. By the time I am settled in, time to upgrade again ,,,, or so it seems! Chased down VLC and it works. Of course I needed LibreOffice and other software that isn't 'auto loaded' like Ubuntu. Even gcc/g++ isn't on the desktop spin.... But I am mostly there now tonight. Will tackle Virtual Box another day....

Just thougt I'd put this out there if someone else was thinking of going to the new AMD chip. F26 does work here.

ocratato
9th August 2017, 06:46 AM
Did you see this: http://techreport.com/news/32362/amd-confirms-linux-performance-marginality-problem-on-ryzen

srakitnican
9th August 2017, 06:49 AM
Just thougt I'd put this out there if someone else was thinking of going to the new AMD chip. F26 does work here.

Some users are having stability issues under heavy workloads with Ryzen currently. Until that is resolved I would delay shopping for a new CPU.

antikythera
9th August 2017, 08:59 AM
should be resolved by a microcode update at some point. the custom order Ryzen systems I've built so far have been destined for Windows only usage. Same goes for any brand new hardware from any manufacturer though, I won't recommend a Linux user the components until they are proven and these teething problems ironed out. It's simply not a route I'm willing to go down with my business.

rclark
9th August 2017, 03:54 PM
I would agree ... for a business stay away as the Ryzen chips have only been around for a few months. Even the new motherboards are having some teething problems. But, mine is a home computer that I use for my development projects, so I don't mind a glitch here and there when they pop up (none has so far, other than the Ubuntu issue which will work it's way out eventually). I don't plan on overclocking it anytime soon. Writing this on the new Ryzen system.

BTW, I did see the article above before buying the system. Just wanted a 'bleeding' edge system for once to play with :) . I think I still will get 'useful' work out of it! And any glitches will get fixed over time.


Oh, the system is nice and quiet. The CPU is a lot more energy efficient than the Phenom II X4 that it replaces. Fan just doesn't kick on. BIOS was showing 52C.

lsatenstein
9th August 2017, 04:03 PM
The noted problems with Ryzen have been with executing 9 concurrent GCC compiles. I have never done more than two concurrent ones. I had a response from AMD that it is not a mother board problem. That AMD is working diligently on a permanent fix.

The Ryzen problem is a teething one, just as Intel has/had a teething problem with some 'Lake systems. In the meantime, my target for my Ryzen upgrade is on schedule.
I question if I should go with a 1700 system or a 1600 systems, as you have done. I am in contemplation mode.

Mr Clark, I really thank you for your posting. Do you mind to produce an update after two two weeks of use?
.

rclark
9th August 2017, 07:30 PM
I will plan on giving an updates as this progresses.

I didn't go with the 1700 mainly because it costs ~80 bucks more. The performance from the 1600 with 6 cores and 12 threads is way better than my original phenom processor. I was pretty satisfied with the original, except for VM performance. I am hoping this system will help in that department. Down the road, I can always upgrade the processor if it becomes necessary.

rclark
11th August 2017, 04:11 AM
VirtualBox 5.1.26 is up and running. Much much faster than the old system. Seem close to 'native' speed now even with four VMs running :) . Did have to enable SVM and SMT in the board BIOS for it to work. For some reason they were disabled. So far so good. VMs are running WinXP, WIN7, F24 and F25. Using 20.3G of RAM out of the 32G.

antikythera
11th August 2017, 09:40 AM
good to know. yes it's odd that the hardware virtualisation functions are disabled by default but the same applied for my own mainboards from both ASUS and Gigabyte. I can't see how it would hit performance to have them enabled by default even if the user isn't running a VM.

what GPU did you go with when building this system? nevermind re-read your post, the old faithful nvidia :)

rclark
16th August 2017, 03:29 PM
Went through and update cycle/reboot no problem with new kernel (4.12.5-300.fc26.x86_64 #1 SMP) now on-line. No problems (no seg-faults, NVidia continues to work). Haven't shut down the computer since I loaded Fedora, just letting it burn in... so to speak.

VLC has a little problem. Oh, it plays everything ok. But sometimes when you go to load a new video, it locks up. You have to use the kill command to remove the application. I am sure it has nothing to do with Ryzen.
Firefox has crashed a couple of times now. The one time I was scrolling down while in a youtube page. Just locked and a message popped up saying that it had crashed.
08/18/17 Today at lunch, F24 VM found running at 100% on one processor. Had to kill the VM, as session was locked up.

pingwu
19th August 2017, 08:45 PM
Just thougt I'd put this out there if someone else was thinking of going to the new AMD chip. F26 does work here.

Thanks for the exciting report!

I am really looking forward to the "Raven Ridge" APU (Ryzen + Vega) chip. Because of the opensource amdgpu, all our future computers will be AMD based.

We began using AMD machine since near the end of last year. Our first AMD-based system, AMD FX 8350 + RX 480, works really great--thanks in part to the amdgpu.

rclark
20th August 2017, 10:11 PM
Well, bad news .... I am seg faulting all over the place now. I've updated to latest BIOS, etc. and problem exists. Not distro dependent. Seg faults can happen anywhere. Very easy to reproduce. From what I am reading this is a Ryzen CPU problem.... Seems to be getting worse now. Even locks up the computer. Anyway... Not to happy.

dd_wizard
20th August 2017, 10:22 PM
Have you tried turning off hyper-threading?

dd_wizard

lsatenstein
20th August 2017, 11:46 PM
Well, bad news .... I am seg faulting all over the place now. I've updated to latest BIOS, etc. and problem exists. Not distro dependent. Seg faults can happen anywhere. Very easy to reproduce. From what I am reading this is a Ryzen CPU problem.... Seems to be getting worse now. Even locks up the computer. Anyway... Not to happy.

Is it with Gnome, or another interface or with an application?

rclark
21st August 2017, 01:44 AM
I used LXDE with Fedora. You can see the 'fault's sometimes when the system starts to boot and hangs. If it does run, you see the fault with the applications you pull up. Say firefox, or a terminal, or whatever. The application may just disappear all of a sudden. Or in case of 'dnf', it will not download correctly, get checksum errors or say unable to install 'x' app... Check dmesg and you see the seg fault, core dump, etc... It is frustrating. Let it run long enough now and it hangs (keyboard/mouse freeze up). Have to hit the reset to get going again.

Yes I have disabled SVM and SMT. I think SMT disables the hyper threading. I disabled the speed boost as well, just running straight 3.2Ghz

I am running 'stock'. Not over clocking. The first time I flashed the BIOS, it didn't take. F6 was still being used on reboot (or F7 was corrupted and it reverted back to F6) . Tried again, and this time it 'took'. So running F7 now.

Look a lot like what these guys are seeing:

https://community.amd.com/thread/215773?start=960&tstart=0

lsatenstein
21st August 2017, 02:57 AM
As speed of execution increases, the timings for ram become more critical. Have you verified your ram characteristics against what is published for your Ryzen CPU? If there is a timing issue, segfaults will occur.

There was also a comment that the Ryzen cpu's in the 6 months they are out, have had engineering changes (new microcode).
You could get an RMA if your Ryzen CPU Chip is one of the earlier models.

---------- Post added at 09:57 PM ---------- Previous post was at 09:53 PM ----------

Is the GCC compiler for your lxde system 7.1?

rclark
21st August 2017, 06:16 AM
For the Fedora system yes 7.1.1.

Yep, RMA sounds like the next step. But sounds like a bunch of hoops to jump through.

The ram I got was the DDR4-2133 F4-2133C15D-32GFX (16GBx2). Basically the lowest DD4 speed. System recognizes it as such. According to packaging, the FlareX (G Skill) was 'designed' for the AMD Ryzen platform.

antikythera
21st August 2017, 07:01 AM
rma? hold that thought for now, there is a microcode coming via uefi firmware update. amd are releasing a firmware patch as soon as it is tested and ready to manufacturers to integrate into an uefi firmware update apply which you'd then be able to flash to your system

lsatenstein
21st August 2017, 02:01 PM
rma? hold that thought for now, there is a microcode coming via uefi firmware update. amd are releasing a firmware patch as soon as it is tested and ready to manufacturers to integrate into an uefi firmware update apply which you'd then be able to flash to your system

Hi Antikythera

Could you clarify one topic for me. The above microcode, from your post, updates the UEFI boot firmware.
What about post UEFI when Fedora gets rolling?

What happens if he tests with a Workstation Live ISO, without doing an installation?
Is it an lxde problem or a general problem?

antikythera
21st August 2017, 03:22 PM
anything that runs on the PC is also patched because it is done at system level rather than OS UEFI is not just a boot loader

rclark
21st August 2017, 03:23 PM
Yeah ... Holding ... Still playing. I pulled one of the ram sticks this morning, and we seem to run on the live-dvd without error (at the moment). Down to 16GB, one stick. Need to run, but will let system run on live DVD until I get back to it tonight.

lsatenstein
21st August 2017, 09:57 PM
So are you intelligent guessing that the problem is with the ram old the mother board?

rclark
21st August 2017, 10:18 PM
I am guessing... Because it is randomly seg faulting in a bunch of different applications, I was thinking memory is something I can try to change things up as I have two sticks, as the manual says the MB will work with just one stick. Not much more I can do hardware wise unless I order up another processor, MB... At this time I can tell you, the system is much more stable. Even reinstalled F26 to SDD to get rid of any corrupted files that may be out there. Started fresh. Install went without a hickup. The only seg fault right now is when I attempt to play a video in VLC. Always seg faults in same place in the nouveau driver. Firefox plays video just fine. Other apps run. dmesg shows no errors, except for when I try to run VLC. SMT bios setting doesn't seem to matter here on or off. I might now try to load the nvidia drivers... But don't have time now.

Is it really memory? Is it the processor? Is is the MB? How can you nail it down? Right now it 'looks' like memory. Yesterday I was convinced it had to be the processor as their Seg faults, looked a lot like mine.... <sigh>. Remember this system ran fine for several days ... before the degradation hit...

---

[edited] Tonight, I loaded daily LUbuntu, then OpenSuse TumbleWeed, and now back to Fedora. So far so good. Nary a glitch. Ran the memtest on the memory and it passed. Leaving on Fedora now as the saga continues. Need to load the Nvidia driver yet... But not tonight... No up to fighting that battle at the moment. Wish it was 'easier'.

Out of curiosity, what is the Fedora official 'extras' repository. I have been using rpmfusion....

---

This morning (8/22/17) all is well. No errors in dmesg. Seems stable. Not pushing the system, but then wasn't before either.

lsatenstein
22nd August 2017, 05:55 PM
I am guessing... Because it is randomly seg faulting in a bunch of different applications, I was thinking memory is something I can try to change things up as I have two sticks, as the manual says the MB will work with just one stick. Not much more I can do hardware wise unless I order up another processor, MB... At this time I can tell you, the system is much more stable. Even reinstalled F26 to SDD to get rid of any corrupted files that may be out there. Started fresh. Install went without a hickup. The only seg fault right now is when I attempt to play a video in VLC. Always seg faults in same place in the nouveau driver. Firefox plays video just fine. Other apps run. dmesg shows no errors, except for when I try to run VLC. SMT bios setting doesn't seem to matter here on or off. I might now try to load the nvidia drivers... But don't have time now.

Is it really memory? Is it the processor? Is is the MB? How can you nail it down? Right now it 'looks' like memory. Yesterday I was convinced it had to be the processor as their Seg faults, looked a lot like mine.... <sigh>. Remember this system ran fine for several days ... before the degradation hit...

---

[edited] Tonight, I loaded daily LUbuntu, then OpenSuse TumbleWeed, and now back to Fedora. So far so good. Nary a glitch. Ran the memtest on the memory and it passed. Leaving on Fedora now as the saga continues. Need to load the Nvidia driver yet... But not tonight... No up to fighting that battle at the moment. Wish it was 'easier'.

Out of curiosity, what is the Fedora official 'extras' repository. I have been using rpmfusion....

---

This morning (8/22/17) all is well. No errors in dmesg. Seems stable. Not pushing the system, but then wasn't before either.

Its RPMFUSION, but I take the Russian Fedora26 version, because it has some grub stuff that I do not see on rpmfustion. The Russian (hosted by Yandex, (http://mirror.yandex.ru/fedora/russianfedora/releases/RFRemix/26/) the competitor to Google) is where I find it. If you decide to take it, only take the DVD version unless you know Russian and Gnome. With the Desktop version you need to find Region and Language, and switch to English.

The Russian version has rpmfusion an things that tend to work out of the box. They have all the sources.

After installation, you could use dnf as before.

rclark
23rd August 2017, 03:02 AM
Thanks. I stuck with rpmfusion. Loaded the Nvidia driver for the 750Ti. No problem. Loaded VirtualBox no problem. Again checked dmesg and all looks well there. Everything appears clean at this point. Loaded a VM and ran/running fine. Interestingly all six cores are working from 1 to 8%, memory at 4.45G used. I do have SMT turned off at this point as I really don't need 12 threads.... Yet. Oh, also loaded VLC and it runs ok under the Nvidia driver. At the moment there are no known 'issues' I am chasing. Running kernel 4.12.5-300.fc26.x86_64 #1 SMP.

Overall I am happy with this setup again. Everything just runs smooth, fast, and quiet compared to my previous system. Now to let 'cook' for awhile and see how it runs over time. I will probably get some different memory to bump me back up to 32G, but for now ... I'll just use what I have that works... 16GB.

I am using htop to watch the system usage. I like this little utility.

I do wish there was a utility to watch the CPU temp, but the last time I loaded sensors, it didn't detect any temps and such.

antikythera
23rd August 2017, 08:04 AM
there is no official extras repository supported directly by fedora project, they are all third party/community maintained. rpmfusion is the most established but there is also now unitedrpms. you'd use one or other, not both because they would conflict otherwise.

copr is the nearest to an official extras repository but again, none of the contents of copr are supported directly by fedora project.

rclark
24th August 2017, 05:22 PM
Thanks for the information. Appreciate it.

System is running like a Swiss Watch right now. So unless I run into any more issues, this will be my last update (unless someone has a question that is).

Really like the system. With an Samsung SSD 250GB loaded with Fedora 26 and a 1TB HDD for the /home (so worth it having /home on separate drive for OS testing and all), 16GB of DDR4, Ryzen 5 1600 3.2Ghz, 6 core CPU, NVidia 750Ti for graphics, good sound, Gbit Ethernet to the home server ... what can I say! The speed of the system as a whole is just awesome. Enjoy 'snappy'! System is so quiet, you hardly know it is there. It's on my desk, next to keyboard / monitor. Even the 'blue' LEDs (set to whatever color you like in BIOS, including off) on the board give off a nice power glow leaking through the case cracks and holes... Not a gamer/miner, so won't be 'bogging' down any of the cores any time soon!

srakitnican
25th August 2017, 06:15 AM
Seems only early batches are affected.

http://www.phoronix.com/scan.php?page=article&item=new-ryzen-fixed&num=1

To monitor CPU temp you can use lm_sensors if there is support for your motherboard chips. There are various utilities that reads from it for example Freon, the Gnome extension.

antikythera
25th August 2017, 07:38 AM
gremlins the new batch. don't get them wet or feed after midnight

I think it is dangerous to state categorically a processor from week XX onward is fixed because it ran for a few hours without any problems and other than that is based on users all with different hardware combinations. unless this statement comes directly from AMD I'll take it with a pinch of salt

srakitnican
25th August 2017, 08:58 AM
That's why I said "seems", data so far points in that conclusion.

https://docs.google.com/spreadsheets/d/1pp6SKqvERxBKJupIVTp2_FMNRYmtxgP14ZekMReQVM4/edit#gid=0

antikythera
25th August 2017, 09:10 AM
I was referring to Phoronix, not your own post. this seems to happen a lot more these days with new hardware. Intel and AMD often have to patch microcode because of memory compatibility issues or other bugs found post release, so the whole issue is really nothing out of the ordinary. it is however a stupid trend they both need to address before it gets any worse.