Jump to content

dkabot

Members
  • Posts

    22
  • Joined

  • Last visited

Posts posted by dkabot

  1. You may want to avoid passing in SMT cores, but you should be fine to put in as many real cores as you want.

     

    QEMU (I think it was QEMU) is actually hardcoded to assume that AMD processors can't do SMT, and will pass through any SMT cores as if they were real ones.

    In my basic testing, while this does improve CPU performance a bit, you also lose some GPU performance for it.

  2. 2 minutes ago, jonp said:

    I was asked by another user to put this update in this thread as well, but it appears the issue relating to Nested Page Tables with AMD platforms has been resolved via a patch created by Paolo Bonzini:

     

    https://marc.info/?l=kvm&m=150891016802546&w=2

     

    Another user actually found the troublesome code segment, supplied a hack of a patch, then Paolo saw the real problem and implemented a proper solution.  That said, it hasn't made its way into a full release of the kernel yet, so we are going to patch it in manually with the release of the next rc.

     

    Needless to say, we are incredibly excited if this patch truly does resolve all these nasty GPU pass through performance problems.  Ryzen offers a strong use-case with unRAID and VMs with GPU pass through so long as it can keep on par with Intel for performance and price.  I'm excited to hear what you guys achieve after the next update.

     

    All the best,

     

    Jon

    We've been discussing this the entire page, though the update is appreciated regardless.

    Looking forward to trying the new build!

     

    I'll be sure to look into a license (been using trials every month or two to check the situation) once things are cooperating.

  3. 10 minutes ago, Southweave said:

    Can't they manually override the code or is it too much of a hassle? (I have never worked with kernels)

    Theoretically they should be able to apply the patch, but I'd be inclined toward thinking they won't.

     

    Edit: I don't have any reason to actually think so or not, just my gut.

  4. 2 hours ago, Southweave said:

    Edit #1: I just realized that it requires editing kernel? If so then I will just wait

    Yes, KVM is in the kernel and as such, this would require building a kernel yourself.

    Considering how old the only documentation for that (on Unraid) is, not something I'd recommend.

  5. Hopefully it's just a matter of time before something gets upstream and Unraid gets an update.

    I'm still tempted by the idea of being lazy and using Unraid for the management aspect, even if I can just install a normal distro.

     

    I suppose someone could try to compile a kernel on top of current Unraid, but considering the documentation, I don't think I'll be doing that.

  6. 6 hours ago, methanoid said:

    Hey dkabot where has your message gone?  Really good news and if a fix arrives will make lots of Ryzen and TR owners happy. I just plumped for two 5820k systems instead :(

     

    dkabot said:

    Saw this linked in the VFIO Discord (cool place, though not Unraid related).

    https://lists.linuxfoundation.org/pipermail/iommu/2017-October/024822.html

     

    Maybe a fix won't be never after all?

    I linked the wrong one, thought I had to remove and repost for some reason, and then forgot to.

    Sorry, that was me being dumb.

     

    There is, however, further good news in that a KVM maintainer has seen this and is trying to make a real patch.

     

    Edit: Sarnex has graced me with the link to this development: https://marc.info/?l=kvm&m=150891787704371&w=2

  7. 4 minutes ago, swiguy said:

    Hi all,

     

    I've been following the Ryzen based threads relatively closely, however I haven't seen much more active discussions based on the KVM NPT bug? I recently bought a Ryzen 7 1700 to replace my aging Xeon platform but I'm starting to regret it already! Video performance is awful, I haven't tried disabling NPT but it seems like a lose/lose workaround anyway. I'm mainly concerned about my Win 10 gaming VM... my other VMs are fine but this is my main VM with which I pass through a Geforce 1060.


    Anyway, I may need to stick with the Xeon platform for now for my unraid usage (NAS and dockers, some testing VMs etc.)... and go bare metal for the Ryzen platform if there isn't a resolution for the NPT bug on the horizon.


    Thanks!


    Chris

    You haven't seen anything because nothing's happened.

    It's still there, and the only "discovery" is that this bug is 10 years old and on other AMD processors.

     

    If you have a more technical hand, you can try other hypervisors which apparently are unaffected, such as Xen or ESXi.

    Xen seems to be very lenient on what you can pass through, but requires modifying nVidia drivers to install them and setup is obtuse.

    ESXi apparently works, and has a far easier way to just hide the VM status (like KVM), but I know of no way to work around my poor IOMMU groupings.

     

    There's also the possibility that some other hypervisor like Bhyve (BSD) will add GPU passthrough, in which case it could probably be used instead, but we have no estimate for if/when that'll happen.

     

    As much as I hate to say it, if you want things to Just Work(TM) get an Intel processor.

  8. On 10/11/2017 at 6:56 AM, ars92 said:

    Ah great at least it works with KVM now

     

    The wait for NPT fix by KVM continues.....

    Yeah, I'd be using Xen right now (which doesn't have the NPT bug at all, so I hear), but it refused to acknowledge that I had SVM on regardless of the fact that it definitely was.

    Running bare Windows is a tragedy, but I'm not the type to dual boot.

     

    Edit: My recollection was wrong. It knew I had SVM on, but due to some ACPI table thing, it was unable to enable AMD-Vi (IOMMU).

  9. On 9/26/2017 at 11:52 AM, dimes007 said:

    Is the change between NPT on or off transparent to the guest?  If I setup a new ryzen environment and get it working when and if a fix is released to allow NPT and GPU passthrough can I just change the settings or will Windows guests have issues and need to be rebuilt.  I have to decide if I'm going to keep my new Ryzen box or not.  ;(

     

    Thanks.

    It's been quite a while since I've touched this, but I'm fairly certain NPT can be toggled without changes to the VMs.

     

    NPT is a host hardware setting, not something you apply to virtual hardware on the guest.

    As such, while performance will differ, as far as the VM is concerned, nothing has changed.

  10. 50 minutes ago, Josecitox said:

    These are my IOMMU groups after updating a Gigabyte AX370-Gaming K7 to the latest beta bios with the AGESA 1.0.0.6 code.

     

    I had to do the ACS Override Enabled (downstream,multifunction) for them to appear like this, with nothing enabled they look even worse than the previous bios lol.

    My performance in gaming tanked though, don't know why i'm getting 50-60fps average in the Rise of the Tomb Raider benchmark when i had around 180-200 on bare metal.

     

    NPT is still busted as ever from what I can tell; it murders GPU performance, but turning it off hurts CPU performance a good deal.

  11. Alright, reverted my board back to latest stable.

     

    While the GPUs were detected in their own group, neither Unraid nor Arch were able to actually pass it through.

    Going into the VM externally with TeamViewer showed the resolution was 640x480 and there were no graphics adapters at all.

     

    Also, the RGB light cycle was messed up, and it would show red instead of purple on my CPU fan.

    Clearly the worst of tragedies.

     

    Anyway, looks promising, be nice to see what happens on stable release.

  12. 29 minutes ago, ufopinball said:

     

    That's disappointing.  With the ACS patch on unRAID, I am able to pass in one of three USB controllers built into the motherboard.  I ended up returning my USB 3.0 PCIe card because it wasn't necessary anymore.  It was less than $10, so I suppose it's no big deal either way.  Still, it would be nice to be able to plug the USB 3.0 card into a PCIe x1 slot, than having to use one of the larger x16 slots ... that's just how the IOMMU groupings broke up.

     

    - Bill

    With ACS on this board, one USB controller is exposed, which is for the two USB 3.1 (one of them Type-C) ports.

    The rest are intermingled with other system components, even with the patch.

  13. I tossed a beta BIOS on my Crosshair VI Hero, disabled the ACS patch, and here are my groupings:

    IOMMU group 0
    	[1022:1452] 00:01.0 Host bridge: Advanced Micro Devices, Inc. [AMD] Device 1452
    IOMMU group 1
    	[1022:1453] 00:01.3 PCI bridge: Advanced Micro Devices, Inc. [AMD] Device 1453
    IOMMU group 2
    	[1022:1452] 00:02.0 Host bridge: Advanced Micro Devices, Inc. [AMD] Device 1452
    IOMMU group 3
    	[1022:1452] 00:03.0 Host bridge: Advanced Micro Devices, Inc. [AMD] Device 1452
    IOMMU group 4
    	[1022:1453] 00:03.1 PCI bridge: Advanced Micro Devices, Inc. [AMD] Device 1453
    IOMMU group 5
    	[1022:1453] 00:03.2 PCI bridge: Advanced Micro Devices, Inc. [AMD] Device 1453
    IOMMU group 6
    	[1022:1452] 00:04.0 Host bridge: Advanced Micro Devices, Inc. [AMD] Device 1452
    IOMMU group 7
    	[1022:1452] 00:07.0 Host bridge: Advanced Micro Devices, Inc. [AMD] Device 1452
    	[1022:1454] 00:07.1 PCI bridge: Advanced Micro Devices, Inc. [AMD] Device 1454
    	[1022:145a] 2b:00.0 Non-Essential Instrumentation [1300]: Advanced Micro Devices, Inc. [AMD] Device 145a
    	[1022:1456] 2b:00.2 Encryption controller: Advanced Micro Devices, Inc. [AMD] Device 1456
    	[1022:145c] 2b:00.3 USB controller: Advanced Micro Devices, Inc. [AMD] Device 145c
    IOMMU group 8
    	[1022:1452] 00:08.0 Host bridge: Advanced Micro Devices, Inc. [AMD] Device 1452
    	[1022:1454] 00:08.1 PCI bridge: Advanced Micro Devices, Inc. [AMD] Device 1454
    	[1022:1455] 2c:00.0 Non-Essential Instrumentation [1300]: Advanced Micro Devices, Inc. [AMD] Device 1455
    	[1022:7901] 2c:00.2 SATA controller: Advanced Micro Devices, Inc. [AMD] FCH SATA Controller [AHCI mode] (rev 51)
    	[1022:1457] 2c:00.3 Audio device: Advanced Micro Devices, Inc. [AMD] Device 1457
    IOMMU group 9
    	[1022:790b] 00:14.0 SMBus: Advanced Micro Devices, Inc. [AMD] FCH SMBus Controller (rev 59)
    	[1022:790e] 00:14.3 ISA bridge: Advanced Micro Devices, Inc. [AMD] FCH LPC Bridge (rev 51)
    IOMMU group 10
    	[1022:1460] 00:18.0 Host bridge: Advanced Micro Devices, Inc. [AMD] Device 1460
    	[1022:1461] 00:18.1 Host bridge: Advanced Micro Devices, Inc. [AMD] Device 1461
    	[1022:1462] 00:18.2 Host bridge: Advanced Micro Devices, Inc. [AMD] Device 1462
    	[1022:1463] 00:18.3 Host bridge: Advanced Micro Devices, Inc. [AMD] Device 1463
    	[1022:1464] 00:18.4 Host bridge: Advanced Micro Devices, Inc. [AMD] Device 1464
    	[1022:1465] 00:18.5 Host bridge: Advanced Micro Devices, Inc. [AMD] Device 1465
    	[1022:1466] 00:18.6 Host bridge: Advanced Micro Devices, Inc. [AMD] Device 1466
    	[1022:1467] 00:18.7 Host bridge: Advanced Micro Devices, Inc. [AMD] Device 1467
    IOMMU group 11
    	[1022:43b9] 03:00.0 USB controller: Advanced Micro Devices, Inc. [AMD] Device 43b9 (rev 02)
    	[1022:43b5] 03:00.1 SATA controller: Advanced Micro Devices, Inc. [AMD] Device 43b5 (rev 02)
    	[1022:43b0] 03:00.2 PCI bridge: Advanced Micro Devices, Inc. [AMD] Device 43b0 (rev 02)
    	[1022:43b4] 1d:00.0 PCI bridge: Advanced Micro Devices, Inc. [AMD] Device 43b4 (rev 02)
    	[1022:43b4] 1d:02.0 PCI bridge: Advanced Micro Devices, Inc. [AMD] Device 43b4 (rev 02)
    	[1022:43b4] 1d:03.0 PCI bridge: Advanced Micro Devices, Inc. [AMD] Device 43b4 (rev 02)
    	[1022:43b4] 1d:04.0 PCI bridge: Advanced Micro Devices, Inc. [AMD] Device 43b4 (rev 02)
    	[1022:43b4] 1d:05.0 PCI bridge: Advanced Micro Devices, Inc. [AMD] Device 43b4 (rev 02)
    	[1022:43b4] 1d:06.0 PCI bridge: Advanced Micro Devices, Inc. [AMD] Device 43b4 (rev 02)
    	[1022:43b4] 1d:07.0 PCI bridge: Advanced Micro Devices, Inc. [AMD] Device 43b4 (rev 02)
    	[1b21:1343] 21:00.0 USB controller: ASMedia Technology Inc. Device 1343
    	[8086:1539] 23:00.0 Ethernet controller: Intel Corporation I211 Gigabit Network Connection (rev 03)
    IOMMU group 12
    	[10de:1380] 29:00.0 VGA compatible controller: NVIDIA Corporation GM107 [GeForce GTX 750 Ti] (rev a2)
    	[10de:0fbc] 29:00.1 Audio device: NVIDIA Corporation Device 0fbc (rev a1)
    IOMMU group 13
    	[10de:13c0] 2a:00.0 VGA compatible controller: NVIDIA Corporation GM204 [GeForce GTX 980] (rev a1)
    	[10de:0fbb] 2a:00.1 Audio device: NVIDIA Corporation GM204 High Definition Audio Controller (rev a1)

    Looks like GPUs can readily be passed through without hacks, but USB, onboard audio, SATA, etc. are still off the table as-is.

     

    EDIT: After trying on both an existing USB and a new one, I now cannot get Unraid to pass through a GPU for some reason. Ver. 6.3.4.

    Going to try a normal linux again now that I don't need to compile a kernel for ACS patch and see how far I can get.

  14. Additional observations about VM-related Freezes:

    • Debian Netinst (non-GUI) freezes when I hit Choose Language.
    • Antergos freezes once you choose your disk partitioning style, but the cursor can still move.
    • When the host freezes, the system is not technically hung; WebUI, SSH and (seemingly) the local console hang, but I can upload an ISO and it stays across the reset.

    Still no closer to any progress (I'd test normal linux but I can't get a compiled kernel to boot...), but the last one was something I thought was interesting.

     

    EDIT: Because I don't want to double post...

    I'm an idiot; I finally decided that since I needed the QEMU Emulated CPU rather than Host Passthrough to get Windows running, I'd try it on Linux.

    Tried Antergos, and the installer itself errored out, oh well.

    Tried Kubuntu, and it actually succeeded, though it failed to restart I could just Force Stop the VM at that point.

     

    Trying to install a 4.11 kernel on Kubuntu made it stop booting, so there's still a good deal to poke at, but at least I get further than everything crashing, it seems.

  15. 11 hours ago, GViz said:

    I was able to get mine up and running without any issues. I am using cpu cores 8-15, it should be the same since you're on the 1700 and i'm using a 1700x. 30GB virtual hard drive, 10 gb of ram (there is a bare minimum of at least 2), stored within the domains folder that unraid creates for VM's. Are you still having issues setting up a vm?

    No; while I did encounter issues with setting up Windows, they were alleviated fairly easily with the instructions I noted.

     

    My current issue is that as a guest, Kubuntu's installer will fairly consistently hang the system, Ubuntu's will occasionally do so and even Windows has managed to do it once.

    Trying to poke at other Linux host setups to see if they change the situation, and so far the only one I got far enough to work (kernel 4.10) hung the guest when I tried to load Kubuntu, with the host then noting that the process for KVM was hung every ~2 minutes.

     

    Because of that, I haven't even gotten into trying disabling NPT or any other workarounds for the GPU hit; just trying to sort out if there's anything to deal with hard locks.

  16. Something odd I've encountered but not seen in here based on some searches: Windows 8.1/10 refused to start the installer on the default VM settings.

    It'd get to the Windows logo flag, I'd get some HDD access for a little while, followed by absolutely nothing at all to indicate it was doing anything.

     

    I followed some instructions I saw elsewhere about doing an upgrade to 10 where they said to set CPU to "Emulated (QEMU64)" with one core, and that got it booting.

    The passed through GPU seems to be irrelevant to this occurring, as the failure and success cases were the same with VNC or a passed through GPU.

     

    It seems once it's been installed and downloaded a round of updates (probably unnecessary?) I was able to switch it back and have it not fail to boot.

    May be common knowledge, I'm not certain, but tossing it in this thread for completeness' sake since it doesn't appear to be mentioned.

  17. 10 hours ago, GViz said:

    So has anyone who has bought the Ryzen chip had any good luck in terms of playing games at around 60 fps? At the moment games just jump all over the place from a steady fps down to the 10-20 fps. Portal and Portal 2 work like a charm, but any game like dying light, witcher 3, borderlands, or anything with some extra umph seems to go all over the place in terms of performance. Running on 3d Mark had no issues. I'm at a bit of a loss on what to do.

    I've changed the minimum cpu frequency to match the max to try and reduce any latency that may be causing any issues, changed some performance metrics, etc. I'm not sure on where else to go from here. 

    I'm loving everything else about unraid, just this last hump i'm trying to get over and this system will be great. Any and all help will be appreciative!

    I do not have firsthand experience, but have read about it in a few places.

     

    For some reason, Ryzen/AMD appears to tank framerates unless you disable NPT (Nested Page Tables).

    Disabling it, though, degrades CPU performance; the extent of which seems to vary on who you ask.

     

    We've only got a few places it's referenced, but the story is still fairly similar all around.

    It's been posted on this thread, at reddit (twice), and on the VFIO mailing list with no response.

     

    Hopefully a less harmful workaround than disabling NPT is found; the second reddit post looks interesting.

×
×
  • Create New...