Announcement

Collapse
No announcement yet.

AMD Working On VirtIO GPU & Passthrough GPU Support For Xen Virtualization

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • #31
    Originally posted by Danny3 View Post
    Why should I pay a thousand bucks for a GPU that can't support SR-IOV and CEC commands?
    just dont spend the money then lol, go buy used or another company, you have two choices now

    Comment


    • #32
      Originally posted by Danny3 View Post
      Isn't the same excuse used to tell us why we don't have CEC commands support also?
      Honestly, I don't buy it as I think the separation can be done with the firmware / driver's software.
      As for the price, they already increased to prices to unbelievable heights.
      Why should I pay a thousand bucks for a GPU that can't support SR-IOV and CEC commands?
      They are asking for those huge prices without anything of value being added.
      CEC one is annoying.

      DisplayPort CEC-Tunneling-over-AUX on i915, nouveau and amdgpu
      Yes if you plug your HDMI monitor in by HDMI to displayport adapter CEC commands can in fact work.

      The HDMI standard body as CEC as optional feature that any one putting HDMI port on something has to pay extra for do you fell like paying extra 50 to 300 dollars for the card.

      Lot of ways just give us consumer cards with all displayport and those who want HDMI can buy a dongle and make this be able to work well by extending the displayport protocol to cover all the HDMI features.

      Danny3 with SR-IOV are you willing to pay 200 dollars more for the same performance. Yes the professional cards are about 200 dollars more for the same performance as a consumer card and yes the consumer card has a smaller bit of silicon in majority of that 200 dollars is the SR-IOV price..

      Intel gpu some of the performance problem is SR-IOV.

      HDMI CEC and SR-IOV have different reasons. HDMI CEC is not on AMD cards HDMI ports because the standard body is of HDMI is a jackass. HDMI CEC is on the dispalyports because the displayport standard body is not a jackass.

      SR-IOV is hardware cost problem. Nvidia also keeps SR-IOV off consumer cards because because it has quite a silicon area cost and SR-IOV always comes with a performance cost.

      We have need a different solution to SR-IOV for consumer/cost effective card market.

      Comment


      • #33
        Speaking of AMD GPUs and passthrough, not only does AMD not offer SR-IOV even on many of the pro cards such as the Radeon Pro W5700, full passthrough of a single card often doesn't work properly either.

        They often fail to PCI reset properly, forcing you to reboot the host if you want to reboot the guest. This was true back when I tried it using an R9 290 (Hawaii), and it wouldn't surprise me if it's *still* true for many of their GPUs today.

        Comment


        • #34
          Originally posted by DanaG View Post
          Speaking of AMD GPUs and passthrough, not only does AMD not offer SR-IOV even on many of the pro cards such as the Radeon Pro W5700, full passthrough of a single card often doesn't work properly either.

          They often fail to PCI reset properly, forcing you to reboot the host if you want to reboot the guest. This was true back when I tried it using an R9 290 (Hawaii), and it wouldn't surprise me if it's *still* true for many of their GPUs today.
          When is important.
          Linux kernel vendor specific hardware reset module for sequences that are too complex/complicated to land in pci_quirks.c - gnif/vendor-reset

          Can confirm that you’re not the only one with this issue. I have an Asus Dual RX 6700XT (Navi 22) with the exact same problem on an Asus Prime X570-Pro. The card refuses to go into D3Hot state again after shutdown. Turning the VM on again after VM shutdown will cause a black screen (code 43 behind the scenes) due to the Restoring BARs problem. At any third attempt to start the VM it’ll go into the classic Unknown PCI header type ‘127’ for device ‘0000:0b:00.0 problem and be completely locked unt...


          Reboot host has not been the only option for quite some time. Vendor-reset and basic scripts when that problem turns up now will force card back into clean state. So things have improved somewhat.


          Comment


          • #35
            Originally posted by oiaohm View Post
            When is important.
            Linux kernel vendor specific hardware reset module for sequences that are too complex/complicated to land in pci_quirks.c - gnif/vendor-reset

            Can confirm that you’re not the only one with this issue. I have an Asus Dual RX 6700XT (Navi 22) with the exact same problem on an Asus Prime X570-Pro. The card refuses to go into D3Hot state again after shutdown. Turning the VM on again after VM shutdown will cause a black screen (code 43 behind the scenes) due to the Restoring BARs problem. At any third attempt to start the VM it’ll go into the classic Unknown PCI header type ‘127’ for device ‘0000:0b:00.0 problem and be completely locked unt...


            Reboot host has not been the only option for quite some time. Vendor-reset and basic scripts when that problem turns up now will force card back into clean state. So things have improved somewhat.
            vendor-reset is spotty at best

            Comment


            • #36
              Originally posted by Quackdoc View Post
              vendor-reset is spotty at best
              Lot of case this is not exactly spotty. There are times were it need to trigger more than once to get the card to reset. Interesting point lot of AMD card needing vendor-reset triggered more than once to reset card equals quite a few cases if you reboot the host you system you risk end up with no output as well vendor firmware fun. Fun right reboot can just dig yourself into a deeper hole. As I said improved somewhat not fixed.

              6000 series another set of vendor firmware issues on AMD. Yes fun where you have 2 AMD cards from the same vendor same model different firmware and one has issues with vm and restarts and the other is fine.

              ODM and AMD need to get these firmware issues sorted out. Its not just GPUs.

              Comment

              Working...
              X