Announcement

Collapse
No announcement yet.

QEMU 1.5 Supports VGA Passthrough, Better USB 3.0

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

  • #16
    Originally posted by GreatEmerald View Post
    Got that covered. It's called VT-d on Intel and AMD-Vi on AMD. And it's not really that uncommon on current hardware.
    Finding a motherboard that really supports it is a pain.
    I have 4 motherboards here all with IOMMU mode and it doesn't work on any of them.

    Comment


    • #17
      Originally posted by GreatEmerald View Post
      Got that covered. It's called VT-d on Intel and AMD-Vi on AMD. And it's not really that uncommon on current hardware.
      Oh, I think it's quite uncommon. Especially on Intel processors. You need a Xeon processor and the corresponding mainboard. For AMD also not all chipsets support it. Which is rather strange, because I would have though it's enough for the processor to support it, since the memory controller is now part of the processor. In any case, it's often hard to find out whether particular mainboard supports it. I often have to download the manual to take a look at the BIOS options.

      It's similar to the situation with ECC. For intel it's mostly Xeon processors that support it. Strangely enough, there are some Pentium/Core-i3 processors that support it too but need the appropriate mainboard (C200 series chipset) which quite a bit more expensive. AMD AM3+ processors support ECC but there are some motherboards that don't. I imagine they just lack the traces to the additional pins for the ECC DIMM. But finding trustworthy information is quite hard. Apparently, most people dont give a shit about IOMMU/ECC. They need to know what they are first...

      Comment


      • #18
        Originally posted by schmidtbag View Post
        Why is this useless? Most x86 based computers today have an IGP and a discrete GPU, yet one of them is almost always inactive. Give the better one to your VM and problem solved.
        Most of the ultrabooks I've looked at just have Ivy Bridge, and you pay a premium in both normal laptops and ultrabooks to get a discrete card alongside it. Of course, if your purpose is to run Linux with Windows on QEMU intermittently for graphics-intensive proprietary apps, you could plan to buy one of these systems specifically for that purpose. And of course it's useful in that case- no need to dual-boot most likely, unless you're a stickler for getting every ounce of performance.

        I'm just waiting to see whether this can apply more broadly or split a single GPU's resources. My guess is that it's either non-trivial or it's basically impossible without the right graphics chip architecture.

        Comment


        • #19
          Originally posted by kobblestown View Post
          Oh, I think it's quite uncommon. Especially on Intel processors. You need a Xeon processor and the corresponding mainboard. For AMD also not all chipsets support it. Which is rather strange, because I would have though it's enough for the processor to support it, since the memory controller is now part of the processor. In any case, it's often hard to find out whether particular mainboard supports it. I often have to download the manual to take a look at the BIOS options.
          What are you talking about? I have an i5-3470, and it supports VT-d. It's fairly high-end, but by far not uncommon, and definitely not a Xeon. Other processors that support it are i5-3550, i5-3330, i5-2500... Heck, even a i5-680 supports it. It's not uncommon in the slightest.

          I also have a MSI B75A-G43 motherboard, and guess what, it supports VT-d as well. Even some Core 2 motherboards support it. So it's again not uncommon, although it's true that it's usually not advertised that much.
          Last edited by GreatEmerald; 05-21-2013, 12:32 PM.

          Comment


          • #20
            @sonicspectre
            Well the only way to divide GPU power between a VM is through the old method of creating a virtual GPU, which we all know is limiting and doesn't perform all that great.


            Originally posted by GreatEmerald View Post
            What are you talking about? I have an i5-3470, and it supports VT-d. It's fairly high-end, but by far not uncommon, and definitely not a Xeon. Other processors that support it are i5-3550, i5-3330, i5-2500... Heck, even in i5-680 supports it. It's not uncommon in the slightest.

            I also have a MSI B75A-G43 motherboard, and guess what, it supports VT-d as well. And even some Core 2 motherboards support it. So it's again not uncommon.
            VT-d is extremely common. The version of VT-d that supports IOMMU is not, and IOMMU is required for GPU passthrough:

            http://en.wikipedia.org/wiki/List_of...rting_hardware

            This list is relatively small. Albeit, the motherboard sections are a bit incomplete. My motherboard is one of the only AMD desktop boards with IOMMU support and it isn't on this list (890FX chipset).
            Last edited by schmidtbag; 05-21-2013, 12:37 PM.

            Comment


            • #21
              Originally posted by schmidtbag View Post
              VT-d is extremely common. The version of VT-d that supports IOMMU is not, and IOMMU is required for GPU passthrough:

              http://en.wikipedia.org/wiki/List_of...rting_hardware

              This list is relatively small. Albeit, the motherboard sections are a bit incomplete. My motherboard is one of the only AMD desktop boards with IOMMU support and it isn't on this list (890FX chipset).
              And look at all of those non-Xeon Intel CPUs (which is what I think the person you quoted was really refuting).

              This is pretty exciting. Xen will probably have something similar soon. I couldn't ever get my setup working with Xen (Intel 3770 non-K) but with the amount of work involved - hammering away at the kernel parameters, rebooting, trying again, discovering yet another configuration file to edit, reboot, try again, discovering that grub2 didn't properly build the new grub.cfg, etc - something much more simple is always welcome. I might have to switch over to QEMU from Xen just to give it a shot.

              Comment


              • #22
                Originally posted by schmidtbag View Post
                VT-d is extremely common. The version of VT-d that supports IOMMU is not, and IOMMU is required for GPU passthrough:

                http://en.wikipedia.org/wiki/List_of...rting_hardware

                This list is relatively small. Albeit, the motherboard sections are a bit incomplete. My motherboard is one of the only AMD desktop boards with IOMMU support and it isn't on this list (890FX chipset).
                Eh? I've never seen anyone making a distinction like that. The IOMMU page says that VT-d is a specification of IOMMU, so how can something support VT-d and not support IOMMU? And yes, the list is clearly incomplete.

                Comment


                • #23
                  Originally posted by peppercats View Post
                  Finding a motherboard that really supports it is a pain.
                  I have 4 motherboards here all with IOMMU mode and it doesn't work on any of them.
                  Same here, I have gone through 3 motherboards with iommu support but it doesnt actually work. A bios update for my asus sabertooth 990fx finally got it working.

                  Originally posted by schmidtbag View Post
                  @sonicspectre
                  Well the only way to divide GPU power between a VM is through the old method of creating a virtual GPU, which we all know is limiting and doesn't perform all that great.




                  VT-d is extremely common. The version of VT-d that supports IOMMU is not, and IOMMU is required for GPU passthrough:

                  http://en.wikipedia.org/wiki/List_of...rting_hardware

                  This list is relatively small. Albeit, the motherboard sections are a bit incomplete. My motherboard is one of the only AMD desktop boards with IOMMU support and it isn't on this list (890FX chipset).
                  One of the only ones? Ive never seen a high-end amd board recently that didnt support it. All of the 990fx series do that Ive seen, the problem is getting it to work properly.

                  Comment


                  • #24
                    Originally posted by GreatEmerald View Post
                    Eh? I've never seen anyone making a distinction like that. The IOMMU page says that VT-d is a specification of IOMMU, so how can something support VT-d and not support IOMMU? And yes, the list is clearly incomplete.
                    Rather than blindly disagreeing with me thinking that all VT-d based systems support IOMMU, I'd really like to see you try getting just plain PCI passthrough working on a Core 2 system on any VM. IOMMU is NOT just a CPU feature.
                    Last edited by schmidtbag; 05-21-2013, 02:02 PM.

                    Comment


                    • #25
                      Can someone post a walk-though or a guide on how to get setup KVM with VGA Passthrough? I've got a 2nd video card and would finally like to get rid of Wine

                      Comment


                      • #26
                        Originally posted by gururise View Post
                        Can someone post a walk-though or a guide on how to get setup KVM with VGA Passthrough? I've got a 2nd video card and would finally like to get rid of Wine
                        https://docs.google.com/document/d/1...K8Y/edit?pli=1

                        Comment


                        • #27
                          Originally posted by schmidtbag View Post
                          Rather than blindly disagreeing with me thinking that all VT-d based systems support IOMMU, I'd really like to see you try getting just plain PCI passthrough working on a Core 2 system on any VM. IOMMU is NOT just a CPU feature.
                          Blindly? Again, you didn't provide any reason for that to be the case. And I don't have such a system (the one Core 2 I do have doesn't have VT-d to begin with). What philip550c says makes sense, though, in that some motherboards have buggy firmware that doesn't actually enable the feature. But it doesn't mean that there are different versions of VT-d, some of which are not IOMMU.

                          Comment


                          • #28
                            Originally posted by GreatEmerald View Post
                            What are you talking about? I have an i5-3470, and it supports VT-d. It's fairly high-end, but by far not uncommon, and definitely not a Xeon. Other processors that support it are i5-3550, i5-3330, i5-2500... Heck, even a i5-680 supports it. It's not uncommon in the slightest.

                            I also have a MSI B75A-G43 motherboard, and guess what, it supports VT-d as well. Even some Core 2 motherboards support it. So it's again not uncommon, although it's true that it's usually not advertised that much.
                            I have to concede here. On closer inspection VT-d is more common that I though. But the later posts show that it's not so simple. I still maintain that it's very hard to pick a CPU-MB combo and be reasonably sure in advance that it will work. And I wouldn't trust it on a non-Xeon setup. Even on Xeon I wonder if there's any difference between the E3, E5 and E7 lines WRT VT-d feature set.

                            Comment


                            • #29
                              Originally posted by GreatEmerald View Post
                              Blindly? Again, you didn't provide any reason for that to be the case. And I don't have such a system (the one Core 2 I do have doesn't have VT-d to begin with). What philip550c says makes sense, though, in that some motherboards have buggy firmware that doesn't actually enable the feature. But it doesn't mean that there are different versions of VT-d, some of which are not IOMMU.
                              schmidtbag IS correct.

                              Comment


                              • #30
                                Originally posted by soupbowl View Post
                                schmidtbag IS correct.
                                By all means, give me an explanation. Or a source of the claim.

                                Comment

                                Working...
                                X