Announcement

Collapse
No announcement yet.

First Batch Of AMDGPU Changes For Linux 4.16: DC Multi-Display Sync, Vega Tuning

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

  • #21
    Originally posted by ptyerman View Post
    @polarathene.
    Sorry, my post was concerning the selling of gimped cards still marketed as full RX 560's. That's false advertising at the very least, and a below the belt move.
    I know I wouldn't be very happy if I purchased a card and then found out it wasn't what I'd paid for but something a full tier below it!
    That's why I'm going to hold off for a while until things become more clear.
    That's not a full tier below it. Its a GPU with 1024 processors and someone might sell you one with 860-ish. That's around 9-10% performance difference in tests on the 460 (where there are bios mods that enable all 1024 processors on a card that is marketed as only 860-ish).

    Comment


    • #22
      Originally posted by polarathene View Post
      We can only wish, but I think it's like NVIDIA's equivalent where it's one of those features that enterprise tend to need rather than consumer so it helps being an exclusive feature to the pro cards?

      I think Bridgman has mentioned in the past there might have been some interest at AMD to offer SR-IOV on a consumer model(high-end) that didn't creep on the enterprise offerings too much, something like only partitioning resources into 2-4 sections I think? Even two would be pretty neat.
      I hope Intel's GVT will give enough pressure to AMD and maybe Nvidia too to provide these features on consumer cards. I know this is really just about market segmentation and not so much about special sauce on certain chips, maybe they even disable/enable it by using fuses or/and any other way (firmware?).

      Many users like me could barely afford a somewhat high end laptop, but it would be very positive to be able to have "vGPU" features when dealing with Virtual Machines and mixed environments (or needing to use software in different platforms due to the work environment or testing or whatever). AMD, please consider it.

      bridgman Could you give us information about Open Source GPU virtualization for consumer hardware?
      Last edited by timofonic; 07 December 2017, 11:33 AM.

      Comment


      • #23
        Originally posted by timofonic View Post

        I hope Intel's GVT will give enough pressure to AMD and maybe Nvidia too to provide these features on consumer cards. I know this is really just about market segmentation and not so much about special sauce on certain chips, maybe they even disable/enable it by using fuses or/and any other way (firmware?).
        FWIW, GVT is a pure software implementation; the drivers in the VMs just proxy to the driver on the physical GPU. If you want something like that, you can use a paravirtualized solution like virgil.

        Comment


        • #24
          -56722 lines of code. Niiiice. I love changes that reduce the codebase.

          Less code- less bugs, and easier to change and improve.

          Comment


          • #25
            Originally posted by chithanh View Post
            AMD admitted that it is a problem that customers are confused about which version of 560 they are buying
            Buyers are not "confused", they are deliberately deceived. There is no way in hell that these three things all happened "coincidentally" and "unintentionally":
            (1) AMD changing the specs on their pre-existing web-page on RX560 to suddenly allow for less compute units
            (2) Asus marketing an "RX560" with < 1024 compute units
            (3) PowerColor marketing an "RX560" with < 1024 compute units
            One of the above happening - ok, might be some odd mistake. All of the above happening: A deliberate attempt to lure buyers to an inferiour product.

            I don't know where you live but here in the EU any consumer can demand compensation or his money back if the product which they bought does not meet the advertised specs.
            Sure, if you notice that you have been deceived. AMD and their accomplices now probably bet on "very few people following the IT press will notice they were deceived, so let's not proactively inform any buyer, and keep the swag!".

            Comment


            • #26
              Originally posted by agd5f View Post

              FWIW, GVT is a pure software implementation; the drivers in the VMs just proxy to the driver on the physical GPU. If you want something like that, you can use a paravirtualized solution like virgil.
              I thought GVT to be far more sophisticated. I didn't look at the code, but their techie marketing stuff. The lack of a proper implementation without using remote desktop software worries me too.

              Virgil and such are a joke, it doesn't solve the need to depend on API reimplementations (DirectX, OpenGL, Vulkan, Metal, whatever...). I consider the key of having proper virtualization is able to get near-native performance on 2D+3D+video+whatever GPU stuff and not needing to deal with the proprietary operating system graphics API but run the stuff transparently in the GPU itself.

              I thought GVT was real GPU virtualization, able to use a virtual GPU with very low overhead compared to native usage. It's disappointing if what you say it's true, it probably is, but it's hard for me to accept it.



              Originally posted by Jike Song
              One clarification for rest discussion, is that we're talking about GVT-g vGPU (here which) is a pure software GPU virtualization technique. GVT-d (note some use in the text) refers to passing through the whole GPU or a specific VF. GVT-d already falls into existing VFIO APIs nicely (though some on-going effort to remove Intel specific platform stickness from gfx driver). :-)
              Some extra info related to Xen (Nvidia appears there too, yes...): http://xapi-project.github.io/featur...evolution.html
              Last edited by timofonic; 07 December 2017, 02:51 PM.

              Comment


              • #27
                Originally posted by Brophen View Post
                Maybe then we can get some consumer VEGA cards with sr-iov? Please AMD? Bridgman help me out here lol
                Signed! Come on AMD! You don't play a big role in enterprise! Hurt Nvidia by offering it on your GPUs And then invent/implement a fast streaming solution for zero/thin clients to hurt VMware

                Comment


                • #28
                  Jumping into the RX560 vs RX560D controversy here with some info I found. I wasn't even aware of this issue until reading this comment thread. From doing some online sleuthing on the subject it looks like AMD pulled a classic Nvidia and simply took the RX 460 ( low end card...had one installed in my HP Bristol Ridge APU desktop ) and rebadged it as the RX 560. Looking at the head to head comparison on GPU Boss the RX 460 and the original RX 560 ( now called RX560 D ) are absolutely identical in every respect except for....

                  TDP of 75 W for the RX460 and TDP of 65 W for the RX560 D
                  2 GB of GDDR5 for the 460 and 4 GB of GDDR5 for the 560 D

                  So basically just slight tweak upgrade to power draw ( less power for the 560 D ) and RAM.

                  They even admit that the same chip family is used for both the 460 and the 560 D ( AMD Baffin )

                  The Baffin family is part of the 400 series of AMD gpus ( Polaris ) with Baffin being the replacement of the Curacao gpu which powered the RX 270 back in the day.

                  So...it looks like the NOW badged RX 560 vs the RX 560 D is like Nvidia doing a 1050 and a 1050 Ti. with the RX560 D being the 1050 equivalent and the NOW badged RX560 being the equivalent to a 1050 Ti.

                  Here is the comparison of the RX560 D vs the RX 460 from GPU Boss



                  And here is the comparison of the RX 560 D vs the RX 560 also on GPU Boss

                  Comment


                  • #29
                    Argh!
                    Yes, AMD offered a slightly slim-down version of the 560 (different SKU number). It was to be sold as RX 560D. A model with 75 W max iirc. so PCIe slot power only, which might be attractive under circumstances. Also possibly to offer a cheaper model (even though it's probably the same HW). Some vendors seemed just to put an RX 560 without any note, which is not a good thing because customers should be informed and be freely choosing between low power consumption model (with less active units, though) or the full bang.

                    AMD adressed the issue already. (Unlike a certain competition that sold defective chips knowingly(!) for years. Or that came up with false specs from the start and didn't admit it until it was so screaming obvious that even the heavens took notice of it.)
                    It's wasn't really good for PR, surely, but it's being handled now.
                    Stop TCPA, stupid software patents and corrupt politicians!

                    Comment


                    • #30
                      Originally posted by timofonic View Post

                      I hope Intel's GVT will give enough pressure to AMD and maybe Nvidia too to provide these features on consumer cards. I know this is really just about market segmentation and not so much about special sauce on certain chips, maybe they even disable/enable it by using fuses or/and any other way (firmware?).

                      Many users like me could barely afford a somewhat high end laptop, but it would be very positive to be able to have "vGPU" features when dealing with Virtual Machines and mixed environments (or needing to use software in different platforms due to the work environment or testing or whatever). AMD, please consider it.

                      bridgman Could you give us information about Open Source GPU virtualization for consumer hardware?
                      SR-IOV is a hardware feature iirc, I had a discussion about this on reddit with Bridgman months ago(or maybe it was on Phoronix), SR-IOV is similar to Nvidia's GRID or something. All I remember is it is not something that AMD can just offer via some driver update. The Intel GVT stuff had 3 types, the one most people know of and that is desirable(splitting into vGPUs) only supports Broadwell and up due to hardware reasons iirc, this capability had been in the works for a while. You'll more than likely find that AMD would have to implement some architecture change to properly support vGPU or bring SR-IOV to consumer cards.

                      Comment

                      Working...
                      X