Announcement

Collapse
No announcement yet.

Ryzen 3 2200G Video Memory Size Testing On Linux

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

  • Ryzen 3 2200G Video Memory Size Testing On Linux

    Phoronix: Ryzen 3 2200G Video Memory Size Testing On Linux

    One of the discussion items in the forums this week was about the video memory allowance for the Vega graphics on Raven Ridge APUs as well as efficiences or inefficiencies around the TTM memory manager as used by the AMDGPU kernel driver. Here are some vRAM size tests with the Ryzen 3 2200G...

    http://www.phoronix.com/scan.php?pag...200G-vRAM-Size

  • #2
    tl;dr use 1GB for now

    Comment


    • #3
      Interesting results! It seems like the 8GB systems is a bit limiting to give away 2gb for the gpu, and overall 1gb seems to be best tradeof. I guess if you look at the ram usage is too high for the games that perform better under 512 or 1gb for vram. Now it would be interesting to see the same test with a 4gb ram and a 16gb ram system.

      Comment


      • #4
        In Windows, there is no performance drop at all, even when you drop to 64MB. This is because it automatically allocates to system memory, which the GPU was already using in the first place. It seems Linux doesn't do this, or not efficiently anyway.

        Comment


        • #5
          Originally posted by schmidtbag View Post
          In Windows, there is no performance drop at all, even when you drop to 64MB. This is because it automatically allocates to system memory, which the GPU was already using in the first place. It seems Linux doesn't do this, or not efficiently anyway.
          Yeah, I was about to say the same. Techspot did a nice test on this, and the worst cases never seem more than 2 FPS drop.

          But they also warned that some software may not recognize things correctly, and can refuse to launch if they think only 64 or 512MB is available for video.

          Anyway, 8GB minus 1GB is still a very good amount of memory for today office use, unless you are one of those freaks with hundreds of tabs open on a browser.

          Comment


          • #6
            Originally posted by schmidtbag View Post
            In Windows, there is no performance drop at all, even when you drop to 64MB. This is because it automatically allocates to system memory, which the GPU was already using in the first place. It seems Linux doesn't do this, or not efficiently anyway.
            We support this on Linux just fine. The missing piece is enabling display from system memory (rather than carve out "vram") on Raven. So far we've only enabled this by default on carrizo and stoney.

            Comment


            • #7
              Originally posted by agd5f View Post
              We support this on Linux just fine. The missing piece is enabling display from system memory (rather than carve out "vram") on Raven. So far we've only enabled this by default on carrizo and stoney.
              Hate to break this to you, but if there's a "missing piece" then you do not in fact "support it on Linux just fine".

              Comment


              • #8
                Ok - so what is the point about associating video memory with FPS?

                When the engine is old it waits on missing textures to render the image.

                When it‘s better it just displays resources that are available which results probably in more FPS with less memory but black spots.

                A modern engine like the one used for Call of Duty: Modern Warfare 3 will just dynamically switch your textures to lower res textures.

                So you better compare the image quality as an indicator for sufficient video memory size or just measure the usage making sure it always stays below 80%.

                Comment


                • #9
                  Originally posted by Spooktra View Post

                  Hate to break this to you, but if there's a "missing piece" then you do not in fact "support it on Linux just fine".
                  I was referring to your comment about the driver being able to allocate and use system memory. This has been possible going back to pre-radeon hardware. The ability for the display hardware to scanout from scatter/gather system memory is a relatively new feature in hardware and is only now being enabled.

                  Comment


                  • #10
                    Originally posted by Spooktra View Post
                    Hate to break this to you, but if there's a "missing piece" then you do not in fact "support it on Linux just fine".
                    You guys are talking about different things. Alex was responding to "automatically allocating to system memory when VRAM is not available", while display placement is not an automatic thing - just the last remaining difference in memory usage between Windows & Linux.

                    That said, there are still enough performance differences with reduced "VRAM" carve-out that there may still be some swapping going on between "system" and "VRAM" pools... not sure, but that seems a likely explanation.

                    oooverclocker's point about game engines automatically adjusting quality based on available VRAM is an interesting point (one of the things we discovered at Ryzen launch was that a couple of games even raised quality when they saw >4 cores) but I *think* that would translate into higher FPS with lower carve-out, not lower.

                    Comment

                    Working...
                    X