Announcement

Collapse
No announcement yet.

AMD Radeon RX 5500 XT 8GB Linux Gaming Performance

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

  • AMD Radeon RX 5500 XT 8GB Linux Gaming Performance

    Phoronix: AMD Radeon RX 5500 XT 8GB Linux Gaming Performance

    Last week AMD launched the Radeon RX 5500 XT graphics card as the sub-$200 Navi 14 graphics card in versions with either 4GB or 8GB of GDDR6 video memory. In our launch-day Radeon RX 5500 XT Linux testing the benchmarks of this budget 7nm graphics card was done using the 4GB review sample, but with Phoronix readers being curious about the 8GB version, I bought the GIGABYTE Radeon RX 5500 XT GV-R55XTOC-8GD for some additional Linux testing. Here are those results.

    http://www.phoronix.com/vr.php?view=28650

  • #2
    Strange - unlike Windows, the 8GB model doesn't really offer any performance benefit. Not sure if that's a good or a bad thing - it could imply Linux has better memory management, or, it could mean that the drivers aren't properly taking advantage of all the available memory.

    Comment


    • #3
      How is the power comsumption measured? Is it just software values?
      If yes, then the values of GeForces and Radeons can't be compared, as the latter only measure GPU (+HBM), while Nvidia measures the whole card.

      Comment


      • #4
        Originally posted by schmidtbag View Post
        Strange - unlike Windows, the 8GB model doesn't really offer any performance benefit. Not sure if that's a good or a bad thing - it could imply Linux has better memory management, or, it could mean that the drivers aren't properly taking advantage of all the available memory.
        How can Linux kernel's memory management play a role here? The kernel has almost 0 say in what gets put into Video RAM. The drivers probably use a bit of that but if you run a game/benchmark the app is doing all the VRAM control (directly or indirectly). I suspect this issue is more of a driver/reporting issue and that the 8GB wasn't actually being utilized on Linux because of it.

        Comment


        • #5
          The Sapphire PULSE Radeon 5500 XT has a BIOS switch that lets you choose between stock or OC settings. The latter apparently gives 1-2 % extra performance at about 10 % higher power draw, and is the default setting. Are you testing the card with the default OC BIOS or have you changed it to the stock BIOS?

          Also, some typos:

          ... you need to be using the yet-to-be-stablized Linux 5.5 kernel ...
          ... coming in at the %169 price point ...

          A very interesting test, thanks! So far I'm glad I bought the 4 GiB version.

          Comment


          • #6
            Originally posted by Almindor View Post
            How can Linux kernel's memory management play a role here? The kernel has almost 0 say in what gets put into Video RAM. The drivers probably use a bit of that but if you run a game/benchmark the app is doing all the VRAM control (directly or indirectly). I suspect this issue is more of a driver/reporting issue and that the 8GB wasn't actually being utilized on Linux because of it.
            I didn't say the kernel's memory management played a role... I was implying the driver's memory management. Though re-reading my post, I guess I can see why you'd think otherwise.
            Last edited by schmidtbag; 12-16-2019, 02:15 PM.

            Comment


            • #7
              Originally posted by Almindor View Post
              How can Linux kernel's memory management play a role here? The kernel has almost 0 say in what gets put into Video RAM. The drivers probably use a bit of that but if you run a game/benchmark the app is doing all the VRAM control (directly or indirectly). I suspect this issue is more of a driver/reporting issue and that the 8GB wasn't actually being utilized on Linux because of it.
              Not sure I agree.

              With OpenGL the app can hint but AFAIK the driver makes all the decisions about whether buffers are placed in system or device memory. With Vulkan the hints are a lot stronger but AFAIK the driver still makes the final decisions.

              Comment


              • #8
                They really need to drop the price on that card. On windows as well, the 1650 super is kicking it's ass.

                Comment


                • #9
                  I have made observations which are few but by occasion made me tend to think that Linux gaming on Radeon cards might indeed need less VRAM.
                  I wouldn't take these observations too seriously but it would be interesting to find out substantially.

                  However, the reason that both cards share the same speed in these tests is that the VRAM limit either hasn't been reached (only tested in 1080p) or this limit has been simply circumvented by discarding objects. I have seen this behavior with Windows games plenty times: When the VRAM limit was reached, the game simply exchanged the textures with blurred base textures to still reaching the same FPS. Especially tricky was Call of Duty Black Ops 3, where the texture quality had been changed dynamically in order to stay below 4 GiB VRAM.

                  Expecting a drop in FPS by lack of VRAM is technically a wrong assumption because the FPS drop can be can be circumvented by reducing the image quality. Unless a person observes the quality, no one would notice that the VRAM limit had been reached, resulting a decrease of gaming experience.

                  Edit: BTW, I honestly recommend to (pretty much) always buy the version with more VRAM. Having to live with less FPS is way less pain than having to deal with unknown issues due to lack of VRAM.

                  Edit2: In the second diagram for Shadow of the Tomb Raider we see that the RX 550 XT 8GB is slightly ahead of the 4GiB model. That could mean that there were some moments where the VRAM had been exceeded.
                  The unnoticeable, measured disadvantage in the other games might be due to overhead for managing more memory or due to a higher power consumption, reaching the board power limit more often. It's an almost unmeasurable drawback for having more resources available.
                  But in summary, what I read from these diagrams is that the drivers have some potential left to improve for this chip.
                  Last edited by oooverclocker; 12-16-2019, 02:57 PM.

                  Comment


                  • #10
                    Originally posted by schmidtbag View Post
                    Strange - unlike Windows, the 8GB model doesn't really offer any performance benefit. Not sure if that's a good or a bad thing - it could imply Linux has better memory management, or, it could mean that the drivers aren't properly taking advantage of all the available memory.
                    the few windows benchmarks I saw are about the same as here, with the 4gb version getting about equal or 1% better results than the 8gb card.

                    Comment

                    Working...
                    X