Announcement

Collapse
No announcement yet.

AMD Is Aiming For Radeon RX 5700 "Navi" Support In Linux 5.3 + Mesa 19.2

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

  • #21
    Originally posted by entropy View Post
    I was about to replace my Phenom II and Radeon 7950 with Zen 2 and Navi.
    What do you think should I at least wait for the new Navis in 2020 instead?

    It's said they'll feature ray tracing HW and a significantly different architecture. (?)
    At least for SONY's PS5 it's official that RT hardware will be included.
    While I agree RT is not (yet) a killer feature, I'm a bit worried that for the next gen games
    that will target PS5 and whatever MS comes up, the Navi iteration might be a much better match.

    Yeah, I'm thinking long term, as you can tell from my current specs.
    If there is no other reason than your eagerness to get fancy new HW, then yes.

    Comment


    • #22
      Originally posted by debianxfce View Post

      You do not need to buy anything if you have a fullhd monitor as the 7950 card suggest. It has a RX 560 performance. I have an academic interest for cheaper NAVI 5000 models after the 5700 model and the performance per dollar value must be good before I buy anything. My RX 570 is enough fast for 2K and 4K gaming. Freesync and LFC are great features.
      You can just upgrade to Zen 2 and leave your GPU alone for the time being. It's enough for 95% of games in FullHD. The Zen 2 upgrade,however, would already give you a massive upgrade.

      Comment


      • #23
        Originally posted by Veto View Post
        Except that Ryzen is not really x86, but AMD64 + lots of extensions (e.g. AVX), which is arguably a quite different ISA as well...
        Yep, although I think the analogy still holds... our GPU ISA has evolved over the years as well but you can still see common elements.

        You are right about AMD64 changing the programming model though (more and wider architectural registers) - I should have worded that differently.

        The point I was trying to make though was that even the changes from x86 to AMD64 were small in comparison to the massive changes in micro-architecture that have provided ongoing performance improvements. It's not a perfect analogy because GPUs are SIMD/SIMT behind the scenes while CPUs are superscalar, but it's close.

        Originally posted by Veto View Post
        I always found the commitment to the GCN ISA interesting, and wonder how long you are going to stick to it. With the history of rapidly changing GPU architectures and ISAs, it is impressive if you manage to pull of some sort of standardization at GPU ISA level that is both efficient and sustainable (with extensions of course)! I would guess RT will require some additions
        AFAICS most GPU vendor ISAs have largely converged already, and most of the changes now are in the underlying implementations rather than the programming models themselves.

        The main difference in programming model IIRC is that we include scalar instructions/registers alongside vector (SIMD/SIMT) instructions/registers while some other vendors do not. Anything past that gets down into what you might call extensions (operations, addressing modes etc..), and they change regularly between generations.
        Last edited by bridgman; 06-01-2019, 07:00 PM.

        Comment


        • #24
          thanks for the clarification bridgman
          can you share anything about the "next-gen", is it gonna be called RDNA too or something different?

          Comment


          • #25
            Sorry, but all the new product info goes through marketing first... I'm just keeping an eye on things to help keep our driver code aligned with marketing plans.
            Last edited by bridgman; 06-01-2019, 06:53 PM.

            Comment


            • #26
              Originally posted by bridgman View Post
              AFAICS most GPU vendor ISAs have largely converged already, and most of the changes now are in the underlying implementations rather than the programming models themselves.

              The main difference in programming model IIRC is that we include scalar instructions/registers alongside vector (SIMD/SIMT) instructions/registers while some other vendors do not. Anything past that gets down into what you might call extensions (operations, addressing modes etc..), and they change regularly between generations.
              Thanks a lot for your answer. Your informative posts are one of the reasons it is worthwhile frequenting this forum.

              Comment


              • #27
                Originally posted by entropy View Post
                I was about to replace my Phenom II and Radeon 7950 with Zen 2 and Navi.
                What do you think should I at least wait for the new Navis in 2020 instead?

                It's said they'll feature ray tracing HW and a significantly different architecture. (?)
                i'm sure 2030's videocards will be even better

                Comment


                • #28
                  Originally posted by pal666 View Post
                  i'm sure 2030's videocards will be even better
                  Your reading comprehension is shite, lad.

                  Comment


                  • #29
                    Originally posted by bridgman View Post
                    .
                    Can we have a Radeon7 edition with full 4096 shader cores? it should be ~7% faster than the normal 3840shader core one.

                    this should help AMD in benchmark comparisons to compete with Nvidia.
                    Phantom circuit Sequence Reducer Dyslexia

                    Comment


                    • #30
                      Originally posted by Qaridarium View Post

                      Can we have a Radeon7 edition with full 4096 shader cores? it should be ~7% faster than the normal 3840shader core one.

                      this should help AMD in benchmark comparisons to compete with Nvidia.
                      https://www.gamestar.de/artikel/amd-...s,3344763.html

                      Radeon Pro Vega II und Radeon Pro Vega II Duo release

                      Radeon Pro Vega II as 4096 shader cores and 32gb HBM2

                      and the "DUO" has

                      "
                      • 128 Compute Units
                      • 8192 Stream-shader cores
                      • 1,7 GHz
                      • up to 28,3 FP32 TFLOPS
                      • up to 64 GByte HBM2
                      • 1 TB/s
                      • 4096-bit ram Interface
                      • Thunderbolt 3 (up to 40Gb/s)
                      "
                      Phantom circuit Sequence Reducer Dyslexia

                      Comment

                      Working...
                      X