Announcement

Collapse
No announcement yet.

AMD Acquiring Another Company To Bolster Its AI Play

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

  • #11
    Originally posted by ezst036 View Post
    Aren't they going to need to mature their OpenCL stack further for this?
    I don't think so. OpenCL usage isn't really widespread these days. Or did I miss something?

    Comment


    • #12
      Originally posted by oleid View Post
      You mean GPU drivers?
      What does this have to do with AI stuff?
      Or what drivers are you taking about?
      Yes, GPU drivers. As far as I know, AMD's powerhouse "AI" products are primarily GPUs. I'm aware of the Alveo series but I get the impression that's more of a development platform.

      Comment


      • #13
        I saw the blurb on my financial news feed this morning.

        I think it's a horrible idea.

        AMD has been on a spending spree, they spent something like 125 million on that one AI company, then they spent another 665 million in cash on another company and now nearly 5 billion on this.

        They bought ATI in 2006, 18 years ago, and despite all this time, for the most part, they are still a few steps behind NVIDIA.

        I see no reason to expect that any of these acquisitions are going to be any different.

        Not to mention there are analysts that think NVIDIA will gave a significant pullback in its valuation, not because of competitive pressure from AMD and Intel but because 40% of NVIDIA's revenue comes from a handful of companies, like Microsoft, Amazon, Meta and Alphabet and all these companies are working on their own in house developed AI chips.

        If NVIDIA is facing downward pressure from competition created by their own customers, how is AMD going to gain market share.

        I half expect them to one day announce that they are buying S76.

        Comment


        • #14
          Originally posted by schmidtbag View Post
          Yes, GPU drivers. As far as I know, AMD's powerhouse "AI" products are primarily GPUs. I'm aware of the Alveo series but I get the impression that's more of a development platform.
          I was more referring to the userspace mesa code. But you probably mean the userspace ROCm code?

          Comment


          • #15
            Originally posted by schmidtbag View Post
            AMD,
            Just fix your damn drivers (which includes fully supporting devices older than last-gen), and re-instate ZLUDA. Those aren't too much to ask. In fact, proper driver support is the bare minimum. Your hardware is plenty good enough for AI, but nobody wants to buy it because of the abysmal state of your drivers and the looming threat of prematurely dropping support of devices. People are looking for an alternative to Nvidia, and you're not offering one.
            I bought Witcher 3 the other day and one thing it needs tweaked is the tessellation level clamped to 16x in order to use NVIDIA Hairworks on my 6700 XT. It's the difference between 85-90% usage (16x), 90-95% (32x), and 95-100% (64x) in and around the bar at the start of the game on DX 11 with Ultra+ settings at 3440x1440-60. Setting the tessellation is easy enough for DX11 in dxvk.conf, but I'm not sure if there's a way for DX12 and vkd3d-proton or Vulkan or OpenGL in general.

            Using DX12 in that same spot results in 100% GPU usage and thermal throttling until I lower things down to Medium-High settings. Just not worth the performance hit or drop in fidelity for HDR.

            Assuming it's possible, it'd be nice if there was some sort way to configure a lot of those kinds settings in a more centralized location, graphically or otherwise.
            Last edited by skeevy420; 19 August 2024, 01:54 PM.

            Comment


            • #16
              Originally posted by oleid View Post

              You mean GPU drivers? As in amdgpu?
              What does this have to do with AI stuff?
              Or what drivers are you taking about?
              It includes ROCm, too. AMD needs better support on consumer hardware for AI.

              Comment


              • #17
                Originally posted by oleid View Post
                I was more referring to the userspace mesa code. But you probably mean the userspace ROCm code?
                Yes; AMD's Mesa code (particularly OpenGL) is pretty solid. RADV is solid too. Everything else (which includes Windows), not so much.

                Comment


                • #18
                  Originally posted by skeevy420 View Post

                  It includes ROCm, too. AMD needs better support on consumer hardware for AI.
                  I'm with you that ROCm needs better consumer GPU support. But I wouldn't call this driver support, as it is userspace library code. Hence the question.

                  Comment


                  • #19
                    Originally posted by sophisticles View Post
                    They bought ATI in 2006, 18 years ago, and despite all this time, for the most part, they are still a few steps behind NVIDIA.
                    If you are referring to gaming, then I guess AMD is now ahead of nVidia, even if it's not yet showing in terms of the number of PCs running AMD vs nVidia graphics. The days of dGPU are now numbered as people move towards lighter and smaller laptops that are based on low power iGPUs, and with AI and NPUs those low power APUs will be even more feasible for producing AAA quality graphics. So, in few years most gamers will move towards those iGPU based products where nVidia right now has no presence.

                    Comment


                    • #20
                      Originally posted by oleid View Post

                      I'm with you that ROCm needs better consumer GPU support. But I wouldn't call this driver support, as it is userspace library code. Hence the question.
                      IMHO, ROCm blurs the line between driver and userspace because it's vendor provided and something that people have to have in order to do GPU compute.

                      Comment

                      Working...
                      X