Announcement

Collapse
No announcement yet.

AMD Announces Radeon Open Compute ROCm 3.0

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

  • #21
    Originally posted by bug77 View Post

    I prefer the Windows way: plug stuff in, detect it, install drivers and make stuff work.
    What you have just described is horrible by comparison.
    Yes ,I love the way that installing Windows fails in the middle because the OS has no DVD driver for the DVD drive you are currently installing off of. Brilliant!!!

    Comment


    • #22
      There should be ROCm with NAVI support soon.
      https://www.amd.com/en/press-releases/2019-11-13-amd-radeon-pro-5000m-series-mobile-gpus-bring-high-performance-amd-rdna AMD now confirms that Navi 14 has indeed 24 Compute Units or 12 Workgroup Processors. At least 16" MacBook Pro will include the full blown model as an option

      Comment


      • #23
        Originally posted by bridgman View Post

        In fairness, if the range of supported Windows versions big as the range of Linux distro versions "the Windows way" would probably be worse than the Linux way. It's not an intrinsic advantage of Windows, just the combination of stable driver ABI and much smaller range of OS versions to support.
        It's a huge usability advantage. While you may debate whether it's intrinsic or not, users will keep flocking to Windows*.
        On Linux, not even Nvidia's drivers are automatically installed. But once you install them, you get support for the whole enchilada: 3D gfx, HDMI sound (never used/needed it, but it's supposedly there) and the ability to tap into your card's compute resources. With AMD you get a mainlined gfx driver (huge step forward), but then you're left wondering why you can't get compute outside of Ubuntu 18.04, because you're unaware there's a script for Arch Linux that can point you in the right direction. It's just sad.

        *I may have said Windows but MacOS works the same way, leaving Linux as the redhead kid on the block.

        Comment


        • #24
          It will take a while for application fully use Vulkan so OpenCL is currently a better option on the open source part. amdpgu-pro version runs well on APU although unofficially as it does not recognize Ryzen APU yet. ROCm version is slower but it is currently the only OpenCL running on Davinci Resolve.

          Comment


          • #25
            This is one of those tools that I've wanted to try for a long time now, but I haven't been able to simply because it literally supports no distros outside specific versions of Ubuntu and RHEL. I'm not going to waste hundreds of hours researching if I can get a frankestein installation to work or if my only options are either the docker (which segfaulted on its included tests when I gave it a quick try) or installing those distros on bare hardware.
            Last edited by Aeder; 19 November 2019, 12:32 AM.

            Comment


            • #26
              Originally posted by bug77 View Post
              I prefer the Windows way: plug stuff in, detect it, install drivers and make stuff work.
              i prefer the linux way: it just works out of the box

              Comment


              • #27
                Originally posted by bug77 View Post

                I prefer the Windows way: plug stuff in, detect it, install drivers and make stuff work.
                What you have just described is horrible by comparison.
                Ahh, the windows way... where I can't even install the AMD graphics/CL drivers unless I re-map my Pictures/Documents folders back to c:\users\myaccount because the installer chokes if you've got any of the main user folders pointed somewhere other than the default (my Documents are on the spinning rust D: drive, and my pictures are on a mounted network drive with lots of backup/redundancy in place).

                Comment


                • #28
                  And to sum it up.

                  Windows do NOT run on supercomputers which this announcement talking about.

                  Comment


                  • #29
                    Originally posted by zxy_thf View Post
                    To be fair, CUDA is just slightly better; its installer won't stop you from continuing, but it's very likely you will have to stop due to incompatibility with newer gcc or newer kernel.
                    You don't have to use their installer. They offer packages for several major distros.

                    Comment


                    • #30
                      ROCm upstream integration into leading TensorFlow and PyTorch machine learning frameworks for applications like reinforcement learning, autonomous driving, and image and video detection.
                      AMD really needs to beef up its video decoder blocks, for this. A Tesla T4 (which is really the same TU104 that you find in a RTX 2080) can decode 1080p H.265 at over 2000 fps!

                      https://developer.nvidia.com/nvidia-...#NVDECFeatures

                      Besides releasing ROCm 3.0 for SC19, the rest of AMD's press release is mostly EPYC news
                      That's a shame. I was hoping for something about Arcturus.
                      Last edited by coder; 19 November 2019, 02:51 AM.

                      Comment

                      Working...
                      X