Announcement

Collapse
No announcement yet.

NVIDIA Open-Sources TensorRT Library Components

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

  • #21
    Now that TensorRT is open source, maybe someone can abstract it as a layer over SyCL or OpenCL instead of CUDA?

    Comment


    • #22
      Originally posted by betam4x View Post
      That out of the way, of course they aren't going to start open sourcing stuff. Then AMD and Intel will realize how stupid easy it is to integrate on current gen hardware. As it is, NVIDIA already does a LOT of stuff in software. Even the DirectX 12 implementations for certain GPUs are done mostly by software. That is one of the reasons why NVIDIA drivers are proprietary on Linux today and you can't find a decent open source equivalent. If NVIDIA created an open source driver, all those dirty little driver secrets would be spilled out.
      I think it isn't about "Nvidia doesn't want to open source their driver!" but about their approach to Nouveau driver. I don't think they should throw away all and help writing Nouveau but just release all needed firmware and some documentation to let developers write proper and working driver. Nouveau on Maxwell+ hardware is lacking firmware to do basic stuff like proper power management.

      Comment


      • #23
        Originally posted by skeevy420 View Post

        I made a joke and you replied by name calling. You're the one trolling in this context.
        if it was a joke, it was a very bad one.

        Comment


        • #24
          Originally posted by yoshi314 View Post

          they throw you a carrot, but they still keep the leash.
          I don't know why people care about CUDA these days, considering that OpenCL exists, and it's a standard controlled by the Khronos group. It's as irrelevant as DirectX.

          Comment


          • #25
            Originally posted by DoMiNeLa10 View Post
            I don't know why people care about CUDA these days, considering that OpenCL exists, and it's a standard controlled by the Khronos group. It's as irrelevant as DirectX.
            CUDA is exactly used for the same reason DirectX is used. It's the most popular API in its field.

            Comment


            • #26
              Originally posted by jojo7887 View Post
              They're not as open source friendly as much as we'd like to, but something is still better than nothing. I don't wanna sound like an Nvidia fan boy or anything, as I still consider AMD to be king when it comes to open source friendliness, and they're thriving at that.
              No, RISC-V is king when it comes to open source friendliness, and they're thriving at that.

              Comment


              • #27
                Originally posted by skeevy420 View Post
                Oopps, I did tripped it again.
                I typed some dumb stuff, got lost in the post

                Comment


                • #28
                  Originally posted by schmidtbag View Post
                  Well, CUDA is also open-sourced, but you still pretty much have to have Nvidia hardware. So basically they throw you a carrot, keep the leash, and they pull the carrot away on a string as soon as you get a little too close to it.
                  They put out a big fat stinky carrot, you chew on it only to realize it's not a carrot.

                  Comment


                  • #29
                    Originally posted by Vistaus View Post

                    No, RISC-V is king when it comes to open source friendliness, and they're thriving at that.
                    RISC-V is hardly thriving. I've never seen anything use it. Period.

                    Comment


                    • #30
                      Originally posted by DoMiNeLa10 View Post

                      I don't know why people care about CUDA these days, considering that OpenCL exists, and it's a standard controlled by the Khronos group. It's as irrelevant as DirectX.
                      For the same reason why OpenGL/Vulkan isn't as widespread in use as DirectX: DirectX, Cuda, and the like have actual dollars thrown at them to support continual development. While I haven't used CUDA (I haven't had the desire to learn it), it is likely a superior API to OpenCL and since NVIDIA owns the GPU and GPU Compute segments, it makes more sense to use an API that has the support of the GPU Vendor vs no support at all. If you use OpenCL and find a bug in the API, who do you turn to? Khronos Group will likely defer you to the GPU Vendor, the GPU Vendor will likely ignore you or refer you back to Khronos.

                      OpenGL was/is bad because it allows for vendor specific extensions. This means NVIDIA can effectively implement their own API that is incompatible with AMD/ATI. DirectX has a standard API you have to use. If you wanted to add extensions, you have to talk to Microsoft, and they will create a reference implementation for you and competitors to follow. NVIDIA STILL tried to work around this by implementing Gameworks, which is a software library, optimized for NVIDIA GPUs, and is intentionally NOT optimized for competitors. They then PAID developers to use it to cripple the competition.

                      Vulkan has continued this trend. The only advantage/strength of Vulkan and OpenGL is cross platform compatibility. People who claim the API is superior are wrong. Khronos doesn't have the clout to dictate a standard API like Microsoft does. The Linux community could have done it, but IMHO it is far too late for that. Even Apple has clamped down on it's 3D API. They now dictate the use of their own API for everything, which pretty much killed the use of NVIDIA GPUs on the latest version of the MacOS. They can get away with this because they own 10% of the personal/professional computing market, and a large percentage of the mobile market. Khronos has no devices and therefore no marketshare. Their APIs are merely a suggestion.

                      Comment

                      Working...
                      X