Announcement

Collapse
No announcement yet.

Intel's OpenCL Beignet Project Is Gaining Ground

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

  • Intel's OpenCL Beignet Project Is Gaining Ground

    Phoronix: Intel's OpenCL Beignet Project Is Gaining Ground

    Beignet is the controversial project to provide OpenCL/GPGPU support for modern Intel GPUs on Linux. Since the first Beignet release in April, this open-source Intel OpenCL project has been making lots of progress...

    Phoronix, Linux Hardware Reviews, Linux hardware benchmarks, Linux server benchmarks, Linux benchmarking, Desktop Linux, Linux performance, Open Source graphics, Linux How To, Ubuntu benchmarks, Ubuntu hardware, Phoronix Test Suite

  • #2
    Beignet reminds me a lot of Mir

    Just in terms of the sheer, "we don't like upstream so we're going to reimplement everything from scratch" NIH syndromeness.

    At least this project will still have the same OpenCL API that apps can all share. I'm half surprised they aren't making their own incompatible API that will "better fit Intel hardware".

    Comment


    • #3
      Originally posted by smitty3268 View Post
      I'm half surprised they aren't making their own incompatible API that will "better fit Intel hardware".
      They are aware such approach would only alienate developers, since they'd need to support yet another back-end for their software. Otherwise, they might follow that path (for example, if they were the ones drawing the path to follow, instead of just another walker on such path, like they do on other areas creating their own instruction sets everyone else has to follow to be compatible).
      Also, I don't think it's a matter of not liking upstream (even if upstream would be perfect for them, I think they'd avoid it) but rather trying not to benefit their competitors from their work, partially (and only partially) killing the idea of open source development. Kind of what Mir with GPLv3 tries to do, IMO.
      I do like them for focusing on free drivers, though.

      Comment


      • #4
        Argh

        This seems so stupid...
        Why can't they just develop for Gallium3D instead?

        Comment


        • #5
          Can you even use MESA driver with Galium3D? Can you mix different GPU driver with another one for OpenCL?

          Comment


          • #6
            Originally posted by smitty3268 View Post
            Just in terms of the sheer, "we don't like upstream so we're going to reimplement everything from scratch" NIH syndromeness.
            If Intel does not like Gallium3D for whatever reason, then they are free to write their own things. This is no better or worse than Ubuntu with Mir.
            Originally posted by Redi44 View Post
            Can you even use MESA driver with Galium3D? Can you mix different GPU driver with another one for OpenCL?
            To find out, try to run beignet with the ilo driver. Then report the results here.
            Last edited by chithanh; 18 August 2013, 11:01 AM.

            Comment


            • #7
              Originally posted by chithanh View Post
              If Intel does not like Gallium3D for whatever reason, then they are free to write their own things.
              I find it funny when someone assumes a critic is equal to "OH, GOD, BAN THEM FROM LIFE!!!". They are free to do whatever they like, and people are free to like or dislike such things. And that leads to controversy. Some people will probably like it, and some people will probably dislike it. And there's people in between, like me.

              On my post, which I wrote drunk and missed a few facts, I want to point out that another reason I didn't talk about is the fact that they already invested lots of work on their classic mesa driver, so it's normal they don't want to ditch it. The corporate reasons are still something pretty much likely, and partially fair to the companies (probably not the users, but to the companies, since they don't help Intel either when working on their closed source drivers). And considering how they help a lot on infrastructure pieces, they do help a lot of competitors' products users.

              Comment


              • #8
                I suspect that this is at least partially to support embedded systems. Having OpenCL makes a lot of sense for some embedded applications and linking in vendor specific libraries (e.g. Intel's ipp and mkl) that match the target CPU you've picked is a common approach. Since AMD has been implementing OpenCL on their Fusion APUs perhaps Intel is getting some market pressure to do something similar. Nvidia's CUDA is the 800lb proprietary gorilla here so I fail to see how supporting an open API is a bad thing. Also, see the article 'Intel's Mesa Team Has Grown About 10x In Three Years'. Its possible we're seeing disjoint pieces of some larger strategy so its WAY to soon to grab out pitchforks and storm the castle over this.

                Comment


                • #9
                  Originally posted by 6L3ZZ View Post
                  I fail to see how supporting an open API is a bad thing.
                  Nobody stated such thing. The critics are on not using Gallium to share more code with other hw providers, not on adding support to anything.

                  Comment


                  • #10
                    Originally posted by 6L3ZZ View Post
                    I suspect that this is at least partially to support embedded systems. Having OpenCL makes a lot of sense for some embedded applications and linking in vendor specific libraries (e.g. Intel's ipp and mkl) that match the target CPU you've picked is a common approach. Since AMD has been implementing OpenCL on their Fusion APUs perhaps Intel is getting some market pressure to do something similar. Nvidia's CUDA is the 800lb proprietary gorilla here so I fail to see how supporting an open API is a bad thing. Also, see the article 'Intel's Mesa Team Has Grown About 10x In Three Years'. Its possible we're seeing disjoint pieces of some larger strategy so its WAY to soon to grab out pitchforks and storm the castle over this.
                    Yeah, talk about splitting hairs, for Linus' sake! Intel is producing, as usual, an open source solution for Linux. They honor the standard APIs. But they choose a path that doesn't seem the best in terms of code reuse by other OS partners. But hey, they need to move quickly and chase NVIDIA, I would assume that they assessed the situation and chose what they feel best works for them. Was that soooo bad, really? Chill, my friends, chill!

                    Cheers!

                    Comment

                    Working...
                    X