Announcement

Collapse
No announcement yet.

Would A Kickstarter Open-Source GPU Work?

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

  • #81
    Originally posted by intellivision View Post
    That would prevent it being used by open source projects that have an aversion to copyleft licenses such as the BSDs.
    It should be released under a more permissive license so these projects can more readily utilise the code.
    "My thoughts are that if someone makes a change to the core it should be available for all." Hence GPL.

    The GPU design being GPL does not stop it being used on with BSD based OS's at all, it could even be used on Windows without an issue. I think GPL is a great licence for this project I probably wouldnt bother supporting it if it used a more permissive license.

    Originally posted by fbruno View Post
    1) The OpenGL level is c1998 technology 1.x?

    2) 25K for 2D w/ VGA. 85K for 3D w/ 2D & VGA. (need to confirm the 3D #s)
    Running on a low end FPGA (Cyclone or Spartan), it runs as fast as a 1998 Graphics accelerator. This can be tweaked. For an ASIC it would be potentially much faster.

    3) It would be GPL or similar. My thoughts are that if someone makes a change to the core it should be available for all.

    4) We would be targeting something in the 300-500K range.

    5) Look at T2R4 performance numbers and it is better than that, but fairly close.

    6) We have some other IP that we might put in for stretch goals. Hardware VNC is the nicest. Get a VNC connection for your BIOS screen on up.
    We'd probably release the board design files if someone wanted to build their own FPGA board or sell them.

    7) The #9 T2R4 driver works out of the box. It's not optimized though.

    I hope this helps. The IP does exist. If the kickstarter were successful, the turnaround would be fairly fast. We'd need to put the licensing headers in place, but if my partner goes along with this, we'll start them ahead of time. And also documentation, making sure it's up to date.
    Thanks for the info. This gives us a much better base to make comments on. I for one hope that you can convince you partner this is a good idea and are looking forward to backing it.

    Comment


    • #82
      Originally posted by LinX View Post
      Sun Sparc T1 and T2 are completely Open Source... HERE

      It would require some work but eventually it would be possible to make a completely open source PC, you'd just need money and maybe a Silicon Foundry along with a Fab.
      oh wow that sound nice actually, having a 64 bit processor as open hardware and adding an open GPU on the same die as what AMD did that would give more room for many different OSes to flourish easily (having in mind Haiku OS), imagine entirely a ultrabook-like using open source APU. a single driver src code for many OS without having delays of release of documentation and support. Intel so far is the only company that is actually doing something usefull to the open source community with its CPU and gpu driver although the performance doesn't shadows the windows drivers yet.


      I would fund a project that would combine a t2 spark cpu with an open GPU targetting openGL4.x and openCL 1.2 at first. oh dedicated mem is not required i can share my 16 gb of ram for a gpu like i did for sandy bridge on my desktop.

      Comment


      • #83
        Originally posted by intellivision View Post
        That would prevent it being used by open source projects that have an aversion to copyleft licenses such as the BSDs.
        It should be released under a more permissive license so these projects can more readily utilise the code.
        i agree with you a MIT license would be easier to use but i would add some clause making copyleft available in case of modification of the core code. does it make sense to you?

        Comment


        • #84
          Originally posted by Setlec View Post
          i agree with you a MIT license would be easier to use but i would add some clause making copyleft available in case of modification of the core code. does it make sense to you?
          I'd love to hear thoughts on licensing. I'm not averse to letting people use the code in any project they want, even if commercial. My main concern is that any modification from the register interface on one side, to the memory interface on the other should be made available. Pretty much most modifications of the core.

          If someone puts an AXI interface in place of the PCI, it should be made available.

          If someone changes the interface to a DDR3 or GDDR5, the IP for those interfaces shouldn't necessarily be made available (they are likely existing IP), but any changes to interface to it should be.

          A prudent example would be if someone put this in an Altera SOC or Xilinx SOC, making AXI interfaces for memory and processor interfaces, they should be given back to the community.

          Comment


          • #85
            Hi Frank,

            Best wishes to your sick relative. I hope he or her gets well soon.

            I've been working on the OpenRISC project for three years besides my FPGA/ASIC day job, and I must say that an open source GPU is really interesting, especially an ASIC proven with one with OpenGL drivers. There are a few open source graphics accelerators available already, but this seems far more complete than any other I have seen

            I used to work for the company that runs opencores.org, and a few years ago we tried to start a fundraiser to produce the world's first Open Source ASIC (opencores.org/donation), based on an OpenRISC SoC. So far it hasn't worked out all that well. I've reflected a bit over that, and there are a few things that could have been done better.
            The timing was really bad. This was done just before kickstarter had taken off, so the crowd funding hype hadn't really started. It was also difficult to make people understand what this was all about, as it wasn't something that could compete with an off-the-shelf ARM SoC. I also think we failed to provide use cases and cool demos. The stark truth is that moving pictures sell, and I think that a 3d engine could have made this project much more interesting to the general public. The Open Source OpenRISC ASIC hasn't been produced yet, but what we did was to produce a quite affordable FPGA board that was tailored for a FPGA implementation of the SoC and could therefore be populated with ASICs later on.

            So, here's my idea. Make a kickstarter to produce a dev board with displayport (mainly to avoid HDMI license problems) containing a reference SoC with your IP hooked up to an OpenRISC and some other peripherals. We have most of the cores, like ethernet, uart, ac97, spi already ASIC proven and available. I really think it will be hard to just sell an IP core to the general public without also providing some hardware. It would cost you to produce the board, but that's what the kickstarter money is for, and instead of selling the IP, you sell the boards and provide the IP for free. This is a bit similar to what Selling a board worked for parallella, although they already had an (closed-source) ASIC to begin with

            Also, if it makes sense financially, make a stretch goal to turn it into an eASIC or an Altera Hard Copy. That could give you some of the benefits of an ASIC without paying the full cost.

            And finally some notes regarding license. This is a bit of a grey area for RTL code, but based on what you have written, I would go for LGPL. The general view is that RTL code released under LGPL requires you to disclose changes to the LGPL-licensed IP, but doesn't affect other cores on the same chip. Some years ago, the lawyers at a large company that invested heavily in OpenRISC at the time came to this conclusion, and both we in the OpenRISC project and the open hardware guys at CERN (ohwr.org if you are interested) share this view. The OpenRISC can now be found in Samsung digital TVs and the Allwinner A31 SoC for example, and none of those companies have had any problems with using it. I know other companies that use GPL for their cores, but this is solely for the purpose of dual-licensing. They know that companies wont put a GPL-licensed core with their proprietary IP, so they can sell commercial licences. A BSD/MIT license would of course work too. A fellow Open RISCer came up with a new license especially suited for RTL code. You might want to check that out too (http://www.juliusbaxter.net/ohdl/)

            Whatever the outcome is, I'll probably support a kickstarter run. After running around at conferences and preaching the gospel about the benefits of Open Source IP, I should probably put my money where my mouth is

            Good luck

            Comment


            • #86
              Hi Olaf,
              Thanks for the response. The only issue with building a board like that is that they already exist. Most of the Altera and Xilinx boards have some form of display out (usually HDMI). One thought would be to maybe rework the interfaces to connect to OpenRiisc and provide something tailored to a Spartan or Cyclone eval board. Then software could get linux running with decent drivers.

              To be honest, the windows drivers for XP and before have support for DX and OpenGL. The open source linux driver is 2D and very basic. There was a company that did a 3D driver, but it is closed source. The good thing is the register set is exposed and the core is free to modify, so it would be great to get some real software people to tackle it.

              We also have some really cool stuff we may throw in, like hardware VNC.

              I supported Parallela. I think we have an advantage: we would open source it and people can do what they want. The disadvantage is we don't have much to offer in incentives for kickstarter.

              Thanks again. I'm meeting with my business partner on Friday and hopefully something will come of it.

              Comment


              • #87
                If there already exist suitable boards, then that's less work for you, no? You could get a bulk discount on one, then offer it preinstalled with the opened core as a kickstarter perk.

                Comment


                • #88
                  Originally posted by curaga View Post
                  If there already exist suitable boards, then that's less work for you, no? You could get a bulk discount on one, then offer it preinstalled with the opened core as a kickstarter perk.
                  That is true. The boards are expensive, though. So possibly it comes down to making our own w/ just the components we need to make it work. Something to think about.

                  Comment


                  • #89
                    Originally posted by fbruno View Post
                    That is true. The boards are expensive, though. So possibly it comes down to making our own w/ just the components we need to make it work. Something to think about.
                    curious about the cost of those boards, why are they expensive and what do they offer as specifications?

                    Comment


                    • #90
                      Originally posted by Setlec View Post
                      curious about the cost of those boards, why are they expensive and what do they offer as specifications?
                      They are generally expensive as they support many peripherals to show off the FPGA capabilities. They also have very high density connectors which allow expansion. Typically they include FPGA development tools. Also, a part large enough for OpenRISC and the Graphics would be on the order of the largest inexpensive parts.

                      Expensive is relative. Here are some examples: http://www.mouser.com/Embedded-Solut...ols/_/N-8mwzo/

                      Bear in mind the board required would be 100K LE or greater, Cyclone or Spartan family.

                      Comment

                      Working...
                      X