Announcement

Collapse
No announcement yet.

Would A Kickstarter Open-Source GPU Work?

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

  • movieman
    replied
    Originally posted by Caledar View Post
    It would be sad if one company in its own interests is allowed to hold a patents that make opensource projects impossible.
    That's, you know, kind of the point of patents: preventing other people from doing what you're doing without giving you money.

    And yes, they do make much open source hardware impossible.

    Leave a comment:


  • fbruno
    replied
    Hi All,
    My name is Frank Bruno and I'm the one who initially asked Michael his thoughts on the project.
    I'm actually tending to a sick relative right now, so i only had a chance to read the first few posts the other day. I will go through all the others in the coming week since I am very interested in everyones thoughts. To answer a couple of initial questions:

    1) Why not just do the kickstarter?
    I am one of two founders of the company in question. I am all for this, but I am trying to convince my partner this is worthwhile.

    2) What is the nature of the IP (paraphrased)

    My partner and I came from #9 and designed the four chips they did.
    The IP is a DX6/7 (possibly8) and OpenGL 2D/3D controller.
    The essentials:
    PCI (we have a version w/ AHB/ APB for ARM)
    CRT controller
    Full IBM compatible VGA (passes all Displaymate tests)
    2D engine:
    -Lines (error and non)
    -BitBLT
    -Solid Fills
    -Patterned fills
    - Blending

    3D engine:
    -Bilinear and Trilinear texture mapped triangles
    -Stretch copy mode (video)
    -Floating point parameters
    -Z buffering
    -Fog
    -Blending

    We also have other IP (Hardware VNC controller and some other stuff)

    It's all verilog w/ complete test bench.
    Drivers exist for Linux.
    T2R4 Windows drivers work as is (register compatible).
    The 2D design is ASIC proven.
    The 3D design is working on the board.
    This is very different than open graphics. These designs actually work running real OSes.

    The 2D controller fits in a small FPGA 20K LE or more.
    The 3D is about 85K LE part (I think, have to verify)

    It's a good starting point if someone wanted to try to switch to OpenCL or equivalent, but all of ouyr 3D functionality is a fixed function pipeline.

    Please feel free to follow up or message me your thoughts. I hope to be more active next week.

    Thanks,
    Frank

    Leave a comment:


  • Kivada
    replied
    Originally posted by timothyja View Post
    Well we really need more details on what they are aiming to do to make any worthwhile comments. However I would assume they would definatly not be aiming for a top end GPU.



    Sure it can take a look at the most successful projects page: http://www.kickstarter.com/discover/most-funded

    Many projects have gone into the millions with some notable ones:
    OUYA Console - $8,596,474
    Pebble ewatch - $10,266,845

    I wouldnt expect this project to reach such high numbers but I guess it depends on what they are going to offer. Probably the closest project I can think of to this one is the Parallella: http://www.kickstarter.com/projects/...r-for-everyone and that reached $898,921
    And those projects didn't actually create any new hardware, all they did was repackage existing parts in a new way.

    To make a new GPU from scratch it's cost at minimum around $100 million and at least 3x that much after dealing with all of the patent and licensing issues. And this would likely get us a first few runs of $1000+ parts that would be slower then a bottom end GPU from near a decade ago.

    I'm sure my numbers are near an order of magnitude short since I remember reading a while back that it would take around $300 million in dev time to rebuild the OSS graphics stack and get all of the GPU drivers into a decent state.

    Then theres the problem of finding enough people with the necessary skill set to design the GPU, these people are rather rare and likely already quite busy doing so for their actual job.

    The entrenched players are there because their investments till profitability where made in the 80's and 90's, these days they are just building of of thteir previous tech, trying to catch up to them would be near impossible without spending huge amounts of money at a loss for several years.

    So, short of Redhat buying S3 or some other small time GPU company and building off what they have I don't see how this can work well enough to become a self supporting project.

    In the end it's like breaking into the auto industry. It's a billion dollar risk to enter the game with no promise of payoff unless you do something very novel like Tesla Motors has done. They had a crazy billionaire engineer willing to risk everything on his dream car and it actually worked. His cars prove that electric cars can be fast, have good range and look awesome(Though I'm not feeling the "gull wing" doors on the Model X, which otherwise looks sweet) and he's kicking Ford's Focus Electric and Nissan's Leaf asses up and down the block.

    Leave a comment:


  • timothyja
    replied
    Originally posted by kertoxol View Post
    I don't know if a kickstarted open source gpu would work, but i know what would work: kickstarted open source gpu drivers (aka Nouveau and Radeon)
    I tried to get something like that off the ground see: http://phoronix.com/forums/showthrea...-For-a-1-a-Day

    The issue for me was tax. I could not take the donations directly and pay developers as I would have to pay tax or setup a non profit organisation. I did send some emails to some existing linux not for profits but never go a reply. I guess you could start up a campain if you were a dev with the skills to do the work. My idea was to try find some devs via odesk.

    Leave a comment:


  • Ibidem
    replied
    Originally posted by c117152 View Post
    This open GPU idea isn't "development of ideas or outside the box thinking". It's taking something we have on the shelf, from a half dozen companies, and reinventing it.
    If this was an initiative to buy the patent rights of an established and successful GPU, and then release them under a copyleft open hardware public license, I would be all for it. This way manufacturers would keep using it since it's already a success, but would also contribute improvements over time while keeping the tech accessible and affordable.
    And still, the amount of money necessary to pull something like this off is quite substantial. AMD sold their own to Qualcomm for 64M (http://www.engadget.com/2009/01/20/a...lcomm-for-64m/) and it was considered a bargain at the time.
    How many people need to go RTFA?

    This is the developer of an established GPU (though not a major one) asking for funding to release it under some sort of open hardware license.

    Leave a comment:


  • timothyja
    replied
    Originally posted by silix View Post
    People being happy with wasting money on a delusional project doesnt, in itself, make it a non - waste...
    I will say this one last time, Kickstater is about funding projects which are of interest to ones self. It still amazes me that people like yourself feel they have the right to tell people what they are allowed to do with their money.
    The project is not delusional the only thing delusional is your expectations. I've already provided a link to one such project which shows it can be achieved: http://www.logicbricks.com/logicBRIC...ideo-Clip.aspx

    From a pure interest and educational point of view this is worth some of my personnal money. Where else can you look at and tinker with the design of a GPU?
    Last edited by timothyja; 07-17-2013, 06:06 PM.

    Leave a comment:


  • duby229
    replied
    Originally posted by Caledar View Post
    Are patents the killer of opensource development projects? It would be sad if one company in its own interests is allowed to hold a patents that make opensource projects impossible.

    One company working for its own interest and that of its shareholders (make more $$$ exploit when it can etc..) v.s. humanity in its FOS style?
    Your thoughts?
    Unfortunately I'm not philosophically inclined....

    But I think you are correct about patent law. All that requires for a patent to be filed in the US is for it to have some possible application. It doesnt really matter how flawed the design is as long as it has a use of some kind. There are US patents for some ridiculous things.

    Leave a comment:


  • Caledar
    replied
    Are patents the killer of opensource development projects? It would be sad if one company in its own interests is allowed to hold a patents that make opensource projects impossible.

    One company working for its own interest and that of its shareholders (make more $$$ exploit when it can etc..) v.s. humanity in its FOS style?
    Your thoughts?

    Leave a comment:


  • riklaunim
    replied
    It would be worth it if:
    - it would work somewhere like Intel HD 3000 / 4000 (or better)
    - would be easily Linux supported

    The it could replace all those "weird" graphics on ARM-based single board computers that are supported nicely only on Android.

    Leave a comment:


  • artivision
    replied
    Originally posted by silix View Post
    People being happy with wasting money on a delusional project doesnt, in itself, make it a non - waste...
    no, this is more about "jumping on the bandwagon" than anything...

    the "idea" to be developed is all but new;

    the chip to be designed is not some new exotic architecture, rather something that is able to DECENTLY run EXISTING graphic API's - and that, in order to do so, will inevitably and necessarily employ a design not too dissimilar (wrt the structure and organizazion of the functional units) from its very competitors (*);

    and, implementing a gpu in an FPGA instead of an asic, is not "thinking out of the box", it squarely falls into the definition of "prototyping" as it's precisely what is done before manufacturing a custom chip (implementing the same logic but allowing for lower costs, higher economies of scale and, often, higher clockspeeds - fpga's are usually clock limited, except very costly ones)

    (*) moreover: what nobody has thought of until now, is PATENTS
    this is no more a free/ open source sw project for an important yet commoditized component as in the case of your beloved unix like kernel and OS - protected by sw patents not being enforceable everywhere, and by prior art and portfolios on the part of Oracle (formerly Sun), IBM and others
    this is about hardware, a physical product that is built and sold, for which you cannot say patents dont apply and keep a straight face
    especially given who are he ones who hold related patents
    unless you're deliberately building something that will assure not to infringe on patents AMD or NVidia or the likes (who you can be sure will screw you till forever if you do) hold, by NOT being competitive on feature, performance, or both
    but then, what was the point of it again?

    everything that is required to implement any state of the art architectural or feature is patented by one or another, so a chip that is competitive ends up infringing one or more patents that would have to be negotiated and paid royalties for, there's no escape...
    moreover, with nvidia not even publishing their own specifications to avoid implementation details leaking out, if an open source hw design happened to have coinciding details, do you really think nvidia wouldnt crush it immediately?



    Implementing a GPU ISA to an FPGA is wrong. Having a 20 layer GPGPU FPGA like Tabula's, that is capable as an ASIC, and then have it decide the best configuration on-the-fly for each source-program-algorithm is the best, that is called sophisticated_computing. Taking a MISC pseudo-ISA and automatically make it more complex blocks. Full programmable or partially makes no difference. If you try for an ASIC with a full GPU-ISA, you will have patent problems for sure and high maintaining costs for new shader-models. If you don't have a full SoC with Cpu an wireless as well, you will extinct inside 1-2 years. Also you need to consider that Leased-Free tech is also free. You may pay a 10% from your profits for 3-5 years but its free. You just have to use this or convince some one to have his technology under that.


    For me you can take Loongson L3C as a base CPU core: 1024bit-Fmac-Fpu, 10+dmips/mhz, 40million-transistors(with L1+L2), china has not patents for maths, MIPS has only patents for 2-4 instruction that they didn't use on early Loongson. Then you can have a Gpu near Vivante or Creative Stem-cell Apu, with Tflop/watt (Toshiba had Tflop/watt Dsp on 65nm), Vivante has 50Gflops/50-80mwatt on 28nm. You can also have it cemi-upgrade-able(Microcode for example) for new shader-models and instructions. Then add wireless between 802.11ac and Wimax2, and share capabilities, even unofficial. Also make ADSL routers with the same share capabilities, and program them to give the 80% of the speed to their boss when needed and 100% share when not. Then you can start an IP to IP web service, so we all have different IPs even if we are connected via Shared-Wireless. At the end do cheap pocket-PCs with gestures via camera or infrared or other tech, and 720p pico laser projectors.

    At the end if you just want to make your fun, don't even start it, you will be hated. If you want to go free, then respect the 2 sacred rules: 1.Technologically Advanced, not common, ISA is for monopoly. 2.Destroy the equal closed one.
    Last edited by artivision; 07-17-2013, 01:31 PM.

    Leave a comment:

Working...
X