On Thursday 12 May 2005 23:43, Mike Matera wrote: > Tim, I have some thoughts to add about your whitepaper. I hope that you > (and the rest of the gang) find them interesting... > > The RTL should be open source from the get go. I have five observations > that make this case. Taken together I want to show that development of > open source RTL will promote community involvement and open standards > while creating profitable markets.
But will it give us a sustained line of open-source friendly hardware over, say, the next ten years? > Licensing hardware IP cores is a bad business. HDL code like Verilog > and VHDL require enormous support, well beyond what C++, for example, > would require. The behaviors of modules are too complex and not > abstract enough to simply offer users RTL and a manual. Paying > customers will expect support. In addition RTL optimized for one > technology is often very poorly suited for another. This is true of > FPGA to ASIC conversions. A good FPGA design should use almost > exclusively synchronous resets because timing is hard to control and > synchronous resets cost you nothing. However in an ASIC timing is much > easier to control and synchronous resets cost you transistors. A good > design may use exclusively asynchronous resets. This would require > potentially extensive changes to RTL level code. There's a long list of > companies that have gone out of business trying to sell hardware IP. Of > the companies that still do virtually all sell hardware IP targeted only > for their ASIC or FPGA technology. But why does not being able to sell the IP mean that you have to open source it? Yes, paying customers will expect support, but I doubt stealing "customers" would. And how are you going to convince investors that this is not going to happen? > Companies that develop cards based on OGP hardware IP could, as a > stipulation of the license, be required to give back any enhancements > they make to our code. The license could be flexible enough to let them > develop their own private modules but any user of an OGP based graphics > card should have the option of loading pure open source hardware into > their system if it is FPGA based. Thus we take advantage of our own > success. If OGP becomes the standard graphics platform we can be > guaranteed a rich and diverse set of developers. But who will check that? Are you going to be putting that competitor's chip under a microscope to see if they used the OGP IP without abiding its licence? > Further if the OGP reference design becomes the standard it opens the > market to smaller companies that will be able to produce a high end > graphics card without having to do the work of designing its algorithms > and driver. There's a C model of the 3D renderer that is completely open source already. It describes the algorithms used. We've already used it to experiment with accuracy issues and clever ways to implement a reciprocal function. > This will allow hardware vendors to compete on what > hardware vendors should compete on; time to market, performance, quality > and price. Having exclusive access to the knowledge of how to build a > 3D engine will no longer be a barrier to entry into the graphics card > arena. (Imagine what PCs would be like if every motherboard vendor had > to make their own processor, BIOS and operating system) Today this > barrier is what keeps two players on top and everybody else (including > the open source community) fighting over scraps. But who is going to invest enough money to start actually producing the chips and the cards? > It is unlikely that the OGP will be able to deliver a traditional 3D > renderer that is more powerful than one of the big vendor's versions. That was never the goal of the project. > The scale of the ASICS that go into the high end cards (300M > transistors, 130nm process's) is simply not attainable to anyone but a > large company. It would require an enormous FPGA (>1B transistors) to > rival the computing power with a traditional 3D design. Such an FPGA > alone could cost two times more than the super-high-end graphics card. By that time, there wouldn't be an FPGA of course, it'd be an ASIC. And we might never get there, indeed. But then, if you want to play games, you can always buy a PlayStation. > Also what is the benefit of going after middle of the road cards? Most > middle of the road cards are adequately supported in free operating > systems. But if you buy 200 new GNU/Linux desktops for your company then you're not going to have the option "second-hand G400" on your suppliers feature sheet. > The complaint that existing drivers, both open source and > proprietary, for high end graphics cards have poor performance brings us > right back to developing for the high end. The potential of open > hardware lies in a user's ability to tailor graphics hardware to their > application. (Imagine upgrading the graphics algorithms on your card as > you upgraded to the new MESA GL which takes advantage of them; all with > a software installer!) Open source is a strength not a weakness. A > programmable open source graphics card could take advantage of > innovation that today cannot exist. That's nice, but right now, I'd rather have something that will not cost too much, and be able to power my next free desktop. Oh, and it would be cool if it could accelerate my Blender edit view, and let me play a game of Tuxracer now and then. > Finally if you intend to sell the hardware IP for profit you should hire > engineers and pay them a salary. My interest in volunteering is to > provide the community, not your investors, with the benefit of my labor. > I know that you have to live, work and eat just like me. My intention > is not to offend you. I want to point out that it is not realistic to > expect that people will volunteer for someone else's for-profit > enterprise. My day job pays me to make investors money. That is a good point. What about a non-profit? > The OGP has tremendous potential not only to inspire innovation but to > make people a lot of money. You still seem to have forgotten to tell us how exactly it is going to make anyone any money, or how I'm going to get my open-spec video card. Lourens
pgp3XuAT5JxaC.pgp
Description: PGP signature
_______________________________________________ Open-graphics mailing list [email protected] http://lists.duskglow.com/mailman/listinfo/open-graphics List service provided by Duskglow Consulting, LLC (www.duskglow.com)
