This is why I have started with a precisely-defined aim: Build a detailed GPU simulator for research purposes that includes timing, energy, and reliability models.
SESC is, in a way, a simulator of a fictitious CPU. (Technically, it's the MIPS R10K ISA, but the timing and energy models are very approximate.) Nevertheless, SESC, along with SimpleScalar and several others, is great for many kinds of research. We've used it extensively and modified it extensively. Similarly, OGA2-sim is going to start out as a simulator of a fictitious GPU. But over time we will incorporate increasingly realistic timing, energy, and reliability models. This will make it a valid and attractive research tool. Those researchers who give back to the project will add valuably to our design. Once we have established ourselves as the dominant source of GPU simulation, we can leverage this as a way to entice both research and corporate entities to fabricate our synthesizable version. This will lead to establishing ourselves as the dominant source of a superior GPU design. So to summarize, the plan I've laid out is to FOCUS on a useful piece of software that we can do REALLY WELL, which is an obvious and necessary stepping-stone towards making real hardware anyway. On Fri, Jun 1, 2012 at 1:24 PM, "Ing. Daniel Rozsnyó" <[email protected]> wrote: > On 06/01/2012 07:02 PM, Dieter BSD wrote: >> >> Terry writes: >>> >>> Well, the way I understood the economics, the expensive thing was to get >>> the chips produced. >> >> Specifically, the big problem is the cost to have the mask made >> (photolithography). Machines to make the mask are very expensive. >> And you want the design to be perfect so that you don't have to >> come up with another $2M to make a rev2 mask. >> >> I suspect that if one could borrow the machine, the incremental >> cost to make a mask isn't that much, but that's just a guess. >> So if we could recruit someone that works in photolithography >> at a friendly company/university it might be possible to get a >> mask made for less? >> >> Alternately, are there any organizations with sufficiently deep >> pockets that would benefit from a documented graphics chip? >> Perhaps concerned with security but without the clout to demand >> info from ATI/Nvidia? Both the chip itself and closed drivers >> are subject from security holes, either bugs or intentional. >> In addition to security, OGP's chip could have features like >> being proven correct, and ECC, to decrease the possiblilty >> of errors. It's kinda important to get the right answer when >> designing a bridge. If the chip is sufficiently energy efficient, >> there should be demand for it for use in battery powered devices, >> and even AC powered devices benefit from lower power use and heat >> generation. >> >> And there is the crowd funding approach. >> >> Other ideas? > > > I would start on the beginning, who REALLY needs this kind of graphics chip? > If I look to catalogs to select an embedded processor, nearly all of them > got some graphics support. And as the embedded hardware is intended more and > more to Linux, it has some sort of driver, so you can use it (think of any > mobile, tablet or set top box). > > Or is the whole OGP a Stallman like neverending process/initiative? > > To not be just negative - from my point of view, the ones who can benefit > from accelerated graphics, are the users of micro-controllers - the market > is fairly big and as the user interface requirements grow, the developers > have to use bigger devices - and so far there is quite a big space between a > single purpose mcu (<= arm M0-M3) and a full featured linux system (arm > A8-9). Could you target this market? But then you of course do not need any > shaders, just 2D and basic 3D. > > Without a precisely definable aim (and knowing who the users will be), it > just seems like wasted time/effort. > > D. > > > > > _______________________________________________ > Open-graphics mailing list > [email protected] > http://lists.duskglow.com/mailman/listinfo/open-graphics > List service provided by Duskglow Consulting, LLC (www.duskglow.com) -- Timothy Normand Miller http://www.cse.ohio-state.edu/~millerti Open Graphics Project _______________________________________________ Open-graphics mailing list [email protected] http://lists.duskglow.com/mailman/listinfo/open-graphics List service provided by Duskglow Consulting, LLC (www.duskglow.com)
