On Sun, Dec 09, 2012 at 02:00:51PM -0500, Timothy Normand Miller wrote: > I really like this Steam idea. If we could get them to participate in the > project, that would be phenomenal. Should we wait until I've had time to > code the reference stream processor? > > Also, Jeri's accomplishments are certainly impressive. But doing our own > fabrication is billions of dollars outside of the scope of this project, > You may be able to make a single IC in your garage with a few dozen > transistors on it, but that's a far cry from mass production with high > yield of chips with hundreds of thousands of transistors at 45nm or smaller > geometry.
Let's just go for 130nm process, and make GDS files with alliance and the http://www.vlsitechnology.org/ cell libraries. That doesn't cost anything but time, which, in theory, we have plenty of. I spent an hour or so and I think I managed to run a flow to make a 4 bit adder GDS file. As part of the energy characterization research, if you can code up a single stream processor, and we fab it on 130nm with 300 test points for characterizing the energy use, we have something of real value to the academic research goals. I think it would be a usefull (and low cost) benchmark to be able to make a few thousand research shader chips that grad students can play with running different microcodes and clock speeds and watching a thermal camera image of the package, or even put a window on it like the old UV-EEproms used to have... Sounds like an NSF grant to me... In say 10 years, someone at a fab that's closing down will make me an offer.. If you bring a truck, you can have all the 130nm equipment you can haul away. Heck, I might even get paid to 'dispose' of it. That's the zero-dollar, very long time version. Billions if we ship next year, -$1200 (disposal payment) if we ship last year's technology in a decade from now. _______________________________________________ Open-graphics mailing list [email protected] http://lists.duskglow.com/mailman/listinfo/open-graphics List service provided by Duskglow Consulting, LLC (www.duskglow.com)
