Are you referring to HQ? HQ's intended functions are VERY limited. We're working on (and off) a shader engine design that is completely different from HQ, designed for graphics workloads. HQ needs to be small because the XP10 is a small FPGA. Also, an int mult in HQ would be useful, but the XP10 doesn't have any dedicated multiplier blocks.
Personally, I think that FPGAs would be great for solving lots of parallel problems, where we design FPGA logic to replace software. The problem is the learning curve for writing HDL. Cray had Xilinx FPGAs in some of their supercomputers, but they went largely unused because people didn't want to write HDL. There is some research into conversion of C to HDL, but the results aren't general enough yet, and most of the time and money is spent on GPGPU instead. Your draft looks interesting. I'll have to read it a little later today. BTW, my Ph.D. research is on the reliability problems that face low power and highly parallel systems. See my home page. On Wed, Sep 15, 2010 at 1:20 PM, David Hilvert <[email protected]> wrote: > I'd noticed that you've specified a fairly general-purpose processor > logic core in your design, omitting a multiplier from this for now. > Have you considered designs exposing multiple programmable processors, > or do you think this would be feasible for an initial (or perhaps later) > design? > > There seems to be a fair amount of interest in using commodity hardware > for solving highly parallel problems, apparently including an earlier > post on this list, and since graphics seems to consistently fall into > this class, it seems natural to wonder how general the design could be. > > The question is mostly motivated by working on some fairly parallel > problems for a little while; I've put together a summary of some > relevant ideas below, but it is a very general and non-technical draft, > and I suspect that others will have better, more complete ideas than > these. > > http://auricle.dyndns.org/agc.html > > > > > _______________________________________________ > 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)
