There are folks out there doing just what you describe. http://www.conveycomputer.com/technology-01.html http://www.conveycomputer.com/technology-02.html http://www.conveycomputer.com/Resources/PersonalityDevelopmentKit.pdf
Full disclosure: Since the time I was involved in OGD1 stuff I have changed jobs and now work for Convey Computer. It is one of the reasons I have been quiet since discussion popped up again. Several of the topics mentioned align very much with what we do so I have to ensure I don't run into conflicts of interest with what I say. --wpm ________________________________ From: ""Ing. Daniel Rozsnyó"" <[email protected]> To: [email protected] Sent: Wednesday, May 30, 2012 2:02 PM Subject: Re: [Open-graphics] Support integer ops in GPU? On 05/30/2012 08:43 PM, [email protected] wrote: > On Wed, 30 May 2012 14:26:56 -0400, Timothy Normand Miller wrote: >> Someone pointed out to me that integer ops don't have much utility in >> GPUs. Memory may be in integer form, but ARGB in 8888 format, for >> instance, would get converted to/from vec4 format (or our equivalent) >> as part of the memory access. If we have no need for integer ops, >> that would simplify the architecture. So can anyone make any >> arguments as to why it would be strictly necessary to support integer >> operations? > > loop counters ? > compute array indices ? > helpful for crypto acceleration ? > (we're speaking about making an energy-efficient > system so if it means reusing it for other common > tasks, great) > > just wondering. > When you introduce a programmable device, it won't be energy efficient if it is used to task which it was not designed. Surely, I can emulate an x86 on my 8bit AVR, but the efficiency is.... you know where :) If you are trying to develop a new architecture, here is an idea: - create specialized blocks for realizing tasks very effectively (in means of time, power) - allow these blocks to be connected in a user/programmer defined way - allow to replicate blocks to increase performance The realization of that might be a runtime reconfigurable FPGA like structure.. OR if you are rather into hardware, the chip can bear a very large number of these blocks of various complexity and computing power, each of them be clock-gated so instead of dropping them out of the design you just switch it off. Beat the efficiency of that. (what none of these blocks could do is a task for a processor core) (Once I wanted to do an PhD in CS and such an architecture was a basis of merging software and hardware together - whole apps would run in hardware, having stdin/stdout piped from one to another, a | bash operator would become a fifo and so..) 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) _______________________________________________ Open-graphics mailing list [email protected] http://lists.duskglow.com/mailman/listinfo/open-graphics List service provided by Duskglow Consulting, LLC (www.duskglow.com)
