Hi, Timothy, First congrats for your PhD and new faculty job!
I think we may focus on next-generation LOW-POWER GPU design and implementation (not necessary mobile GPUs). Lots of interesting things could be done along this direction: - How to load balance of ALU/TEX instructions? - How to design low-power TEX caches? - How to design efficient power gating algorithms for GPU Shader Cores? - How to accurately predict power for clamping GPU clock in case power exceeds TPD? - etc. Since we have OGP, it will help us come up a more low-level (not only C-model like other academia papers did), more robust evaluations for proposed algorithms or low-power GPU arch. I would say it's good in the academia area, especially for your grant applications and publications. Thanks, Xiaohan 2012/5/27 Timothy Normand Miller <[email protected]> > Hi, everyone, > > I'm sure you've noticed that the OGP has been rather dead for a good > long time. I'm hoping to find opportunity to change that in the not > too distant future, and I'm looking for some suggestions and > discussion on this. Let's keep in mind that although the OGP is > ostensibly about graphics, many of us are interested in open hardware > in general, and I don't consider any open hardware project to be off > topic on OGML. > > Having completed the requirements for my Ph.D., I am now, as Jeremy > Clarkson would say, "a Doctor of Engineering." Additionally, I am > pleased to accept the position of Assistant Professor in the Computer > Science department at Binghamton University in New York. There, I > will be teaching Computer Architecture and doing research in that > area, with a heavy emphasis on energy efficiency. You can get my CV > at https://www.cse.ohio-state.edu/~millerti/timothy-miller-cv-online2.pdf > to see what I've done so far, and Binghamton hired me to continue > doing that. I will also be at least indirectly associated with E3S > (http://www2.binghamton.edu/e3s/), collaborating with many engineers > of similar and different disciplines, making data centers more > efficient, and I will focus on the microprocessor. > > Now, I see shader-based GPUs as just another kind of many-core > processor, and with scientific computing taking heavy advantage of the > streaming and parallel nature of GPU cores even in the data center, I > see lots of opportunity to explore interesting GPU architectures and > GPU energy efficiency. The OGP has specifications for a shader > engine, and it would be interesting to see what we can do to the > design to optimize it in different ways. With research funding, we > can even turn FPGA-based prototypes into real standard cell ASICs. > > I like to generalize, though, and blur the boundaries between > different types of specialized microprocessors. AMD is already doing > this with their Fusion products; in a future generation, GPU shaders > will have virtual memory addressing, completely eliminating the > distinction between "graphics memory" and "main memory." When doing > research, we can be fanciful and not assume that the main CPU is > always x86, allowing us to be a bit more creative. > > As a research professor, my main job is to produce publishable, > fundable scientific results. (Fortunately for me, being a better > engineer than scientist, most of the innovation in my area is actually > engineering.) I have lots of ideas I plan to put into grant > proposals, but few of them are things that I think would align well > with the goals of the OGP. So what I'm interested in discussing is > research opportunities that WOULD align with the OGP. That would be > doubly awesome as far as I'm concerned, being able to do interesting > research, publishing in traditional channels, as well as contribute to > the FOSS community. Projects often work out best when there are many > more groups that benefit. It would be awesome to develop a feedback > loop between computer engineering research and open hardware. > > Additionally, I'm interested in just general, open discussion on these > topics. > > Thanks. > > -- > 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) >
_______________________________________________ Open-graphics mailing list [email protected] http://lists.duskglow.com/mailman/listinfo/open-graphics List service provided by Duskglow Consulting, LLC (www.duskglow.com)
