Computational Science Stack Exchange is a question and answer site for scientists using computers to solve scientific problems. It only takes a minute to sign up.
Sign up to join this community
I am searching for the most important literature and slide references for modeling current and future energy costs of floating-point operations and data transfers across the CPU, memory, network, and storage. I have marked this question as a community wiki, and I would prefer that you limit each answer to the following format:
Title, Authors, Location/Conference/Journal, Year, DOI/URL, Summary of what information the paper or presentation provides relevant to my request.
There have been numerous panels and reports about exascale computing that should have these sorts of numbers. I would look for reports that came out from DoE (or were written for DoE). Also, there may be relevant stuff at http://www.exascale.org .
There's a lot of information online that search engines are capable of finding. Here's three hits to get the ball rolling:
IEEE MICRO 2011 paper "GPUs and the future of parallel computing" shows that state-of-the-art CPUs and GPUs incur ~1700pJ and ~225pJ, respectively, for each floating-point operation.
For data storage/access cost in different memories (e.g. SRAM, PCM, eDRAM), one has to use tools like CACTI and DESTINY (I am co-developer of DESTINY). Also, McPAT tool can be used for modeling energy consumption of entire processor, while DRAMSim2 is for memory alone. What you may be looking, may be achieved through combination or modification of these (they are open-source). There are many others, but from these, you can find connected references through Google scholar.
