> What's on-topic here is that 120 Watts is a challenge to dissipate, > and you've only covered your CPU. Add to that another couple of CPU > dies, and your main memory, and all your conversion losses, and then > your GPU, and you've got a row of furnaces in your datacenter.
Absolutely. I've been trying to figure out how to explain for years that the fastest supercomputer in the world is probably going to be the one with the highest Bits/Joule benchmark. (http://bitspjoule.org) (on the furnace note, I would like to have some GPUs in my basement running the simulations mentioned below to heat my house this winter) <snip> > .... But I have other pressures. I have two major research > directions. One is energy efficiency in general (related to my prior > research), which I'm working to apply to GPUs, and then there's this > OpenShader sub-project, which is infrastructural. Ideally, I'd do > OpenShader first because it would facilitate all other research in the > area, but practicality may force us to compromise. The order and > proportion to which I focus on each one is going to be a function of > the help I can get, because I personally can only scale so far, and > Mark works on this entirely in his spare time. There needs to be more > than two people on this. > > What is the straw man here? The fact is, this infrastructural work is > going to be of immense value to researchers in GPU-related areas. > Whether or not you see it as feasible to turn this into consumer-grade > hardware is another matter. But if I have a complete design on my > desk (part of the whole process of back-annotating a simulator with > real timing and energy information), I am in an excellent position to > have this funded and fabricated. What do you need to make the OpenShader project go faster? What I'd like to be able to do is click a 'buy now' link and have some sub-$500 board show up in the mail a few weeks later that I can download a bitstream to, and hook up to some random HDMI monitor/TV, and start reading email, play videos, and maybe do some basic 3d. Bonus points if the board can also function as a bitcoin miner when it's not displaying graphics, as that helps with the heat my house part. If I start dreaming, the board I'm imagining is about the size of a beagleboard, arduino or raspberry pi, is supplied 48V power from POE (power over ethernet), and talks X-windows, or runs native linux, and also has the ability to supply power to my monitor as well (my samsungs take 14V at 2 amps) If I can't get a board, then maybe I could at least download the entire design and start figuring how what needs to be written or integrated together so this back-annotated simulation you mentioned above can be developed. -- Troy, da hozer _______________________________________________ Open-graphics mailing list [email protected] http://lists.duskglow.com/mailman/listinfo/open-graphics List service provided by Duskglow Consulting, LLC (www.duskglow.com)
