Looking at the two articles provided, (and later doing a quick cross-check among other news providers) I couldn’t help but notice immediately how much more press the United States’ Titan received compared to the Chinese Tianhe-2. though I imagine the people’s republic’s ever-present translucent cloak of secrecy, and Nvidia’s want to market exorbitantly expensive GPUs to gamers, may have slightly dis-balanced the stories. Regardless, the other strange point that I noticed, was the the Chinese Tianhe-2, is run almost entirely from Intel CPUs, and in fact has no mention of a single GPU involved. Which, begs the question of what kind of power consumption would that computer require. Based on the power consumption, and relative performance of Jaguar computer to the Tianhe-2 provided in the listing provided by top500.org listing and the BBC article provided (above), the best guess I can manage is over 107.4 megawatts of power. Which to me, is completely mind-boggling.
For perspective, let’s say the grid cut out in that particular part of china where the computer is, so they have to put it on backup generators. Which, at the time, for whatever reason, only consumer-grade diesel generators are available. If they were quickly cart in some Cummins RV QD 8000 generators. Based on the information provided by Cummins, to safely power their supercomputer, they would need 14 generators, running at nearly maximum capacity. Which really isn’t too bad… except for when you consider that these aren’t some fit-in-your-minivan-for-some-wake-forest-tailgating generators. This emergency generator train would be sucking about 14 gallons of diesel every hour, and stacked atop each-other would be over 26 feet tall. On the consumption side, it’s even more staggering. According to eia, the power required by the Tianhe-2 could also be used to power over 82,263 average american homes. Which, is quite a bit to stomach in a country that has had historical issues keeping up with the consumption of it’s populace, and that the consumption of the average american household is certainly greater than that of an average Chinese household.
Switching focus to ideas for a project, I’d like to throw up the idea of creating a structural model simulation. Perhaps something similar in concept to a particular demo created by some nvida-affiliated developers, but not nearly as massive in scale. Something as simple as a structure of ridged parts, that if influenced with enough force would split would be effective in this means and fairly stunning to watch. Which can be done with even such a simple concept as a pillar, that when even slightly hit would sub divide into two rectangular prisms, and then those prisms divide and so on until the objects are small enough that their area calculation is below a certain threshold that they are considered insignificant and removed. All the while, all the objects are acting with rigid response to collisions, and being effected by a realistic gravity, and behaving in motion with a realistic inertia based on current area. All of which, I would imagine can be achieved using only parallel algorithms for the blocks, with a single over-arching program assigning names and resources to each newly created sub-division of mass. An idea that I’m surprised how elegant simple to code (I would think), and how complex to observe it would be. The real question is deciding how the blocks would divide. It could just be splitting the longest dimension in half, or choosing a dimension to split at random, or splitting the impacted side. Which, no matter I’m excited to see rendered.