how many flops? i believe current super computers require 6 ot 9 tflop of double precision.
even the best of the current consumer cpus onlt puts out less than .2tflop of single,
and our gpus that can put out 3-4tflops single grind to about 1tflop of double
so, again, the question is how many flops can this push.
Comment aside, I think it is great that things like this "supercomputer" are available for a normal person at this time, people will always discuss if the power is enough, how much it can deliver, etc. but I tink the idea itself is great and the creativity to case it with Legos just makes a it a little more fantastic
To continue my previous thought: I believe the hardware and software are already there to make a "desktop" parallel computer made of 128, 256 or even more system components, which individually is a complete smart phone circuit board-sized micro computer. Such a parallel computer would be very power-efficient as much as smart phones; a single component is sufficient for daily work and other system components are called incrementally as more computing power is demanded.
If you added in a few more nodes and used faster chips like the tegra, this might make for a very power efficient and redundant supercomputer that could be used on submersibles and rovers on mars and other planets that wouldn't require as much electricity and could handle loosing a few nodes.
Whoa, what's with all the negative comments about this? Yeah your home computer is more powerful, yeah you can build a computer with these specs for less, that isn't the point. The point is students and engineers at a university did this, for educational purposes and just for kicks and thrills. Universities always do things like this: things that teach the students involved things, and things that are pretty cool.
You aren't supposed to appreciate the power, you're supposed to appreciate the fact that these guys put together these little $35 independent computers and linked them all together to work in tandem to make a more powerful computer. I think it's pretty fascinating.
I think the comparison is correct.
Comparing the 64CPU setup to a regular PC would be incorrect as it is more like a multiCPU supercomputer.
The real supercomputer is faster, bigger and more expensive but this setup is not so much of interest for it's raw performance then for it's unique programming challenges.
I mean from a strictly practical point of view it is useless, like the first airplane and the first railroad, but all 3 are great learning platforms.
Once Nvida comes up with an HSA-compatible Tegra4 chip, I will definitely build my own. This is the coolest project. I will however use a different parallel technique than MPI.
A AMD APU-based one is another good option, but the energy saving is the novel reason to be playing around with this technology, so a ARM APU will be necessary before I continue with my version of this project.