I think there are quite a few graphics and multimedia tasks that will run pretty well on it. so while the raspberrypi is very cool and cheap, but a bit disapointing as a desktop (speaking as a pi owner), the parallella will be as fast as a normal desktop for some things (the ARM cpu is already several times faster than the pi's armv6).
of course the 16 and 64 core versions are not actual supercomputers by todays standards (i work with 48core opteron machine and would not call that a supercomputer). its a step on the way to 1024 and 4096 core chips. the limit in big supercomputing is power usage. i have seen HPC clusters where 1000s of 3 year old servers are thrown out because its cheaper to replace them a few hundred new machines, than pay for the electricity to run them.
there are many approaches to how to improve flops/watt. you can assume that standard CPUs will get a bit better every year by them selves, or you can try to come up with a whole new way of doing things. one of these is GPUs where you make huge use of SIMD (single instruction multiple data), which is great when you want to do exactly the same operation to each data value, but hopeless when you don't. another is put lots of full x86 cores on a single die, like intels MIC. epiphany is sort of a halfway, lots of simple but still capable independent cores on a chip. They also think that a network like memory system will be more efficient then a cache hierarchy. its hard to know whos right. time will tell.
There are mainly three reasons for me to prefer the Parallella: firstly the openness, secondly the cost, and thirdly the low power needs. The architectural differences may come in later. This won't mean the GPU loses its place; but if it succeeds it may lower the price for that MIC eventually.
For the cost it might be fun to play with. It really ends up depending on how easily it is to port existing code over to run on it. 90gflops does seem kind of low considering there are terraflop boards out there (at dramatically higher cost).
How well it handles double precision is of most interest to me, but that does push the 1GB ram a bit. Single precision would be more interesting to those wanting to try to leverage this thing as a 3d video accelerator.
NO double precision. Only single precision, and not all operations are performed in hw. (division is not handled)
What all here miss is application in mobile. Where low power and low cost can bring some benefits. That board will be one of (as this company hope) many devices with that technology.
I can see it as "addon" or "replacement" for current FPU coprocessors in mobile.
Guys, don't forget that 16 and 64 cores is only to get started. They plan for ~1.000 cores in two years.
Practical (in theory) solutions are still in labs.
But I still doubt the price of the switchers for the 2GB/s links, clearly there won't be many consumers of this kind of switchers.
Gbps Ethernet switchers are much cheaper, while they are not sufficient for large-scale clusters .