R-ten-K wrote:
The thing about the Blue Gene is that it offers a density of computation that is very very very hard to beat. Both Columbia and Blue Gene are geared towards similar algorithms, so there is no clear design win from Columbia's more flexible communication/memory hierarchy in each of the members of the cluster when it comes to compare it against BlueGene.
In any case, Blue Gene L provides 100+ Tflops in less space than columbia
In any case, Blue Gene L provides 100+ Tflops in less space than columbia
Not all algorithms are the same. Not every problem can be divided into small bits for MPI to shuffle around. Or not every one has the man/hours to divide a problem into small bits.
Another adventage for the Altix is the overall memory. Columbia has 20TB of RAM compared to 8TB for the Blue Gene (it's not publisized, but you can add it up, 512MB/node, 2CPU/node 32768 CPUs). And Columbia is no where maxed out, by putting the maximum density RAM, you could go 24 times the RAM.
Everything is relative on what you want to do with you supercomputer.