That caught my eye, also. I'm assuming they are still referring to their bogus study on TCO.
Yeah, plus I'd say 99% of the HPC binaries are for Unix. Someone would have to port the programs to Windows. We use several programs that were discontinued by the authors from 1999. Source code ain't available.
I'd love to see them try that. TCO on a cluster is totally different from what they're used to doing. They're used to saying that training will kill the value of any switch, but in this market everybody's already used to UNIX variants.
Well, except for Cornell, which has the one Windows system on the Top500, down at #310. But Microsoft is about to start pumping money ($400K/year) into them so they can set up a bigger one. You know you're hurting when you have to pay someone to use your stuff.