The thing about building high performance clusters with Windows is:
(1) No one wants to waste time screwing around trying to understand the various licensing conditions.
=> http://www.microsoft.com/hpc/en/us/how-to-buy/volume-licensing.aspx
With Linux, I download a distro and use it on as many machines as I want; real or virtual. If I want commercial support, I'll find a reputable party and pay for their services. If I choose to do things DIY, I can save money and learn something. (I often go for the latter because I love to learn new skills that could be put to practical use.)...Regardless, I'm in control of MY systems. I decide what's best for ME.
(2) The money you saved in NOT investing in Windows for your HPC implementation can be diverted for more crunching nodes (hardware).
(3) The inherit advantage of using Linux is the ability to modify the software (at a source code level), to meet your needs in a precise manner.
HPC goes with Linux like a glove because it plays to Linux's fundamental advantages.
Microsoft trying to fit Windows into markets like HPC and tablets doesn't work well because it doesn't play to Windows's own strengths. Its analogous to spending millions on trying to fit a square peg into a round whole...MS needs to seriously modify Windows until it doesn't even look like Windows; so as to fit into those markets. (Something it won't really want to do, as it costs time and money. Not to mention the need to come up with a different business model for it to be profitable.)
"Furthermore, Microsoft claims that Windows HPC Server is 32 percent to 51 percent less expensive than Linux-based HPC systems over five years."
Microsoft makes claims all the time. Its never realistic when you experience things in the real world. (Its lies to make their solutions look good).
I learned back when MS initiated their "Get the Facts" campaign (server comparison against Linux), they tend to intentionally create studies that result in their products looking better.
=> http://en.wikipedia.org/wiki/Studies_related_to_Microsoft
Thanks to Microsoft's marketing department, I've grown to be skeptical (always questioning) studies conducted in order to promote products or ideals/beliefs outside of the computing realm. So I guess MS isn't all bad! They've given me awareness!
Then again, I question the use of Microsoft solutions in mission critical applications...Just look at the following example.
The London Stock Exchange once used MS-based solution for about 4 years. It cost LSE US$65 million to build and implement the platform...The problem was, it wasn't reliable throughout its 4 yrs of service.
The final straw that broke the camel's back was on Sept 2008; The platform suffered a major outage that lasted for 7 hours. (This is right when the US Govt was rescuing Freddie Mac and Fannie Mae! Imagine not being able to trade during that period?!)...This also affected the Johannesburg Stock Exchange, as they use LSE's system!
You have to understand that 7 hours is several lifetimes; when you realise trading is seen from the millisecond perspective!
This was intolerable for LSE, as their previous system was running 6 years straight without a single outage! The following 12-months after the outage incident, the LSE gradually ripped out the MS-based platform and replaced it with a *nix based solution. (Consisting of Linux and Solaris with an Oracle back-end database).
Result? They saved approx US$15 million compared to the MS-based solution. (Due to reduced hardware requirements, licensing costs, and technical staff)...Not to mention they will save another US$17 million per year (for 2010 to 2012) because the *nix based solution is 6 times faster in trades.
In the LSE's case, not using Microsoft based solutions for really important things actually saved them money. Heck, it turns out the savings pays for itself in a much shorter time!
I find that a common theme in general. Switching to Linux is painful in the short term, but saves you quite a bit in the long term...The good news for gamers is that there is a way to implement the use of DirectX 10/11 support. (Not the way we've seen in Wine via DirectX-to-OpenGL translator. But via a new graphics stack that is currently underdevelopment.)