If you've seen the headlines saying Microsoft has added a Windows Server supercomputer to its Azure cloud service, I'm sure you're confused.
Microsoft, you may believe, does not make a supercomputer.
The definition of "supercomputer," you may believe, includes the words "not made or significantly influenced by Microsoft."
In this you'd be wrong. The definition does not contain those words.
It does say that modern supercomputers, rather than run a workload against a dozen or two high-performance vector processors, are actually massively parallel processing units that break up a workload into many small parts and run each on all the available servers and (where available and configured correctly) underworked Windows 7 machines.
Microsoft's HPC server is almost -- almost -- SETI@home.
On Azure, the HPC function is actually an extension of Microsoft's decision last week to allow customers to launch their own instances of Windows Server on the Azure service, rather than just running .NET or Visual Basic applications directly on Azure with comparatively little access to the operating resources their applications get.
The newest version is designed to break up technical workloads just as the previous version, but it also includes support for better graphics to allow for better performance for modeling applications, the ability to spread work units onto workstations, connections to network-based applications and specific support for large compute or update jobs in Microsoft Excel.
By allowing customers to launch instances of HPC Server, Microsoft isn't so much giving them a brand new operating system as allowing them to spread even enormous workloads across larger virtual cluster of machines.
In the physical world the difference between the version of your OS that can MPP across a cluster of servers and workstations and one that can't is stark. Among virtual machines, where workloads are already distributed and resources assigned logically rather than physically, the line is fuzzier.
If you had a workload prepped to run as a series of MPP processing units, you could probably run it against Azure without the HPC addition, as long as the number of work units wasn't too large.
With the HPC, that's not a worry.
It's still not a real high-performance computer, but it does give much more granular control to users over the number of VMs their workloads are spread across, the resources they use and the amount of time even compute-intensive jobs spend inside the blue cloud rather than chugging around on the cluster inside your own data center.