That is probably because he has a very tiny penis. Iirc it was Freud who wrote on the topic of these kinds of inferiority complexes and how diseased minds deal with them.

Was not aware FreeBSD wasn’t good for superclusters, just assumed it was does do to the higher scalability, learn something new everyday

Not exactly HPC, but Mac’s do have an interesting grid computing system built in. Having used it, XGrid is pretty awesome. If I recall for a while UVA had the most FLOPs because they had something like 2,000 Macs in a cluster.

Only relative to an identical cluster running a non-free OS.
Take a look at, say China’s Earth Sim 2. 1024 cores for 130 Teraflops at ~100 GFLOPs per core, vs Xeons and their 18 GFLOPs per core, given that AFAIK Linux doesn’t run on NEC’s custom vector processors that power ES2, a Xeon would have to cost 1/5th that of an EC for a free OS to impact the upfront cost of hardware, and that excludes the cost on your power bill of having 5 times as many cores, and that’s further excluding other hardware costs, fewer processors equals fewer nodes equals less space in your data centre, and less storage and memory (you are after all, equipping your nodes equally), and you’re even getting a much more efficient cluster out of it (at 95% effic, ES2 is the most efficient cluster in the top 500).

Then there’s the OpenSolaris approach, the SparcxVIIs powering JAXA cost considerably less than NEC’s Vector CPUS (since they aren’t custom hardware), and output about ~45GFLOPs per CPU, vs ~18 more Xeons, 2.5 quad-xeons to one quad-SVII, and again, you’re looking at better power efficiency overall, fewer CPUS, fewer nodes, and a free OS, and at 91% effic, iot’s the second most efficient in the top 500 by a considerable margin (iirc IDRIS is next at ~80% on 4 times more cores for comparable overall output), and you’re looking at between 40% and 70% effic for the rest). And that’s only going to get better once Fujitsu finally unveils their VII/fx line which is supposed to pump out 120 GFLOPs.

So it depends on how you look at it, if you’ve got a cluster with 300,000 Xeon cores, Linux will probably save you some money, but if you’re looking at building a cluster to achieve a given level of output, while minimizing cost, you’re generally better off going less, more powerful hardware*, which is less than fully supported by Linux.

* There are caveats to this approach, however – you can’t expect IBM for example, to use NEC or Sun hardware in their BlueGene/RoadRunner/etc cluster systems – It ends up costing them less to go with their own Power/PPC/Cell hardware, same applies to any other HPC vendor who manufactures their own hardware.

2. It’s a fully monolith kernel, so when properly configured it’s as fast as possible.

That’s faulty logic at best, man, Monokernels aren’t inherently faster than Microkernels (that is to say a poorly designed monokernel is not inherently faster than a well-design Microkernel – look at any virtualization solution using a hypervisor that claims ~98% bare-metal performance – the hypervisor is little more than a glorified microkernel). Besides, individual node performance is capped by interconnect performance – the most streamlined kernel in the known universe won’t show you any performance gains when paired with a shitty interconnect, and even beyond that, there’s efficiency to take into account, Linux has yet to show that it can operate as efficiently in this role as Super-UX or Solaris.

4. Windows and Mac aren’t made for supercomputers, so it makes no sense to even be comparing them in this area.

Granted. Though MagicCube (powered by WinHPC) shows some promise. Cost ultimately works against it here, though.

5. People who work with super computers are highly trained and payed individuals, they can get the damn thing to work.

Granted. not necessarily related, Though I find it funny that FORTRAN is still the most dominantly used functional language for HPC. If nothing else, I’d say that GCC having a FORTRAN compiler by default is an advantage here.

6. This does not in any way shape or form make linux remotely good on the desktop, in fact, it proves the opposite.

Of course.

7. All of this could be achieved with FreeBSD, but people are trained to be linux sysadmins unfortunately.

Eeeh. Yes and no. FreeBSD isn’t designed with HPC in mind, though it scales higher than Linux, it just isn’t fit for superclusters, you’d have to work with much more, smaller nodes because the SMP isn’t all that great past 16 cores (to the freetards, that’s 16 cores per node, not per cluster) which becomes rather inefficient when you’re in the tens of thousands in terms of core count, not to mentions ramps up the cost of hardware regarding storage and memory per node, as well as power consumption, especially given that like Linux, it has less than full support (to put it nicely) for beefier, more efficient more cost-effective hardware, so you’re stuck with commodity hardware.

That being said, Linux is popular in HPC because commodity intel hardware is popular in HPC.

My server runs Linux. My phone runs Linux. My desktop does not. I use whatever is best for the job rather than base my decisions on bizarre ideologies. You are like a looney vegan that claims that they somehow have nicer food despite limiting themselves to a subset of it.

Sure, make your choice based on your warped idea of freedom, but stop pretending that isn’t the reason. Plus answer the question – do you believe that being able to edit and redistribute ebooks is q fundamental right?