Those numbers are plainly wrong, but even if they were correct, I wasn't talking about Linux users, but linux users that use only GPL stuff like Phoronix do. Not even all Linux distros are GPL-only, let alone the complete portfolio of Linux programs. That's why Phoronix is niche even inside the Linux community, which is a niche itself. So you are talking of a niche inside a niche.
What numbers? No, Phoronix do not use "only GPL stuff". In fact those benchmarks were run over a OS which includes propietary and closed source stuffs. For me their test suite is good. Softpedia even call it the best benchmark suite.
Why do you spend all this effort against one fair benchmark suite, whereas (as other poster noted before) you ignore completely biased benchmarks such as sysmark?
The same biased logic regarding your niche argument. It is not true, but moreover, you don't apply it to other stuff. For instance, about less than 5% of gaming PCs has a GPU with 2GB VRAM, less than 1% have 3GB and about 0.5% have 4GB.
The number of users with a GTX Titan card (6GB) is a niche inside a niche (gaming PC). By your own logic no review site would be testing niche cards like the Titan, neither using them in gaming benchmarks. Which is nonsense.
The only people who care about this fact are fanboys. The rest of us just buy what we need and offers the best perf/$.
Agree with you on that ordinary people don't care about brands.
Regarding fanboys. The other day someone tried to convince me that Intel > AMD because a $600 chip was about a 70% faster than a $140 APU. And in the graphics card forum I hear often people claiming that the Xbox One ($400 console) is **** because they can beat it with two GTX-680 in SLI conf. in a i5-3570k.
Well galego does have a point about Intel Complier.
Agner was barking about this problem for years now.
http://www.agner.org/optimize/blog/read.php?i=49#269
And he made a mathematical librarie so you can compare FX chips to Xeons.
http://www.yeppp.info/home/yeppp-performance-numbers
Yes, his research was basic to demonstrate that Intel cheats both compilers and benchmarks. Agner provides some examples of biased benchmarks that use the Cripple_AMD function.
I always ask the same question and nobody answers it for me in a convincing form: If intel had so good products as some believe, because does it need to cheat benchmarks such as sysmark, cinebench... before confronting to AMD, VIA, or Nvidia?