Ubuntu 10.10 vs. Mac OS X 10.6.5: A Competitive Race

[phoronix]

Phoronix: Ubuntu 10.10 vs. Mac OS X 10.6.5: A Competitive Race

Back in May, we published benchmarks of Mac OS X 10.6.3 vs. Windows 7 vs. Ubuntu 10.04, along with other times, looking at the performance of Apple's Mac OS X operating system relative to Ubuntu and other Linux distributions. In most of those articles though we used Mac Mini computers, but now with a new Mac Book Pro in our labs with the latest Apple/Intel hardware, along with the most recent versions of each operating system, we have carried out a new set of tests that is also more in-depth than our earlier published benchmarks.

http://www.phoronix.com/vr.php?view=15530

[Tgui]

The disparities in the "SmallPT GPU" test are interesting. Perhaps the OpenCL-GPU compilers for the tested operating systems differ in their abilities to optimize and the Linux one is generating less efficient GPU code? Maybe context switches or data transfer between the GPU and CPU on Linux are more expensive/hindered?

I'd think both operating systems would generate the same GPU binaries given the compiler is provided by NVIDIA?

My experience is CUDA though.

[Plombo]

I'd think both operating systems would generate the same GPU binaries given the compiler is provided by NVIDIA?

I think Apple provides its own OpenCL implementation. (And since Apple actually pioneered OpenCL, it's believable that they've put more of an investment into optimizing their implementation than Nvidia has.)

[Tgui]

I think Apple provides its own OpenCL implementation. (And since Apple actually pioneered OpenCL, it's believable that they've put more of an investment into optimizing their implementation than Nvidia has.)

That would make sense since they're the ones actively using OpenCL in their GUI and whatnot.

Good point.

[deanjo]

I think Apple provides its own OpenCL implementation. (And since Apple actually pioneered OpenCL, it's believable that they've put more of an investment into optimizing their implementation than Nvidia has.)

I believe something else is happening here but the only way to confirm it would be for the tests to be reran utilizing FreeBSD for comparison as well.

[thefirstm]

Was 32-bit or 64-bit Ubuntu used?

[mtippett]

The disparities in the "SmallPT GPU" test are interesting. Perhaps the OpenCL-GPU compilers for the tested operating systems differ in their abilities to optimize and the Linux one is generating less efficient GPU code? Maybe context switches or data transfer between the GPU and CPU on Linux are more expensive/hindered?

I'd think both operating systems would generate the same GPU binaries given the compiler is provided by NVIDIA?

My experience is CUDA though.

OpenGL and OpenCL are handled differently between Linux and OSX. The vendor provides the whole pipeline for Linux. For OSX, Apple provides the pipeline, and the vendor provides one or more of the sub-layers. So if Apple hasn't enabled GPU accel for one of the paths, the vendor won't even see it. That might be the case where there is a large delta between the two.

[Kano]

I would use a nvidia driver update.

http://www.nvidia.com/object/quadro-...03-driver.html

This driver even officially supports the used gfx card.

[BlackStar]

I would use a nvidia driver update.

http://www.nvidia.com/object/quadro-...03-driver.html

This driver even officially supports the used gfx card.

This won't change anything, as the OpenGL "state tracker", if you will, is handled by Apple not Nvidia. Nvidia only controls the bottom end of the stack - which is why Mac OS X is still limited to OpenGL 2.1 and performance is so low.

A Linux/Gallium3d vs Mac OS X comparison would actually be more interesting.

[timwu]

I have installed both OSes on my ASUS 1201N. For daily operations like Internet browsing with google chrome, ubuntu 10.10 out performs a lot than snow leopard 10.6.5.