Here is a short description of my situation – I need a portable machine for CUDA and iOS development, so any of the newer MacBook Pros with NVIDIA GPUs seemed like a no-brainer at first. However, the CUDA-requiring project already has a huge code base from a large number of contributors that is developed on CentOS 5.x/6.x. CentOS is also the only platform that is officially supported.
I obviously need a Mac because of the iOS project, but is MBP the right machine for the CentOS+CUDA based one? Did anyone get this to work properly, or is buying two separate machines (MacBook Pro + a PC laptop) the only solution?
P.S. I am mostly worried about the GPU-switching support between MacBook Pro's cards on Linux, so any first-hand experience from existing MBP owners is very welcome.
Best Answer
I don't think you'll be able to do this. I tried RHEL and it locked up on load. Same results on Fedora 16. I did successfully install Fedora 17 but was unable to get dynamic video card switching to work. Nor was I able to get the Nvidia card to work. I believe you have to run dual X servers and map the second card to your DFP out based on other posts. Additionally, heat and power consumption are increased since it never powers off the NVidia card. I did a lot of playing with Bumblebee and switching configurations for the Nouveau driver since the Nvidia driver yields a black screen. However, it wasn't switching properly. I believe the best I ever had it working was to run on the Intel card with hardware acceleration using the Nouveau driver. I would not recommend it as a CUDA platform at this time. There is a reasonable amount of contrib energy focused in this direction though and I don't think it will be too long before it is working. The Linux kernel has had support related to this being implemented as well.