3 ms·

As a Hackintosh user, I'd recommend an NCASE M1 v5 or the Shuttle SZ170R8 for a significantly better chassis. With the NCASE, you can get a LGA 2011-v3 mini-IT

As a Hackintosh user, I'd recommend an NCASE M1 v5 or the Shuttle SZ170R8 for a significantly better chassis. With the NCASE, you can get a LGA 2011-v3 mini-ITX board and get as fast of a computer as you want. Though truthfully, I don't think there's much of a point. Likewise, an old NVIDIA card for a Hackintosh? Also not much of a point. The web drivers are so bad. Who knows when Apple will ship another NVIDIA chip in a Mac? If you need "CUDA stuff," use Linux or Windows. Software like Octane is so buggy and suffers from worse performance on Mac anyway. Final Cut and After Effects both support OpenCL acceleration. Besides, the RX 480 is $189. If you're doing "VR stuff," well pity on you if you're developing for an Oculus on a Mac. The Vive doesn't, and probably never will, support Mac. Whatever VR solution Apple releases will obviously run great on the video cards they support, so that again strongly points to purchasing an AMD card at this time over your alternatives. With regards to this specific build, a high DPI display will greatly improve the enjoyment of this computer. The Dell P2715Q is the best deal. Mac OS has such good HiDPI support compared to Windows (and especially Linux). Enjoy the features you pay for! Truthfully, I'm hard pressed to see the point of a Hackintosh, and I own one.

atemerev 10y ago
CUDA stuff is awful on Windows (try setting up Tensorflow/Keras there — nothing but overwhelming frustration). So I have to use Ubuntu and miss my happy days with macOS.
- 21 10y ago
  I think that's exaggerating it a bit. I managed to install Tensorflow with CUDA on Windows just by following theirs official installation guides. I also managed to compile Tensorflow with GPU support from source, also by just following the official guide. Sure, it's a bit more work than "sudo apt install".
  - atemerev 10y ago
    It was 5 months ago; perhaps things happened to change since then. :)
    - aseipp 10y ago
      I installed Tensorflow 1.0.1 for Windows for the first time like, 3 weeks ago. I've never installed it before on any system and I never even had Python3 on this machine. I had to use Windows because it has my new GTX 1080. Short story: worked great, running in about 5 minutes. (With VSCode, I even get autocomplete out-of-the-box on the TF libraries, imports, etc.) I think it was much different before 1.0. Also, building on Windows is a little more difficult. But just installing it really isn't. And Keras was far more annoying because it wants scipy, which wants BLAS, and I don't use Anaconda. I haven't gotten around to that one yet...
      - sorenjan 10y ago
        > And Keras was far more annoying because it wants scipy, which wants BLAS, and I don't use Anaconda. I haven't gotten around to that one yet... Christoph Gohlke to the rescue: http://www.lfd.uci.edu/~gohlke/pythonlibs/#scipy http://www.lfd.uci.edu/~gohlke/pythonlibs/#scipy I also don't think that the numpy installed via pip is using mkl, so his version of numpy should be faster too: http://www.lfd.uci.edu/~gohlke/pythonlibs/#numpy http://www.lfd.uci.edu/~gohlke/pythonlibs/#numpy
atemerev 10y ago
What also gets me mad is "oh, so you need CUDA? You are a very niche user then, and ought to suffer!" It's when everybody and his dog are into ML, and having a decent GPU is a must unless you want to pay Amazon $2.5 per hour to train your models...
itg 10y ago
a GTX 980 Ti is considered old now? Unless you're working with datasets where the 6GB is limiting, it is quite capable for both CUDA programming & gaming. I tried CUDA programming on Win10 and it is an exercise in frustration, and find macOS a more pleasant environment compared to linux in part b/c of software library and it's what I'm personally used to.
- nottorp 10y ago
  Problem is, you can either spend the same and do more, or spend less and do the same, if you have support for a newer generation of cards. Not to mention that if you're into silent computers (not being able to hear the fans when gaming because of the game noise doesn't qualify as silent in my book) newer nvidia cards are much better.
- Cacti 10y ago
  It's quite capable, but if you need more than 6GB, the other capabilities are essentially irrelevant. Lack of support for the newer NVIDIA cards is increasingly becoming a major issue for many people.
matt4077 10y ago
I have the GTX980 and the NVIDIA driver for macOS works fine. It's just a pip install tensorflow-gpu nowadays.

atemerev 10y ago

CUDA stuff is awful on Windows (try setting up Tensorflow/Keras there — nothing but overwhelming frustration). So I have to use Ubuntu and miss my happy days with macOS.

21 10y ago
I think that's exaggerating it a bit. I managed to install Tensorflow with CUDA on Windows just by following theirs official installation guides. I also managed to compile Tensorflow with GPU support from source, also by just following the official guide. Sure, it's a bit more work than "sudo apt install".
- atemerev 10y ago
  It was 5 months ago; perhaps things happened to change since then. :)
  - aseipp 10y ago
    I installed Tensorflow 1.0.1 for Windows for the first time like, 3 weeks ago. I've never installed it before on any system and I never even had Python3 on this machine. I had to use Windows because it has my new GTX 1080. Short story: worked great, running in about 5 minutes. (With VSCode, I even get autocomplete out-of-the-box on the TF libraries, imports, etc.) I think it was much different before 1.0. Also, building on Windows is a little more difficult. But just installing it really isn't. And Keras was far more annoying because it wants scipy, which wants BLAS, and I don't use Anaconda. I haven't gotten around to that one yet...
    - sorenjan 10y ago
      > And Keras was far more annoying because it wants scipy, which wants BLAS, and I don't use Anaconda. I haven't gotten around to that one yet... Christoph Gohlke to the rescue: http://www.lfd.uci.edu/~gohlke/pythonlibs/#scipy http://www.lfd.uci.edu/~gohlke/pythonlibs/#scipy I also don't think that the numpy installed via pip is using mkl, so his version of numpy should be faster too: http://www.lfd.uci.edu/~gohlke/pythonlibs/#numpy http://www.lfd.uci.edu/~gohlke/pythonlibs/#numpy

atemerev 10y ago

What also gets me mad is "oh, so you need CUDA? You are a very niche user then, and ought to suffer!" It's when everybody and his dog are into ML, and having a decent GPU is a must unless you want to pay Amazon $2.5 per hour to train your models...

itg 10y ago

a GTX 980 Ti is considered old now? Unless you're working with datasets where the 6GB is limiting, it is quite capable for both CUDA programming & gaming. I tried CUDA programming on Win10 and it is an exercise in frustration, and find macOS a more pleasant environment compared to linux in part b/c of software library and it's what I'm personally used to.

nottorp 10y ago
Problem is, you can either spend the same and do more, or spend less and do the same, if you have support for a newer generation of cards. Not to mention that if you're into silent computers (not being able to hear the fans when gaming because of the game noise doesn't qualify as silent in my book) newer nvidia cards are much better.
Cacti 10y ago
It's quite capable, but if you need more than 6GB, the other capabilities are essentially irrelevant. Lack of support for the newer NVIDIA cards is increasingly becoming a major issue for many people.

matt4077 10y ago

I have the GTX980 and the NVIDIA driver for macOS works fine. It's just a pip install tensorflow-gpu nowadays.