Question

1. I have a linux box wired up as my CUDA machine — it has 3 GPU cards in it.
2. I do all my development on an OSX laptop.

How should I set up my development cycle where I develop on my laptop and deploy to the LINUX box?

My IDE of choice is eclipse and as a backup I use vi.

Thanks.

Answer 1

If your Mac does indeed support CUDA, then you should be able to install CUDA on it and do most of your development work locally. You could use a version control system like Git or Mercurial to push your code to the Linux box for all your testing (in case you don’t want to run CUDA locally), but that will make it slow to test small code changes because you will have to commit and push all changes before they will be available on the Linux box.

Personally, I prefer to develop all of my code directly on the target machine, whenever possible. I open up a few SSH sessions and write all the code with Vim. This allows me to run full tests with very little turn-around time. The catch is that you must always have a good network connection to your test box, or else development will be a royal pain.

So, in summary, you have three options.

1. Write code on Mac, run all tests on Linux. Use Git or something to
   push changes. Testing small code changes will be a huge hassle.
2. Write code on Mac, run basic tests locally (you’ll have to install
   CUDA) and occasionally test against Linux. Testing will usually be
   efficient, but you may need to redo performance optimization for
   both platforms.
3. Write code on Linux, by opening SSH sessions from your Mac and
   writing in Vim or whatever you fancy. You need a consistent, good
   network connection with the Linux box. If you have a fantastic
   connection, you could conceivably run Eclipse from Linux using X
   window sharing.