Building a CUDA pc - linux vs windows some quick questions

I am planning on building a computer for use with CUDA (or maybe openCL?) and I am a little bit confused. Last summer I wrote some stuff on CUDA on a mac, but we’re going to buy a real computer for a new project that requires a bit more computational horsepower.

For parts, I am intending to use this set of components: http://www.tomshardware.com/reviews/overcl…u-gpu,2643.html $2k is around the budget I have to spend.

The thing I am not sure of is the OS to use. I am leaning towards windows, as I am a fairly inexperienced linux user. I was also told that in Windows 7 I would need buy a third GPU for actual graphical output. Is this true? I can’t find it anywhere in the documentation, but I remember there being some limitations when you use windows. If I do need a third card, what card should I get? Also, it seems clear that windows has the better debugging environment. Is this correct?

Sorry for the barrage of questions, but any advice would be greatly appreciated. Thanks.

IMO Linux is better for CUDA development (I was never a big fan of Visual Studio as an editor or as a toolchain) and debugging (I prefer cuda-gdb to NSight).

IMO Linux is better for CUDA development (I was never a big fan of Visual Studio as an editor or as a toolchain) and debugging (I prefer cuda-gdb to NSight).

Interesting. I’ll definitely look into the linux options more.

Is the business about needing a third card in windows true?

Interesting. I’ll definitely look into the linux options more.

Is the business about needing a third card in windows true?

So in Win7, you will want two cards (preferably at least one Tesla so you can use the upcoming improved TCC driver). I don’t know why you’d need three.

So in Win7, you will want two cards (preferably at least one Tesla so you can use the upcoming improved TCC driver). I don’t know why you’d need three.

two cards for cuda, and one for rendering the GUI, or so I’m told. I was confused by this as well.

Why would I want a tesla card over the newer fermi ones?

two cards for cuda, and one for rendering the GUI, or so I’m told. I was confused by this as well.

Why would I want a tesla card over the newer fermi ones?

If you’re doing multi-GPU programming, then yes, you’d probably want three.

Tesla cards will be able to run the TCC driver alongside standard NVIDIA WDDM graphics drivers in an upcoming release.

If you’re doing multi-GPU programming, then yes, you’d probably want three.

Tesla cards will be able to run the TCC driver alongside standard NVIDIA WDDM graphics drivers in an upcoming release.