The following are steps to setup CUDA on Windows and test run the CUDA exercises. It will describe the setup process for 64-bit systems.
All instructions in this post will list directories for default install locations. It is strongly recommended that you do not deviate from the default settings.
NVIDIA's documentation for installing CUDA on windows can be found at the link below.
http://developer.download.nvidia.com/compute/cuda/5_5/rel/docs/CUDA_Getting_Started_Windows.pdf

0. Before you start installing CUDA, make sure you have the following. All Microsoft software can be fetched from MSDN/Dreamspark website.
	Windows x64 (of course!)
	Visual Studio (2012 preferred, 2010 will work as well)

1. Download and install the latest NVIDIA GPU drivers for your device. 
	Use the link: www.nvidia.com/Drivers. 
	It is suggested that you download and install the latest stable drivers. However, you may experiment by downloading beta drivers. NVIDIA's beta drivers are quite stable if you want to give it a shot.

2. Download and install the CUDA toolkit.
	Use the link: https://developer.nvidia.com/cuda-downloads for the latest version.
	If you wish to have previous versions installed, make sure you install the older versions before installing the newer versions.
	The older versions of CUDA toolkit can be found at: https://developer.nvidia.com/cuda-toolkit-archive
	Make sure you have the appropriate CUDA* environment variable set. To check this go to Control Panel -> System -> Advance System Settings -> Advance (tab) -> Environment Variables. Also check the PATH variable to see whether the CUDA path is present.
	Once you have completed installation, run the CUDA samples that were installed to make sure everything is running well. Samples are located at C:\ProgramData\NVIDIA Corporation\CUDA Samples\v5.5 (you may replace v5.5 with the version you have installed). 

3. Download and install NVIDIA Nsight for Visual Studio 3.1 (or latest version)
	To download Nsight, you will need to be a registered user.
	The main page for NSight is https://developer.nvidia.com/nvidia-nsight-visual-studio-edition.
	Follow the appropriate links to register. You will have to fill up a questionnaire. Fill it appropriately. An incomplete or bad application is very very likely to be rejected.
	It may take a few days for your application to be approved. So do not leave this step to last minute.
	Once you have been approved, download and install it.
	If by chance you are not approved, you can ask one of you friends to share the downloaded file. But do work on getting approved.
	You will need Administrator access to run NSight. So if you are using a computer (such as in the lab) to code where you do not have admin access, try getting it. Request Patrick to arrange this if you need it.

4. It is recommended that you copy the NV Tools Ext binaries to system directories. Copy
	C:\Program Files\NVIDIA GPU Computing Toolkit\nvToolsExt\bin\x64\*.dll to C:\Windows\System32\
	C:\Program Files\NVIDIA GPU Computing Toolkit\nvToolsExt\bin\Win32\*.dll to C:\Windows\SysWow64\
	Alternatively, you can copy these files to the executable folders of your projects
	Once you have installed all these, start Visual Studio and create a CUDA project. By default, the code created should run out of the box. Run it and make sure everything is working well. Run NSight as well.

The following is specific to the special session to be held on Sept 25 2013.
Running the CUDA Exercises:
	If you are using Visual Studio 2012, use the CUDA_Exercise.sln (uses CUDA 5.5) and for 2012 use CUDA_Exercise_2010.sln (uses CUDA 5.0). 
	Both solutions have been preconfigured appropriately, but feel free to change things around. One thing you may need to change is the compute version.
	Go to Project Properties -> CUDA C/C++ -> Device -> Code Generation. Enter compute_##,sm_##. Where ## is 30 for Kepler, 20 for Fermi, 35 for K20/Titan/780, and 1x for pre-Fermi card. 
	You can check highest compute versions for your GPU https://developer.nvidia.com/cuda-gpus. The number is same as "compute compatibility" without the decimal point.
	If the CUDA version does not match, then right click on the project and choose "Build Customizations". Choose the CUDA version you would like to use.

Anyone wanting to code on Linux has my thumbs up. I have included a makefile for the CUDA Exercises to run on Linux. If you are interested, send me an email and I will personally assist you in installing and ensuring everything runs.

For any questions or trouble installing, you can email me at shehzan@accelereyes.com