CUDA Boot Camp

THE CUDA stuff should be installed on each lab machine in Fisher 
330/331.  

The first thing you are going to do is get a copy of the CUDA 
samples in  the Software Development Kit (SDK) and test the 
install.  Here is how to do this without using the command line.
If you understand "cp -R /usr/local/cuda/samples ." then 

A) Copying CUDA SDK Samples
A.1) If the machine is in Windows reboot it into Linux.
A.2) Right cick on the Computer Icon on your desktop and 
select "Browse Folder".
A.3) Select Filesystem, then usr, then local, then cuda, then 
drag the samples directory to your desktop.  

The last step will take a few seconds but you should now have a 
copy of the samples on your desktop. 

B) Compiliing the samples
B.1) Right click on the samples folder on your desktop.
B.2) Go down to "Open in Terminal"
B.3) Type "make" and hit return
B.4) Lots of stuff should start appearing in 
the terminal window. The computer is compiling 
all of the samples. It will take several minutes to 
finish.  Go look at http://xkcd.com/303/ whiel you wait. 
Hit random a few times on xkcd to kill a few minutes untill
it is finished.  There should be nothing saying "Error".  

C) Testing the install after it has compiled.
C.1) Right click on the samples folder on your desktop.
C.2) Go down to "Open in Terminal"
C.3) Type "./bin/linux/release/deviceQuery" and hit enter.
     You do not actually need to type the whole thing
     C.3.1) Type "./b" and hit tab (completion) three times 
     to get "./bin/linux/release/".
     C.3.1) Add "de" to get "./bin/linux/release/de" and hit 
     tab again to get the whole thing
C.4) You should get a report on the Devices installed on that machine. 
If it says non CUDA enabled devices or gives any warning try a
different machine.

D) Benchmarking the GPU in the machine. 
D.1) Right click on the samples folder on your desktop.
D.2) Go down to "Open in Terminal"
D.3) Type "./bin/linux/release/matrixMul" in the window and hit return.
     You should get a report on the peak flop count in GFlops/s.
D.5) Type "./bin/linux/release/MC_EstimateP" in the window and hit return.
     You should get a report on performance.
     
E) Take a look at the CUDA C programming guide.  
E.1) Google it and it will appear. 
E.2) If you prefer it is in the install directory.  Typing 
     "acroread /usr/local/cuda/samples/doc/CUDA_C_Programming_Guide.pdf" 
     will pop it up in acrobat.  
E.3) See if you can skim the first 20 pages and then focus on Chapter 2
E.4) Generate some questions.