The following machine has been set up for labs. Please contact Jingfei to get access to the machine.
 
IP: 10.173.214.154 (on-campus access: ssh, off-campus access: VPN + ssh) see http://www.noc.ucf.edu/VPN/ for VPN setup
 
The machine is dual-boot with Windows XP and Fedora 8 installed.
CUDA toolkit and SDK (1.1), Nvidia 8800 GTX card (768M RAM) display driver are all installed.
 
Setup your own programming workspace:
In Linux:
Once you login,
"cd Desktop/CDA6938/CUDASDK" (This is the directory where the SDK has been installed.)
Create your own working directory, for example, "mkdir jingfei_projects" (everyone should create his/her own directory)
Under your own directly, you should create your sub-directory for each assigned homework and project.
Refer to "Desktop/CDA6938/CUDASDK/martin_cuda_projects" for reference. You may just copy martin's Makefile and change for your own work.
 
In Windows XP:
Once you login, you may find CUDASDK directory under CDA6938. You need to create your own project under CUDASDK/projects directory.
Just copy the "template" and change it to your own.
 
 
Note: since we share the same SDK directory and the same skeleton programming environment, please make your project name unique so that the generated executable will not mix in the same directory
(e.g. release, emurelease under CUDASDK/bin/linux).
 

--For laptops with CUDA-supported graphics card,
After installation of toolkit 1.1 and SDK 1.1 on Windows XP, the display driver installation may fail and report some error message like "device with cuda support not found"?
 
Solution: download a modified "nv4_disp.inf" file which supports more device models and replace the original one included in the display driver package.
Link: http://www.laptopvideo2go.com/infs/165series/16921/nv4_disp.inf for 169.21
 

Useful tips:

1. Generate ptx files:

In the directory where the C++ compiler resides (or the directory is included in PATH), in windows XP, the directory is something like “c:\program files\Microsoft Visual Studio 8\vc\bin\”, invoke nvcc in the CUDA directory like “c:\CUDA\bin\nvcc –ptx -o a.ptx “c:\my project\a.cu”. You can then examine the output ptx code: a.ptx.

2. Checking the number of registers

C:\UCDA\bin\ptxas –v a.ptx –o a.o