Main Page

From MathCUDA

Jump to: navigation, search

Contents

[edit] ParaCUDA Wiki

  • This wiki contains information on using Nvidia CUDA for technical computation.
  • The MTU Center for Experimental Computation (CEC) and Math Sciences department are setting up a prototype CUDA enabled cluster called paracuda.cec.mtu.edu.
  • The Math department is running a workshop to develop and share expertise with these emerging computational tools.

We meet Thursdays from 10-11am in Fisher 330. Users of all levels of expertise are welcome.

  • Please edit the wiki. Here is a brief wiki Reference Card.
  • The preformated tag is useful for code e.g. 
 struct vector *p;

[edit] ParaCUDA Access

ParaCUDA is a small-scale shared resource intended for developing experience and parallel program prototyping. The seminar and wiki are the primary communication tools.

To access ParaCUDA you must:

  1. Provide a brief description of the background and code organization of your project. This includes
    1. A brief write up on the Wiki
    2. A brief presentation to the seminar.
    3. Pseudo code describing the planned CUDA parallelization i.e. the dimensioning calls <<< >>>
    4. Details on anticipated memory and processor requirements.
    5. Details on anticipated speed-up.
  2. Computational use will be restricted to 30 minutes of runtime.
  3. Share your ParaCUDA experiences with the seminar and Wiki. This includes
    1. A brief write up on the Wiki
    2. A brief presentation to the seminar
    3. Details on memory/processor use
    4. Details on any realized speed-up.
    5. Details on future plans.
  4. Agree to acknowledge ParaCUDA development where appropriate.

Contact A. Struthers for information.

[edit] News!

  • 05/7/09 - The R system has been installed across the cluster. No extra libraries have been installed. Please talk to Allan/Dave about installing R packages.
  • 04/27/09 - CUDA profiler (cudaprof) is installed and works via qrsh. Remember to use ssh -X when you log into the cluster to export your $DISPLAY correctly.
  • 03/27/09 - paracuda now accepts ISO passwords. Logins must be granted as above, but ISO passwords will now work, rather than using your M number or another password.

[edit] Getting started CUDA

[edit] Getting started ROCKS

[edit] Tools and Guides

[edit] Bits of code

[edit] Projects

[edit] Agendas

[edit] F09 Agendas

  1. Tuesday September 8th. Welcome and CUDA BootCamp
  2. Thursday September 17th. Continuation of CUDA BootCamp
  3. Thursday September 24th. Continuation of CUDA BootCamp
  4. Thursday October 1. CUDA Profiling
  5. Thursday October 8. Quick review of Sparse Solvers and Performance (by Bill) See: General-Puspose Sparse Matrix Building Blocks using the NVIDIA CUDA Technology Platform
  6. Thursday October 22.
    1. Brief review of linking Fortran CPU to CUDA GPU code. (Atikan)
    2. Beam Buckling Parameter Study (A CUDA code development case study), Part I (Allan)
    3. Progress on CUDA compile in R (Allan)
  7. Thursday Nov 5th.
    1. Parallel Optimization Progress (Andrew)
    2. Beam Buckling Parameter Study (A CUDA code development case study), Part II (Allan)
    3. Parallel Diffusion Simulation Progress (Beini)
    4. Discussion
  8. Thursday Nov 19th
    1. Installing your own CUDA card (Dave K.)
    2. Synchronicity : Asynchronous calls, Synchronization and streams. (Allan S.)
    3. CUDA optimization (Andrew A.)

[edit] Old Agendas

[edit] People

[edit] Organization

Retrieved from "http://www.mathlab.mtu.edu/CUDA/index.php/Main_Page"
Personal tools