|
The Golden Energy Computing Organization (GECO) invites CSM students to apply for an account on the Mines Nvidia Tesla S1070 system CUDA1.MINES.EDU Nvidia Tesla S1070 system CUDA1.MINES.EDUSpecial purpose computing accelerators, such as graphic processing units, (GPU) have become an important platform for parallel high performance scientific computing. GECO recently took delivery of and installed a Nvidia Tesla S1070 system. This 1U rack mountable system contains 4 of the Nvidia Quadro FX 5600 GPU cards and has a peak computing performance of 4 trillion floating point operations per second, or 4 Teraflops. This new Tesla system is front-ended by one of the RA nodes. Each of the 4 graphics processing units (GPU) on the Tesla has 240 processing cores and 4 Gbytes of memory for a total of 960 cores and 16 Gbytes. The individual GPUs are connected to the front-end node via a PCI connector. |
The GECO Tesla node is connected to one of the RA nodes via two PCI interface cables. In particular, it is connected node cuda1. To access the Tesla you must first ssh to ra.mines.edu then ssh to cuda1. For example:
[victoire:~] floydl% ssh ra.mines.edu Last login: Tue Jul 14 10:07:12 2009 from victoire.mines.edu Rocks Frontend Node - ra Cluster ... ... ... [floydl@ra ~]$ ssh cuda1 Last login: Tue Jul 14 10:06:34 2009 from ra.mines.edu [floydl@cuda1 ~]$
You will want to add the following lines to you .bash_profile file on cuda1 to enable you to access the cuda compilers.
export LD_LIBRARY_PATH=/usr/local/cuda/lib:/usr/local/cuda/cudaprof/bin:$LD_LIBRARY_PATH export PATH=$PATH:/usr/local/cuda/bin export PATH=$PATH:$HOME/cuda/bin/linux/release export MANPATH=$MANPATH:/usr/local/cuda/man
Next copy the example programs to your home directory.
[floydl@cuda1 ~]$ cp -r /usr/local/cuda_sdk cuda
Finally, log out and log back in to set your path correctly. After you log back in go to your cuda directory and type make. This will build the cuda example programs. The examples will be in the $HOME/cuda/bin/linux/release directory.
[floydl@cuda1 ~]$ cd cuda [floydl@cuda1 cuda]$ ls bin common doc lib Makefile projects releaseNotesData ReleaseNotes.html tools [floydl@cuda1 cuda]$ make make[1]: Entering directory `/home/floydl/cuda/common' a - obj/release/bank_checker.cpp.o a - obj/release/cmd_arg_reader.cpp.o a - obj/release/cutil.cpp.o a - obj/release/stopwatch.cpp.o a - obj/release/stopwatch_linux.cpp.o ... ... [floydl@cuda1 ~]$ cd $HOME/cuda/bin/linux/release [floydl@cuda1 release]$ ls alignedTypes dwtHaar1D nbody simpleStreams asyncAPI dxtc oceanFFT simpleTemplates bandwidthTest eigenvalues particles simpleTexture ... ... deviceQueryDrv MonteCarloMultiGPU simpleMultiGPU volumeRender [floydl@cuda1 release]$
The Tesla board is designed primarily for computation and does not support video out. It you try to run any of the examples that produce video out you will get an error.
For questions please contact Dr. Timothy Kaiser at tkaiser@mines.edu