Currently the Math Dept. has a number of multiprocessor systems available for Math faculty and graduate students to run compute jobs on. Available software includes Matlab, Mathematica, Maple, and TecPlot. An MPI environment for parallel programming is also available. Your Math Dept PID and password is used to access these systems.
- Shell server - ssh.math.vt.edu
- Mavramorn - Graduate workstation in McBryde 475
- Undergrad workstation in McBryde 334
- Tesla - powerful computation server with NVIDIA cuda
- Gauss - more powerful compuation server with NVIDIA cuda
- Matlab help
A virtual server that is available for reading email, accessing personal files and web pages, program compilation, and other command line tasks. This is accessible via ssh or sftp, for file transfer purposes.
Please don't run any large or long computational intensive tasks on ssh.math. These should be run on the cluster, tesla, or mavramorn.
A workstation located in McBryde 475 that can be used interactively with a visual display. It has an Intel quad-core Xeon 2.66GHz processor with 4 Gig RAM.
We ask that a limited number of jobs be run on mavramorn concurrently to keep the load on the system manageable.
Running your compute jobs on mavramornnice -n 19 nohup command >& run.out &
This will run a command sending the output to a file called "run.out". You will be able to logout from your terminal (ssh). The command will run with reduced priority, which means interactive users on the system will not be impacted .
A workstation located in McBryde 334 that can be used by undergraduates interactively with at visual display. It has a dual-core Intel 3.0 GHz processor with 2 Gig RAM. A virtual Windows XP environment is also available.
Our newest computation server that utilizes NVIDA cuda technology and high speed SSD storage. It has 4 Intel 15 core 2.8 GHz processors, 1 TB of RAM, and 3 NVIDIA Telsa K40 GPU cards. There is also 11 TB of local SSD storage available. Most of this space is currently unallocated. Please see Bill Reilly about setting up disk space for your computation work.
A computational server that utilizes NVIDA cuda technology. It has 2 Intel 6-core Xeon 3.33 GHz processors, 48 Gig RAM, and 4 NVIDIA Telsa C2050 GPU cards. The CUDA Toolkit 4.1.21 is installed as well as our normal set of compilers. Please see the following NVIDIA documents for more information on CUDA developement.
We also have licenses for Jacket from Accelereyes. This software helps with MATLAB acceleration with GPUs. Please see the Accelereyes web site for more information.
- Getting Started Guide
- Release Notes
- CUDA C Programming Guide
- CUDA C Best Practices Guide
- CUDA Reference Manual
- API Reference
Latest versions of these as well as other CUDA documents are available at the NVIDIA Developer site A copy of the NVIDIA SDK 3.1 is available on telsa in /opt/nvidia/gpucomputingsdk_3.1_linux.run. You will need to install it in your home directory by running the command sh /opt/nvidia/gpucomputingsdk_3.1_linux.run and following the instructions.
If you have matlab programs that will require large amounts of memory or will run for many hours, you would be better off running them on one of the Math Department servers. Your program can run uninterruped for days/weeks like this without tying up your laptop/desktop.
Typically you will want to run matlab using the nohup command. This will allow you to log off the server and keep your matlab job running. You will have matlab save any results that are needed as a .mat file which will be placed in the directory your matlab job is running from.
You may find that running Matlab with the nohup command can create very large output files that contain repeated lines of "Bad file descriptor" or "Error reading character from command line" messages. This can fill up the filesystem quickly.
A workaround for this problem is to run your matlab command like this:
nohup matlab -nodisplay -r MATLAB_FILE >& OUTPUT_FILE < /dev/null &
MATLAB_FILE is your matlab .m file without the .m
To kill off a program running like this you need to do the following:
ps -fu MATH_PID
You need to find the Process ID (PID) of the command you want to kill.
Then use the kill command to kill it off