This page introduces softwares and services which are installed into TSUBAME2.5 experimentally. As they are experimental services, no official supports are provided.
If you have questions or requests, please use this form.
If the software has link to detailed explanation page, please refer that page.
If no such page or instructions are provided, execute following command to put necessary path and library path into your environment.
$ source /usr/apps.sp3/nosupport/gsic/env/(software name)-(version).sh
Details (In Japanese, Created: 2014/12/12)
Profiler and traces are used to investigate application performance bottlenecks.
Details (In Japanese)
GXP makes massively parallel execution of array jobs more efficiently
Details (PDF in Japanese, Created: 2011/04/01)
BLCR(Berkeley Lab Checkpoint/Restart) is available in S and H queue nodes. Currently, BLCR cannot be used in MPI jobs because its installation does not work with MPI library. Please check with your environment before use, even though we don't find any problem at this moment.
GCC 5.1 in installed with enabling OpenACC experimental supoprt
This version is newer that that is installed in TSUBAME2 default environment, and some HPC packages such as PyCUDA are installed.
If you can't find necessary package, or package installed by GSIC is too old, please try this command after setting up external access:
$ pip install --user <packagename>
CMake is required to build application which has CMakeLists.txt in its archive.
This version is newer that that is installed in TSUBAME2 default environment.
Following deep learning frameworks are available
Caffe(Reference): can be used from Python, C++ etc.
Theano(Reference): version 0.8.2 is available via python-2.7.7.sh or python-3.5.2.sh
Chainer(Reference): version 1.16.0 is available via python-2.7.7.sh or python-3.5.2.sh
Torch(Reference): master branch as of 2017/03/07 (yyyy-mm-dd) is available
Computer vision library
A library which collects performance counter values in CPU and GPU
MUST analyses MPI function usage and detects incorrect invocation of MPI APIs.
An application analysis tool that identifies loop nest structures during actual execution and data dependences among them.
Details (In Japanese)