Wiki‎ > ‎

OpenEye Suite

posted Oct 25, 2012, 4:47 PM by Dong Xu   [ updated Apr 26, 2017, 5:09 PM ]
CUDA Implementation in FastROCS TK

Toolkit install/uninstall:

pip install OpenEye-toolkits-python3-linux-x64-2017.2.1.tar.gz
pip uninstall OpenEye_toolkits_python3_linux_x64-2017.2.1


pip uninstall OpenEye-toolkits-python2.7-ucs4-linux-x64
pip install OpenEye-toolkits-python2.7-ucs4-linux-x64-2017.6b1.tar.gz


The internal GPU implementation of FastROCS has been ported from OpenCL to CUDA. GPU computing with CUDA allows OpenEye to better support NVIDIA hardware and facilitates our ability to improve performance. The switch from OpenCL to CUDA does not, however, change the results returned by FastROCS and should not impact the deployment of FastROCS TK on customer machines.

Our benchmark test shows that the 2016.Oct CUDA-based/python3-based FastROCS is not as fast as OpenCL/Python2 based version. Molport database (6.4mil compounds, 45mil conformers), query: Zinc7455.mol2

Client    Server    Time
p2.7      p2.7         27sec
p3.4      p2.7         27sec
p3.4      p3.4         31.5sec
p2.7      p3.4         32sec

OpenEye is planning to phase out Python 2 support by the 2017.Oct release. We're moving to Conda/py34_env

In QUACPAC v1.7.0 Released,  SuSe 11 is no longer supported. We are able to install the Redhat7 version  QUACPAC-1.7.0.2-redhat-RHEL7-x64.tar.gz



All executables should go in 

openeye/arch/suse-SLES11-x64

because packages such as rocs or eon use "suse-SLED11-x64" dir, the files should be moved into "suse-SLES11-x64" dir

As for Omega, its start-up script "omega2" should be edited just under the line

version=`grep -i suse /etc/SuSE-release`

from osver="SLE$osver" to osver="SLES$osver"

===================

some packages (vida, fred) are extracted to "suse-SLE11-x64", it works fine, you can either move it to "suse-SLES11-x64" or leave it.


Comments