FastROCS TK 1.0.1¶
- First version of FastROCS that is designed to only work on a specific hardware and driver configuration. This release will only work on NVidia hardware with the CUDA 3.2 driver. FastROCS will work on any modern NVidia card that reports a compute capability of 1.0 or greater. Typically this is any NVidia card shipped after 2006. For a complete list of cards refer to Appendix A in the CUDA C Programming Guide <http://developer.download.nvidia.com/compute/cuda/3_2_prod/toolkit/docs/CUDA_C_Programming_Guide.pdf>.
The more stringent requirement is on the driver version, CUDA 3.2. For this reason FastROCS will be shipping with the required associated NVidia Linux driver embedded in the tar-ball: devdriver_3.2_linux_64_260.19.26.run
This requirement is also embedded into the tar-ball name as “OpenCL-1.0-CUDA-3.2” in case we need to support multiple versions or rev’ing to OpenCL 1.1 breaks compatibility. If a driver upgrade needs to occur due to some bug in some other third party software, let OpenEye know and we will see how best we can accommodate you.
- Added ShapeSimilarityMatrix.py example to write out a similarity matrix of an input database into a csv file suitable for feeding to downstream tools like cluster analysis. Note, due to the current slow speed of clCreateContext, the example is not crazy performant on small datasets, but on larger sets it is as the cost of clCreateContext becomes amortized.