University of California at Berkeley EECS Instructional Support Group /share/b/pub/cuda.help /share/b/pub/caffe.help Mar 10 2017 CONTENTS: Cuda, Caffe Cuda and GPUs in Instructional Labs Cuda "nsight" Development Platform Cuda, Caffe ----------- Cuda (http://www.nvidia.com/cuda) is is a parallel computing platform and programming model invented by NVIDIA. Caffe (http://caffe.berkeleyvision.org/) is a deep learning framework made with expression, speed, and modularity in mind. It is developed by the Berkeley Vision and Learning Center (http://bvlc.eecs.berkeley.edu/). Cuda and GPUs in Instructional Labs ----------------------------------- Workstations in 330 and 349 Soda have gpus and Cuda software support. Workstations in 349 Soda have Cuda and Caffe in /usr/local/cuda and /opt/caffe. Caffe biniaries and libraries are installed in /opt/caffe/distribute/bin and /opt/caffe/distribute/lib . Caffe users should add the related locations to their PATH, LD_LIBRARY_PATH and PYTHON_PATH environment variables. If you are doing your project with simple Makefiles rather than an "nsight" (Eclipse) based projects, you would just type export PATH=/usr/local/cuda/bin:$PATH to put the "nvcc" compiler & related programs onto your path first. Cuda "nsight" Development Platform ---------------------------------- Nsight (http://www.nvidia.com/object/nsight.html) is an Eclipse-based Cuda deverlopment environment from NVidia. You can run it on the Instructional Linux systems in 330 and 349 Soda with the command /usr/local/cuda/bin/nsight To load the Cuda 8 samples, select menu item "File > New Cuda C++ project", then choose "Import Cuda Sample..." as the executable type. You can copy a fresh project from the samples. It will auto-detect the card's "CUDA Capability" (3.0 for those cards). After building & running the "DeviceQuery" sample project, this is some of the output: Device 0: "GeForce GT 740" CUDA Driver Version: 8.0 CUDA Capability Major/Minor version number: 3.0 Total amount of global memory: 979 MBytes (1026424832 bytes) ( 2) Multiprocessors, (192) CUDA Cores/MP: 384 CUDA Cores GPU Max Clock rate: 993 MHz (0.99 GHz) ... ... ... EECS Instructional Support 378/386 Cory, 333 Soda inst@eecs.berkeley.edu