You can also add -exclusive to reserve the complete node for your work. Use gpu:1 for one GPU and gpu:2 for both. Qlogin -account= -partition=accel -gres=gpu:2 To gain access use qlogin with GPU request. The GPU nodes are located in rack 19, named compute-19-1 through 16, og just c19-1 through 16. Portland Accelerated C/C++/Fortran compiler have support for CUDA. Here you'll find source code, documentation etc. The software is installed at $CUDA, try ls $CUDA

NVIDIA Cuda 5.0, 5.5, 6.0, 6.5, 7.0, 7.5 and 8.0 (the current default) are installed, module name cuda/X.Y, try module avail to see all choices. If you want more information load the pgi/14.4 module and issue the command : pgaccelinfo. With a measured performance of more than 1 Tflops/s for matrix multiplication using single precision data type. Dual Intel Xeon(R) CPU E5-2609 2.40GHz, aka Sandy Bridge with 64 GiB memory nodes each with two NVIDIA Tesla K20 (Kepler architecture) cards.Ģ688 single precision and 896 dual precision cores

Nvidia cuda driver