This page shows libraries & packages that have been tested on Jetson TK1, mentioning whether they work or not and also discussing any known issues, etc.
- CUDA: CUDA GPGPU Compute Toolkit is included when you install JetPack, or can be installed manually by following the instructions). Versions before CUDA 6.0 have not been tested.
- cuDNN: CUDA Deep Neural Network library, with more info on the cuDNN page.
- OpenCV4Tegra: OpenCV4Tegra works well on Jetson TK1 including the CUDA-accelerated 'gpu' module (but not including the 'nonfree' module of patented code). More info about OpenCV4Tegra is on the Computer Vision Performance page. OpenCV4Tegra is included when you install JetPack, or can be installed manually by following the instructions. Other versions of open-source OpenCV or closed-source OpenCV4Tegra should also work.
- ArrayFire: ArrayFire is a high performance software library for parallel computing with an easy-to-use API. Its array based function set makes parallel programming simple. ArrayFire's multiple backends (CUDA, OpenCL and native CPU) make it platform independent and highly portable. ArrayFire is now Open Source and is available under a BSD 3-Clause license. The source code can be found on the ArrayFire GitHub Page.
- OpenCV: OpenCV 2.4.9 works fine when built from source, including the CUDA-accelerated 'gpu' module, as explained in the wiki instructions. Other versions probably also work fine. The Tegra CPU-optimized OpenCV4Tegra by NVIDIA also works, as mentioned above.