“And though she be but little, she is fierce.” – Shakespeare, on receiving an NVIDIA Jetson Nano for his birthday
The goal of
edgyR is to give R (and Julia!) users access to NVIDIA Jetson development kits. Although the target market of the Jetson platform is autonomous vehicles, and the focus of the software is image and video processing, the hardware and software are general enough for a wide variety of scientific computing applications.
The initial release is aimed at R package developers. I’m assuming a knowledge of R, RStudio, Git, and R Markdown. The release contains everything an R package developer should need to develop packages that interface with the unique hardware of the Jetson platform. It also contains a JupyterLab server for Julia, Python and R users more familiar with that environment.
The Jetson hardware consists of a multi-core
arm64 CPU and a CUDA-compatible GPU. The smallest of the family, the Jetson Nano, has four CPU cores, 4 GB of RAM, and a GPU with 128 CUDA cores. The peak speed is 472 billion floating point operations per second (GFLOPS).
The largest, the Jetson AGX Xavier, has eight CPU cores, 32 GB of RAM, and a GPU with 64 tensor cores and 512 CUDA cores. The peak speed of the AGX Xavier is 32 trillion operations per second.
The Jetson software is called JetPack. JetPack includes
edgyR adds R 4.0.2 and RStudio Server 1.3.1056 to JetPack, compiled for
arm64 from upstream source and packaged on a Docker image,
edgyr-ml. The image is based on Machine Learning for Jetson/L4T. As a result,
edgyr-ml includes numpy, scipy, matplotlib, pandas, scikit-learn, onnx, PyTorch, torchvision, torchaudio, TensorFlow, and JupyterLab.
There are (at least) five options:
It is not pronounced like any brand of peanut butter. Otherwise, it’s your choice!