The rCUDA Team is happy to announce the new release of the rCUDA remote GPU virtualization middleware. This new release is a beta version with lots of new functionality although performance is not optimized yet. The new release includes, among other characteristics, a new architecture of the rCUDA server, which has been renewed from scratch in order to provide full support to multi-threaded applications. Additionally, the rCUDA daemon now makes a better use of the GPU memory, allowing more memory for applications. Support for Deep Learning frameworks such as TensorFlow (v1.2 and v1.3) or Caffe has been introduced. Moreover, rendering programs such as Blender are now supported. New libraries, such as cuSOLVER, cuBLAS-XT and NVGRAPH, are also supported, whereas partial support is provided for the NPP library.