A new version of the rCUDA middleware has been released. This version, known as rCUDA v16.11, provides support for CUDA 8.0 in addition to fix several bugs. It also supports P2P memory copies among GPUs located in different nodes of the cluster.

You can meet the developers of rCUDA during SC'16 next month at Salt Lake City, Utah. We will present rCUDA at Mellanox' booth from Nov 14 until Nov 17.

Prof. Jose Duato will give next week a talk about rCUDA in the context of the 5th HPC Advisory Council Spain Conference, which will be held in Leon on 21st September. The talk will present the recent developments within rCUDA as well as its future roadmap.

rCUDA will be presented at Mellanox' booth during next SC'16 at Salt Lake City, Utah, USA, in November. A new demo will be disclosed during SC'16. The demo will show the new features of rCUDA in the area of cloud computing.

The rCUDA Team is developing a plug-and-play scheduler to be used with rCUDA so that multiple virtual machines can automatically (and concurrently) share the GPUs in the system without running out of GPU memory.