The NUPAR benchmark suite targets the modern GPU architectures and tries to utilize the cutting edge features introduced in CUDA and OpenCL.
The benchmarks concentrate on: Dynamic parallelism Concurrent kernel execution Efficient C++11 Atomics Advanced intrinsic support ( CUDA SHFFLE instruction) Shared Virtual Memory (Coarse Grain OpenCL 2.0).
In Development: Shared Virtual Memory (Fine grain OpenCL 2.0) Unified Memory Pipes.
New Revision Due: March 2015. Stay Tuned.