Triad

Description: a version of the stream triad benchmark, measures bandwidth for a large vector dot product operation. This test includes PCIe transfer time and is designed to test the maximum sustained throughput for devices. With current hardware, this throughput is often bottlenecked at the PCIe bus. Triad uses cuda streams and mutliple OpenCL queues to overlap communication and computation.

Problem Sizes: (MB) - always 64KB - 16 MB (doesn't use -s sizing)

Precision: Single