Hetero-mark, a benchmark suite for CPU-GPU collaborative computing

VComputeBench: A Vulkan Benchmark Suite for GPGPU on Mobile and Embedded GPUs

“…Benchmarks play an important role in exposing these kind of differences between hardware architectures, compilers and more importantly across competing programming models. There are several benchmarks available to evaluate CUDA and OpenCL [10], [11] [12] [13] but currently none for Vulkan. To fill this gap and enable our study we propose VComputeBench, a set of Vulkan compute benchmarks that help developers understand the differences in performance and portability of Vulkan and provide guidance to GPU architects in the design and optimization of their drivers and runtime.…”

High-Performance Computation of Bézier Surfaces on Parallel and Heterogeneous Platforms

“…These mechanisms range from distribution of the same processing stage among the computing units [19] to distribution of processing stages (from pipelines) based on optimal mapping to the most adequate computing unit [20,21]. To date, organization of new applications using HCS is an active area of research; in order to understand the challenges and opportunities for HCS for leveraging improved performance over traditional computing systems, benchmark suites like Hetero-Mark [22] and CHAI [23] have been recently developed.…”

Energy efficient execution of heterogeneous applications

“…As a result, CPU-GPU collaborative computing has recently attracted a lot of attention. Applications can possess a number of different collaboration patterns, as demonstrated in recent benchmark suites such as Hetero-Mark [6] and Chai [7]. The set of applications that are part of these suites highlight the performance potential of CPU-GPU collaborative computing.…”

“…Employing the computational power of the CPU, in parallel with the GPU, can improve the performance of a variety of applications. This computation model is becoming a target for a new class of heterogeneous workloads [6,7].…”

“…We perform this study, while running collaborative applications taken from the Chai [7] and Hetero-Mark [6] benchmark suites. What is challenging is that the TX1 has three different voltage-frequency domains, with the memory domain being shared between the CPU and GPU.…”