This study investigates the performance of asynchronous actor programming and synchronous Partitioned Global Address Space (PGAS) versions of graph kernels on a Cloud platform and a HighPerformance Computing (HPC) platform. Using the Bale suite of graph microkernels, we compare the execution times of kernels implemented with OpenSHMEM (synchronous PGAS) and HClib-actor (asynchronous) on both Azure Cloud with Ethernet and an HPC cluster with InfiniBand. Our results reveal significant performance differences between these platforms. While the asynchronous version outperforms the synchronous version in both settings, the performance gap is dramatically wider on the Cloud platform, with the asynchronous version showing up to 1,000x improvement over the synchronous version in some cases. Moreover, we observe highly variable execution times in the Cloud, likely due to shared resource interference and unpredictable data center traffic. These findings highlight the importance of choosing appropriate programming models for different computational platforms, especially as Cloud platforms are becoming more affordable and easier to access compared to traditional HPC clusters. Our work provides valuable insights for both researchers and practitioners in optimizing parallel programming strategies across diverse computational settings.