Graph analytics plays a crucial role in fields like social network analysis, bioinformatics, and scientific computing, where handling large and irregular datasets efficiently is essential. SYgraph is a SYCL-based graph analytics framework designed to deliver high performance across heterogeneous GPUs. It enables a single-source programming model that runs efficiently on Intel, AMD, and NVIDIA GPUs without code duplication or vendor lock-in. SYgraph introduces a Two-Layer Bitmap data layout and a GPU-tailored load-balancing strategy, reducing memory overhead, achieving a low memory footprint, and eliminating pre- and post-processing. By leveraging SYCL’s unified programming model, we designed a load-balancing approach integrated with the Two-Layer Bitmap using Unified Shared Memory (USM) for data management, nd_range parallelism for fine-grained kernel control, and subgroup collectives for efficient workload distribution. We demonstrate how SYgraph achieves competitive performance against leading CUDA-based frameworks and evaluate it across Intel, NVIDIA, and AMD GPUs. SYgraph is currently being integrated into oneDAL to further expand its usability within Intel software ecosystem.