Abstract: Graph neural networks (GNNs) are widely used for learning node embeddings in graphs, typically adopting a message-passing scheme. This approach, however, leads to the \emph{neighbor explosion} problem, with exponentially growing computational and memory demands as layers increase. Graph sampling has become the predominant method for scaling GNNs to large graphs, mitigating but not fully solving the issue.Pre-propagation GNNs (\PPGs) represent a new class of models that decouple feature propagation from training through pre-processing, addressing neighbor explosion in theory. Yet, their practical advantages and system-level optimizations remain underexplored.This paper provides a comprehensive characterization of \PPGs, comparing them with graph-sampling-based methods in training efficiency, scalability, and accuracy. While \PPGs achieve comparable accuracy, we identify data loading as the key bottleneck for training efficiency and input expansion as a major scalability challenge. To address these issues, we propose optimized data loading schemes and tailored training methods that improve \PPG training throughput by an average of 15$\times$ over the \PPG baselines, with speedup of up to 42$\times$ compared to sampling-based GNNs with superior accuracy on large graph benchmarks.