Keeping a vehicle well-localized within a prebuilt-map is at the core of any autonomousvehicle navigation system. In this work, we show that both standard SIR sampling and rejection-basedoptimal sampling are suitable for efficient (10 to 20 ms) real-time pose tracking without featuredetection that is using raw point clouds from a 3D LiDAR. Motivated by the large amount ofinformation captured by these sensors, we perform a systematic statistical analysis of how manypoints are actually required to reach an optimal ratio between efficiency and positioning accuracy.Furthermore, initialization from adverse conditions, e.g., poor GPS signal in urban canyons, we alsoidentify the optimal particle filter settings required to ensure convergence. Our findings includethat a decimation factor between 100 and 200 on incoming point clouds provides a large savings incomputational cost with a negligible loss in localization accuracy for a VLP-16 scanner. Furthermore,an initial density of 2 particles/m2 is required to achieve 100% convergence success for large-scale(100,000 m2), outdoor global localization without any additional hint from GPS or magnetic fieldsensors. All implementations have been released as open-source software.