Although ray tracing produces significantly more realistic images than traditional rasterization techniques, it is still considered computationally burdensome when implemented on a head-mounted display (HMD) system that demands both wide field of view and high rendering rate. A further challenge is that to present high-quality images on an HMD screen, a sufficient number of ray samples should be taken per pixel for effective antialiasing to reduce visually annoying artifacts. In this paper, we present a novel foveated real-time rendering framework that realizes classic Whitted-style ray tracing on an HMD system. In particular, our method proposes combining the selective supersampling technique by Jin et al. [8] with the foveated rendering scheme, resulting in perceptually highly efficient pixel sampling suitable for HMD ray tracing. We show that further enhanced by foveated temporal antialiasing, our ray tracer renders nontrivial 3D scenes in real time on commodity GPUs at high sampling rates as effective as up to 36 samples per pixel (spp) in the foveal area, gradually reducing to at least 1 spp in the periphery.