High-speed steady-state visual evoked potential (SSVEP)-based brain-computer interfaces (BCIs) have been developed to enable the communications between the human brain and external environments. One of the major issues in the real-world applications of SSVEP-BCIs is the laborious and time-consuming calibration process, triggering the development of transfer-learning approaches to leverage existing data from other users. A comprehensive investigation on the inter-and intra-subject variability in SSVEP data is thus needed to provide insight for designing future transfer-learning frameworks for SSVEP-BCIs. We hereby present the first study that systematically and quantitatively assesses the variability in SSVEP data, where the sources of inter-and intra-subject variability at low-and high-frequency range were identified using Fisher's discriminant ratios (FDRs). The insights gained from this work could drive the future developments of transfer-learning approaches to minimize the calibration efforts in high-speed SSVEP BCIs.