Traditional local approaches to continuum damage mechanics are known to lead to mesh dependency due to damage localization. Some corrections for this ill-posed problem exist through localization limiters. The integral non-local regularization is widely used in quasi-static applications, but for dynamic simulations with an explicit time integration scheme this method is often not chosen due to its high computational cost. In addition, this approach presents the problem of spreading damage over time. In this talk, we present a novel approach to non-local continuum damage modeling, for high strain rate applications. The high computational cost of the approach is leveraged through massively parallel calculations. To limit the spreading, we discuss different possible choices in terms of damage laws, regularization variables (i.e., energy, equivalent strain or damage) or weight function used for the integration.