H2D

H2D is an acronym that can denote several concepts depending on the context. In computing, H2D most commonly refers to Host to Device data transfers in GPU programming and architectures. This describes moving data from the host system memory (CPU) to the device memory (GPU) before a kernel or computation runs. It contrasts with D2H (Device to Host) and D2D (Device to Device). The process may involve asynchronous copies, pinned (page-locked) host memory, and streams to overlap computation with data transfer, aiming to reduce latency and increase throughput. In CUDA, for example, explicit copies are performed via APIs that specify memory directions, and performance is sensitive to memory coalescing and bandwidth.

Other uses of H2D are less standardized and vary by field. In some discussions of digital transformation or digital twins, H2D is used to refer to ideas such as “Human to Digital” or “Human to Data,” describing the flow of human information into digital systems for analysis or simulation. These usages are not universally defined and may be proprietary to an organization or project.

If encountered, the intended meaning of H2D should be clarified from context. Related acronyms include D2H (Device to Host) and D2D (Device to Device). In a technical setting, H2D most reliably refers to Host to Device data transfer in GPU and parallel-computation contexts.