omicverse.settings.gpu_init

settings.gpu_init(managed_memory=True, pool_allocator=True, devices=0)[source]

Initialize GPU mode with RAPIDS for accelerated single-cell analysis.

Parameters:

  • managed_memory (bool, optional) – Enable NVIDIA Unified Memory to support oversubscription for large datasets.

  • pool_allocator (bool, optional) – Enable RMM memory pool allocator for faster repeated GPU allocations.

  • devices (int|list[int], optional) – CUDA device index (or indices) to register in RMM.

Returns:

  • None – Sets self.mode to 'gpu' and configures RAPIDS/CuPy allocators.

  • Examples – >>> import omicverse as ov >>> # Initialize GPU mode with default settings >>> ov.settings.gpu_init() >>> # Custom GPU initialization >>> ov.settings.gpu_init(managed_memory=False, pool_allocator=True)