get_gpu_frames

Client.get_gpu_frames(peer_devices=())

Zero-copy CuPy access to the server’s GPU movie buffers (Windows, same computer).

Call before start_acquisition. The returned GpuFrames holds both movie buffers as CuPy arrays (.buffers) and the semaphores; iterating it yields each finished buffer as batch.images (cupy float32, images x height x width, on GPU .device) and batch.info (frame, scanX, scanY, valid per image). The buffer is released back to the server when the loop advances; the server’s processing waits while it is held, so close it when done (or use it in a with block). Requires CuPy. See deapi.gpu_frames.

peer_devices: other GPUs that will work on the frames. They read the buffers directly where the driver allows peer access (.peer_access), and their current streams are synchronized before each release either way.

Only a process on the server computer can map the buffers; from another computer this raises with the server’s reason. Against deapi’s simulated server this returns a SimulatedGpuFrames instead: the same interface, with NumPy batches of the simulated dataset and no CuPy needed.