a new level

emanresu1

Hi forum. please tell me why you have not yet taught the kernel to work on the GPU without the participation of the CPU? what do you see as the difficulty?

emanresu1

I think that a minimum of kernel alterations is needed, it is enough for the kernel to think that the graphics card (GPU) is the CPU. ALU can be created programmatically.

emanresu1

perhaps the question may arise why is all this necessary? the most important thing is that there will soon be a deep integration of neural networks into operating systems. making core code for new neural processors will only complicate the already bad situation in the core when a large number of CPU types like aarhc64 complicates the development of kernels. the simplest and most inexpensive solution is to shift the CPU tasks to the GPU. for laptops, it would be possible to solve the problem of rapid battery discharge in this way. many manufacturers are trying to solve this problem now. a good solution would be to leave only the GPU. the CPU would not consume the battery. this could also solve the problem of overheating laptops. also, the work of servers and workstations whose GPU work performed the role of the CPU due to the huge parallelization capability would bring them to a new level in performance