Tell me please, did anyone compare the performance of the m1 GPU with other video cards, like 3080?
To evaluate how well they perform for the tasks of learning fully connected, convolutional, recurrent layers.
And does it make sense now to think about using the m1 GPU
Take a look at some of the posts on this thread. One compares Nvidia and Apple GPUs. Although both have hardware acceleration, Apple’s hardware acceleration only increases performance by a tiny amount (well, not really “tiny”, but that’s for another discussion). On Nvidia, you can reach 100 TFLOPS of processing power or more because of tensor cores.