This is incredible. I really like how you're able to trace the evolution of the systems as well. It makes me wonder what the next hardware revolution will be.

This is incredible. I really like how you're able to trace the evolution of the systems as well. It makes me wonder what the next hardware revolution will be. It seems like most resource intense applications are bottlenecking at transferring memory. UE5's nanite tech hinges on the ability to transfer memory directly from disk to GPU, Netflix built specific hardware to avoid copying memory between userspace and hardware, and I wonder how much other performance we're missing out on because we can't transfer memory fast enough. How much faster could AI training be if we could get memory directly from disk to the GPU and avoid the CPU orchestrating it all? What about video streaming? I have a feeling these processes already use some clever tricks to avoid unnecessary trips through the CPU, but it will be interesting to see which direction hardware goes with this in mind.