So let me take a step back. Like GPU, it operates at around 20 to 22 terabytes per second kind of speed. When you move to DRAM, it is almost 37 to 40 times slower than what the GPU wants it. 

You go even further down to the storage side, it is almost 800 times slower. So see how much of a bottleneck is created all throughout. When the data is moving from the GPU into the memory, into the storage, the bottleneck just gets amplified.