What happens when a GPU reads memory

When a GPU executes a global memory load instruction, the data travels through multiple hardware components including the warp scheduler, register file, L1 cache, TLB, crossbar, L2 cache, and ultimately DRAM, with each stage adding latency to the operation. Researchers reverse-engineer this path using timing experiments on hardware like the RTX 4090 because NVIDIA doesn't publicly document these low-level details, which are critical for understanding GPU performance.