Part 4/10:
The mechanism at play is largely due to the original design of the GPUs, which excel at performing massive amounts of matrix multiplication. Experts note that when GPUs finish processing their data, waiting periods occur until they receive the next set of instructions. Some engineers have tackled this issue by implementing dummy operations—like continuously executing zero multiplications—to maintain steady power draw, but this method proves to be inefficient.
The preference, instead, leans towards the use of advanced battery packs that can absorb shocks without compromising performance or requiring unnecessary energy consumption.