Unpopular truth: because that's where they need to be to ensure stability and yields.
There is a wide variation in the stability of hardware across various different tasks. An overclock/undervolt that is stable in one task is not necessarily stable in others, as anyone who's overclocked can attest. f.ex Just Cause 3 needed several hundred MHz less than I could get in TF2 or Witcher 3.
The voltage is set where it needs to be to ensure that there's no instability in any program, on any sample in a batch. People look at one task and one sample and assume that their OC must be stable on everything, on every card in the batch. In reality it's not, or not to the degree that the manufacturer requires.
Yes, you can get extra performance on any given sample by eating up your safety margin and pushing closer to the limit of that specific sample's frequency/voltage curve.
But as the economists say: if there were free money laying on the sidewalk, AMD would have picked it up already. They're not stupid, they ship the voltages they need.
There is a wide variation in the stability of hardware across various different tasks. An overclock/undervolt that is stable in one task is not necessarily stable in others, as anyone who's overclocked can attest. f.ex Just Cause 3 needed several hundred MHz less than I could get in TF2 or Witcher 3.
The voltage is set where it needs to be to ensure that there's no instability in any program, on any sample in a batch. People look at one task and one sample and assume that their OC must be stable on everything, on every card in the batch. In reality it's not, or not to the degree that the manufacturer requires.
Yes, you can get extra performance on any given sample by eating up your safety margin and pushing closer to the limit of that specific sample's frequency/voltage curve. But as the economists say: if there were free money laying on the sidewalk, AMD would have picked it up already. They're not stupid, they ship the voltages they need.