So a colleague website just posted this, and since the cat is out of the bag we might as well post it also. A while ago we noticed separated (two) GPU hardware IDs for the same Turning GPU. Let’s call it an A and B model for each GPU. The one is a standard SKU, the other an OC version. And here is how that works.
So let’s take the GeForce RTX 2080 Ti as an example, the GPU in there is a TU102. So basically NVIDIA offers two chips based on TU102, one is the TU102-300 and the other TU102-300-A. That A model is short for an OC SKU. So when a board partner is using the TU102-300 (and not TU102-300-A), then they are not allowed to factory tweak it. Thus such a product would end up at reference clock frequencies and would end up in the cheaper blower cooler style products right? Likely the better yielded GPUs end up as an A model.
Now here’s why I wanted to write a news item on this: people can still manually tweak that non-A model. So you as a person could grab Afterburner or any tool of your preference to overclock yourself. Chances, however, are higher than the overclock might be less than the A model.
I hope that clarifies a thing or two. BTW TPU who reported this is spot on, we’ve verified this weeks ago already with many AIBs. And from what we learned, all of them are simply opting the A (OC) series GPUs.
