You may not realize it, but not all GeForce RTX 2080 and GeForce RTX 2070 graphics cards are created equal. We are not just talking about the custom cooling solutions and PCBs (printed circuit boards), but the actual GPUs inside for each model—NVIDIA separates its Turing GPU dies for those models into two categories, or at least that is how things used to be done. Not anymore, apparently.
We reported on this separation of GPUs last November. Otherwise known as "binning," NVIDIA up to this point has offered its hardware partners a lower end (and presumably cheaper) variant of a GPU for non-overclocked cards, and a higher-end GPU for models that will be factory overclocked. Here's how things shook out up until now...
- TU104-400A-A1: Overclocked GeForce RTX 2080 cards
- TU104-400-A1: Stock GeForce RTX 2080 cards
- TU106-400A-A1: Overclocked GeForce RTX 2070 cards
- TU106-400-A1: Stock GeForce RTX 2070 cards
For both families of GPUs, the "A" revision represents dies that have been binned to achieve higher clockspeeds. Steve Burke at Gamers Nexus had discovered the divided GPU families, and noted in his testing of EVGA's GeForce RTX 2070 XC Ultra Gaming (TU106-400A-A1) and GeForce RTX 2070 Black Gaming (TU106-400-A1) that there was a significant gap in clocks.
"The XC Ultra averages about 1935MHz for the entirety of the test, whereas the 2070 Black experiences clock decay from about 1815MHz down to 1785MHz. The average frequency bounces between 1785MHz and 1800MHz for the 2070 Black. This is a significant difference for out-of-the-box frequencies, and indicates that performance advantage resultant of the better binned GPUs," Burke wrote at the time.
What about now? Igor Wallossek at TomsHardware Germany says NVIDIA has put the word out to "several sales channels" that it is shifting to the use of solely higher end GPUs. The new variants include the TU104-410 (GeForce RTX 2080) and TU106-410 (GeForce RTX 2070).
The neat thing about this decision is they will all be qualified for overclocking. Discretion will be up to each individual hardware partner, but from a consumer standpoint, all cards going forward will be created equal, in terms of the actual GPU inside. That has the potential to mean better overclocking performance on the part of the end user, even on stock clocked cards.
There are some caveats, though. First, even though the GPUs will no longer be separated into higher performing A revisions and lower performing non-A variants, there are other factors that can affect overclocking—PCB designs, the power delivery scheme, cooling, and so forth. And secondly, there will be no way of telling if you're buying a newer card with the latest GPU underneath the hood, or an existing batch that is using one of the older, divided GPUs.
We wouldn't be shocked if some retailers specifically stated which cards are using a newer GPU, though if and when that happens, you can probably expect a price hike. This will work itself out over time, however, as existing inventories are depleted and replaced by newer batches.