NVIDIA GeForce RTX 2070 Speed Binning Exposed: TU106-400A vs TU106-400

EVGA is one of the few companies to release multiple GeForce RTX series cards based on NVIDIA's reference specs, and at reference price points (an important qualifier). That means you can actually buy a GeForce RTX 2070 for $499.99, rather than pay a premium for NVIDIA's own overclocked Founders Edition model, or any of the several overclocked SKUs by NVIDIA's add-in board (AIB) partners. EVGA, for example, sells an overclocked GeForce RTX 2070 for $569.99. As it turns out, there might be more of a difference than just the out-of-the-box specs. More on that in a moment.
For anyone who has not been following the GeForce RTX launch, NVIDIA released three cards: GeForce RTX 2080 Ti, GeForce RTX 2080, and GeForce RTX 2070. All three are available from NVIDIA as Founders Edition models, and unlike the previous generation (Pascal), the FE models come overclocked. NVIDIA also overhauled the cooling solution, trading the blower-style design for a cooler that is similar to what its AIB partners typically use.
NVIDIA also announced reference specs at lower price points, with the presumption its AIB partners would offer both reference and overclocked models. Here's how the pricing shakes out:
- GeForce RTX 2080 Ti Founders Edition: $1,199
- GeForce RTX 2080 Ti Reference: $999
- GeForce RTX 2080 Founders Edition: $799
- GeForce RTX 2080 Reference: $699
- GeForce RTX 2070 Founders Edition: $599
- GeForce RTX 2070 Reference: $499

Now here's where things get interesting. Steve Burke at Gamers Nexus got his hands on two different GeForce RTX 2070 models from EVGA, and specifically these ones...
Burke stripped both cards to get a look at the GPU, and lo and behold, they're slightly different. The GPU underneath the hood on the Black Gaming card is a TU106-400-A1 and the GPU on the XC Ultra Gaming is a TU106-400A-A1. Both are Turing GPUs and they are essentially the same in terms of specs and features. However, it appears that the "A" revision is a binned part that is able to run faster and hold its boost clock for a longer period of time.
It's important to understand that not all chips are the same. That applies to all semiconductors, whether it's a CPU, GPU, or RAM. In essence, you could buy two of the same model processor and see very different overclocking results. This is where binning comes into play. Companies like NVIDIA have the ability to test and sort chips based on how well they handle higher voltages and higher clockspeeds.
At first glance, Burke seems to have discovered that NVIDIA is binning its Turing chips, which in turn is selling two different versions to its board partners (at least for the GeForce RTX 2070). The TU106-400A-A1 likely costs more, and so it ends up in higher priced cards with factory overclocks. Does it matter? It can, as Burke noted when running the two EVGA cards in 3DMark.
Running at overclocked specs versus reference clocks mostly equates to a few frames per second, depending on the game and settings. However, it's interesting that the difference between a stock-clocked card and an overclocked one isn't just the specs, but the actual GPU. Something to bear in mind when/if shopping for a GeForce RTX card.