RTX cards from Nvidia are coming out soon, finishing a long dawn of leaks and speculations. Even the reviews will be out on 19th September after the embargo ends. Yet, there’s some new revelation every other day and the recent one comes from TechPowerUp.
The owners of CPU-Z, made a very interesting find. Every hardware comes with something called a device ID, which helps windows identify the hardware and install relevant drivers, and also allow or block certain features.
The recent Nvidia Turing cards come with two driver IDs, corresponding to every GPU. So basically we can have two IDs for the RTX 2080ti, “TU102-300” and “TU102-300-A“. These IDs can actually correlate to performance as the TU102-300-A is slated to have better overclocking, but also higher pricing.
TechPowerUp also stated in their original article that factory overclocking was forbidden in the cards with the TU102-300 device ID. Although you can overclock it out of the box, but obviously they won’t surpass the other variants of the cards.
This is not something Nvidia implements in the software, but it’s intrinsically a hardware limitation. There’s a slight difference in even similar models of a GPU or CPU, giving rise to something called silicon lottery. This is the reason why some processors and GPUs can attain a higher clock speed than their counterparts.
But companies now implement a process called binning which is categorizing of finished products based on their characteristics, found out using automated tests in the production lab. That’s why even Intel has two variants, the K and the non K processors, overclocking in the non K processors is locked in Intel CPUs because they are slightly inferior in quality to their K counterparts. But most of the time, the difference isn’t much.
Even in this case, with the two device IDs, Nvidia is implementing binning. This will obviously drive up the costs of the factory overclocked cards, but also make sure that you get superior silicon, if you pay for it.
This is not nessasarily a bad thing, because now GPU makers can target a certain user base. If you are someone who is happy with stock performance and won’t overclock, you can buy a cheaper card and save money. But if you are a overclocking enthusiast, you can pay a bit more and ensure that you get quality silicon, which will have a better overclocking headroom.
Nvidia’s AIB partners are also required to only use the TU102-300-A Turing chips in their advanced models, which is also stated in their contract. All founders edition cards so far are the -300-A variants, we are yet to see any -300 variants in custom cards.