Until recently Nvidia has been separating some of its Turing silicon into A dies and non-A dies. This binning process meant that premium overclocked cards in the RTX 2080 and RTX 2070 ranges would feature A die Turing GPUs with better OC potential, and the lower price point 'reference style' cards used the non-A dies. Just ahead of the weekend, a report shared by Tom's Hardware Germany asserted that, going forward, Nvidia partners won't have the choice of A dies and non-A dies any more.
Previously purchased Nvidia RTX 2080 and RTX 2070 graphics cards will have come packing the following GPUs:
- TU104-400A-A1: Premium overclocked RTX 2080 graphics cards
- TU104-400-A1: Basic RTX 2080 graphics cards
- TU106-400A-A1: Premium overclocked RTX 2070 cards
- TU106-400-A1: Basic GeForce RTX 2070 graphics cards
TechPowerUp first noticed the 'segregation' of Turin GPU dies, as indicated above, back in September last year. It said that Nvidia pre-tests all GPUs and sorts them by properties such as overclocking potential, power efficiency, etc and then prices them to AIBs accordingly.
The recent report from Germany says that two new GPUs have already been in production at TSMC for weeks and won't be pre-binned by Nvidia ahead of going to partners. In coming RTX 2080 and RTX 2070 based graphics cards you will be able to open them up to see that they will be based upon the following GPUs:
- TU104-410: RTX 2080 graphics cards
- TU106-410: RTX 2070 graphics cards
The revised silicon from TSMC's 12nm process might be good enough now, and be produced in good enough yields, that the die segregation isn't necessary anymore. It would be interesting to see if things are improved at both ends of the spectrum with the new GPUs on their way to AIBs. At the time of writing there is no word on whether Nvidia will follow through with a revised GP102 die to eliminate non-A dies for GeForce RTX 2080 Ti cards.