NVIDIA Turing GPUs, in addition to hardware ray tracing and architectural improvements, received another important difference from their predecessors. For them, NVIDIA has introduced differentiation in terms of overclocking potential. In fact, the company now supplies two types of GPUs for the GeForce RTX 2080 Ti, 2080 and 2070 video cards, which differ in the quality of the silicon crystal. Chips with the best overclocking potential are more expensive for NVIDIA partners, but they can be guaranteed to be installed in video cards with noticeable factory overclocking, while conventional chips can only be able to work in the nominal mode. This causes a significant variation in the cost of production GeForce RTX cards, depending on whether they are claimed to be factory overclocked or not. However, according to incoming information, soon NVIDIA is going to curtail the initiative to sell selected Turing crystals at a higher price.
According to the editor-in-chief of the German version of Tom's Hardware, Igor Wallosek, from the end of May, NVIDIA will begin to supply its partners with new revisions of the TU104 and TU106 processors for GeForce RTX 2080 and 2070 video cards. They will assume only one version of each type, TU104-410 and TU106-410, which will not have an additional gradation for the verified frequency potential.
Recall that currently the TU104 and TU106 processors are supplied in versions TU104-400A and TU106-400A for cards with factory overclocking and TU104-400 and TU106-400 for ordinary versions of the GeForce RTX 2080 and 2070. However, practice shows that the real differences between overclocking ceiling for different versions of chips is not too noticeable. The improvement of TSMC's 12nm technology, which is used to manufacture Turing generation GPUs, has led to the fact that the microcircuits coming off the assembly line are mostly similar in terms of frequency capabilities, and it makes no sense to somehow sort them further.
For this reason, NVIDIA decided to abandon the preliminary sorting procedure, offering partners to purchase chips of the same type in terms of target frequencies, and, if necessary, organize the selection of more successful specimens on their own. In the near future, the company should prepare a fresh firmware version that is compatible with the new revisions of the TU104-410 and TU106-410 processors and removes restrictions on factory overclocking of "non-overclocker" chips without the letter A in the marking.
It is hoped that the unification of the TU104 and TU106 processors in terms of target frequencies will cause some reduction in the cost of GeForce RTX 2080 and 2070 cards, especially modifications with higher frequencies. The new TU104-410 and TU106-410 chips will be sold at the price of simpler versions of the previous revision, and in addition, NVIDIA is going to reduce the price of overclocker chips TU104-400A and TU106-400A by $50 until they are completely sold out.
Source: 3dnews.ru