Based on ga100 ampere core, NVIDIA has released A100 accelerated computing card and other products. Not surprisingly, in this quarter, the geforce game card for consumer users will also be available. Recently, kopite7kimi, who frequently disclosed RTX 30 series video card messages, also gave a message“I heard that the public version of the ampere core can feel 2.1 ~ 2.2GHz”However, it may be that he is afraid of the subsequent "slap in the face", he cautioned that "it is not sure yet".
We know that in the last generation of RTX 20 series graphics cards, the highest public frequency is the RTX 2080 super, which can be accelerated to 1815mhz. Although it is not difficult to exceed 2GHz for non-public use, the RTX 30 series has gone out of the factory like this. The ampere architecture is said to be a leap forward, and the ultimate potential is certainly more amazing.
It is said that there are three types of RTX 3070, all of which adopt ga104 core, while RTX 3060 is ga106-300 core. At present, ga104-400 core (sku0) has designed two sets of PCB boards, one is pg141, 8GB gddr6x memory, the other is pg142, 8GB gddr6 video memory, both built-in 3072 CUDA, which is consistent with RTX 2080 super.
There is also the general RTX 3070 version of ga104-300, with 2944 CUDA built in, which is consistent with RTX 2080 and matches with 8GB gddr6 video memory. If the pin bandwidth can reach 16gbps, the total bandwidth under 256bit is 512gb / s, which is significantly improved.
It is possible for RTX 3060 Ti / super to integrate 2304 CUDA and RTX 3060 to integrate 2176 CUDA.
Therefore, the performance of n-card will definitely not let us worry. What really bothers us is whether the price will rise or not, and what is the range?