NVIDIA’s next-generation “Ampere” GA103 and GA104 core detailed leaked: 7nm, 20GB video memory

The RTX 20 and GTX 16 Super series have been laid out (except for the RTX 2080 Ti Super), and in the face of AMD’s constant counterattack, NVIDIA’s next-generation products are not far away. It is said that the March GTC conference You can see the core of the new architecture “Ampere”.

NVIDIA's next-generation According to new insider info, the two new core codes of GA103 and GA104 are obviously part of the high-end sequence of the Ampere family. It is also the first time in many years to see the NVIDIA core code ending with the number 3.

GA103: 60 SM arrays, 3480 stream processors, 320-bit video memory, 10 / 20GB GDDR6 video memory capacity.NVIDIA's next-generation

GA104: 48 sets of SM arrays, 3072 stream processors, 256-bit video memory, 8 / 16GB GDDR6 video memory capacity.NVIDIA's next-generation

Both new cores will be 7nm processes, but the foundry is unknown, and given the close specifications of the two, it is not ruled out that TSMC and Samsung can be manufactured separately. After all, TSMC’s 7nm production capacity is already tight, and NVIDIA and Samsung have a good relationship.

Not surprisingly, they will become RTX 3080 and RTX 3070 respectively. Compared with the RTX 2080 Super and RTX 2070 Super, the specifications have been greatly improved. The number of stream processors and the memory capacity have increased greatly. 20GB is already a professional card level. The memory bit width has also increased.

As for whether it will continue to hide some strength when it is first launched, and then release the full Super version, it is not yet confirmed, it still depends on the process yield and competition situation.

In addition, it is the first time for a sequence name such as GA103. Coupled with the core scale, which is still not up to the level of TU102 RTX 2080 Ti, there will definitely be GA102 on it, corresponding to RTX 3080 Ti and Titan.