NVIDIA recently had two high-performance mobile GPUs, the RTX 3080 Ti and the RTX 3070 Ti. The RTX 3080 Ti features the first-appearance GA103 core, which continues to use the GA104 core.
As early as May 2019, the core of GA103 has appeared in front of the public. This core is specially designed for mobile terminals, still using Sanxing 8nm process manufacturing, Ampere architecture. The RTX 3080 Ti opens 58 groups of units, a total of 7424 CUDA, and the number of fully open blood plates should reach 7680.
According to the previous news, the graphics card is equipped with 256-bit wide video memory, but it is rumored that up to 10 32-bit controllers are built-in, physically as high as 320-bit.
The real photo of the core of the GA103 exposed, the full number is "GN20-E8-A1". It's unclear if the GA103 core will land on the desktop, or if so, will it be called the RTX 3080 SUPER? Or downgrade to the RTX 3070 Ti.
But based on the GA103 cores seen, NVIDIA should only open up 8 of them. After the measurement of foreign media, the area of this core is about 496 square millimeters, which is 26.5% larger than GA104 and the number of cores is 25%.
Not only that, but the ampere family's smallest GA 07 core also has the first photo, this QS sample is not available for retail, the surface number "N20Z-GS1-A1", mainly used in RTX 3050 Ti and RTX 3050 notebook graphics cards; The desktop version of the RTX 3050 still uses the GA106 core.
The six cores of the NVIDIA Ampere family have finally come together, and the consumer GA102, GA103, GA104, GA106, and GA107 are all manufactured using Samsung's 8nm process; The GA100 for data centers and high-performance computing uses TSMC's 7nm process and has up to 54.2 billion transistors and 826 square millimeters of core area.
Editor's comments: According to the rhythm of the two-year generation of products, NVIDIA will launch the Hopper architecture in the third quarter as soon as possible, and the GTC conference in April should announce products for the professional market, which can be seen.