NVIDIA introduced the Ampere graphics architecture on May 14 in a show at the GTC 2020. The architecture debuts with the A100 GPU, but for data centers. However, there are some features that will be transferred to future GeForce RTX that we should know about.
Ampere, a unified architecture for data centers and gamers
Unlike Turing and Volta, who targeted different markets, NVIDIA claims that Ampere is a single architecture for data centers and gamers. As a result, Ampere will be able to adapt to any need, making it far more versatile than previous generations of NVIDIA GPUs, where the company focused on environmental protection with a GPU for each specific segment.
Leap to 7 nm
NVIDIA makes the necessary leap to a higher processing node already used by competitors, namely the 7nm. Looking back at the green team itself, the transition from the 28nm process of the GeForce GTX 900 series to the 16nm process of the GTX 10 series resulted in a huge performance boost, which should also be reflected in the arrival of 7nm transistors, which are capable of adding more transistors in a similar space.
More transistors and more cores
NVIDIA is ready to add more transistors and cores with the 7nm process. Tesla V100, which uses 21.1 billion transistors, 5,120 CUDA cores, and 80 multiprocessor clusters in its 815 mm^2 array, the new amp-based A100 adds 54 billion transistors, 6,912 CUDA cores, and 108 SMs in its 826 mm^2 array.
Smarter AI than ever before
The A100 uses third-generation tensor cores that significantly improve performance for medium-precision 16-bit floating-point tasks in the “FP16”, add “TF32 for AI” functions for single-precision tasks, and now support double-precision tasks in the FP64. Applied to video games, this could improve computation in DLSS technology to increase performance without significant loss of quality, among many other features such as RTX Voice.
These are the key benefits we will see transferred to the future GeForce RTX 3000 graphics cards that will be released later this year. We will keep you informed.