NVIDIA Ampere: New architecture unveiled to be available on Gaming GPUs

After several rumors and a number of unofficial leaks, the NVIDIA reveals some of the features of its next architecture.

ΜOne of the first GPUs based on this architecture will be the new one Nvidia A100 and was officially unveiled by the company at the GTC 2020 digital event that ran live a few hours ago.

Η Ampere as everything shows it will be integrated in the consumer or otherwise gaming GPUs of the company but until then, the CEO stated that he will be first in the family of servers and high performance systems used by organizations and companies for a number of applications, such as machine learning.

Until recently, NVIDIA separated the business part (HPC) from the consumer (gaming / workstations) with two separate products, the Turing family of cards that live in the gaming market and the Volta family that meets all the specific requirements of modern supercomputers.

From this year the company intends to sell a product for both markets and under the name 'Ampere', we will see this goal achieved, probably before the end of the year. The top chip that appeared concerns the supercomputers and is called NVIDIA A100 and can cope quite well with modern requirements and promises great flexibility in many workloads.

It is manufactured at TSMC 7nm, a significant difference from the current lineup implemented around 12nm by the same semiconductor manufacturer, while the first company to receive systems (DGX A100 AI systems) will be Microsoft, as announced in the accompanying press release NVIDIA, followed by a few more that will build servers, such as: Dell, Cisco, GIGABYTE, Lenovo, Quanta and a few more.

54 billion transistors find their way inside the new NVIDIA A100 GPU in a total of 6912 CUDA Cores, while at the same time we see the entry of the third generation of Tensor cores with NVIDIA referring to the expanded capabilities, but also the highest performance from the past.

Some of these features such as faster Tensor cores, better data parallelism and the third generation NVLink interface are elements that we will see in the consumer graphics cards which are expected before the end of the year in the market and will cover most of the market. , ie the gaming community.

Source