Published :12/29/2020 3:18:42 AM
Click Count:2103
1. NVIDIA is planning a 5nm GPU.
In the morning news on December 29, according to foreign media notebookcheck, the 5nm GPU that NVIDIA is planning will be named "Ada Lovelace" and will have as many as 18,432 stream processors.
Although the Ampere architecture in the consumer market has just been launched, and the layout of the entire line has not been completed, and due to the shortage of GPU production capacity, RTX 30 graphics cards are currently hard to find in one card, but information about NVIDIA's next-generation GPUs has appeared from time to time In front of us.
After the Ampere architecture, NVIDIA's next-generation product was originally Hopper, but according to the latest news, a new generation of "Ada Lovelace" was added before that. The code name comes from Ada Lovelace, who is the only legitimate daughter of the poet Byron. She is known as the first computer scientist and wrote the first computer program in history.
2. Stream processors increased by 71%.
According to the latest news, the major core number of the Ada Lovelace architecture will be AD102. There will be as many as 12 GPCs, 72 TPCs, and 144 SMs, and each SM continues with 128 stream processors, so there will be as many as 18432. A stream processor, compared with the current 10752 Ampere architecture GA102 core, an increase of 71%. It is not yet confirmed when the NVIDIA AD102 core will be released, and it is expected to be part of the RTX 40 series.
3. A100 is the 7nm chip with the strongest computing performance.
The first GPU to use the Ampere architecture-NVIDIA A100, which is a GPU product designed for scientific computing, cloud graphics and data analysis. Since A100 has more than 54 billion transistors, it is also the 7nm chip with the strongest theoretical computing performance so far. In terms of specific parameters, the A100 has 19.5 teraflops (trillion floating point operations per second) of FP32 computing performance, 6912 CUDA cores, 40GB of memory and up to 1.6TB/s of memory bandwidth. NVIDIA is still enhancing its Tensor core To make it suitable for developers.
Like Volta architecture to create Tesla V100 and DGX systems, this powerful GPU will be applied to a stacked AI system to provide powerful computing power for supercomputers in global data centers. The new DGX A100 AI system is a This is a giant GPU combined with an A100 GPU, and the computing power of the entire system has reached an astonishing 5 petaflops (petaflops per second).
Before the release of the Ampere architecture, DGX A100 has been applied in the research related to this new crown pneumonia, and has played a key role in AI models and simulations with its powerful computing power. This has also successfully tested the practical value of the Ampere architecture GPU. .
