Nvidia, which has been focusing on AI technology in recent years, will see the replacement of old and new products in 2022 for its data center products. At the upcoming GTC 2022, NVIDIA will focus on AI technology, and it is likely to bring a new generation of Hopper architecture GPUs. This is Nvidia’s first multi-chip module design-based GPU, which will be manufactured using TSMC’s 5nm process and CoWoS advanced packaging, supporting HBM2e and other connectivity features, and targeting the data center market.
Although the new computing card based on the Hopper architecture GPU has not been released, Nvidia does not seem to worry about sales at all. According to DigiTimes
, a person familiar with the matter said that Nvidia’s new generation of computing chips will be shipped as soon as the beginning of the third quarter of this year, and is expected to achieve an annual growth target of 200% to 250%. To this end, TSMC’s CoWoS advanced packaging also needs more materials, increasing the order volume by about three times.
It is rumored that the number of transistors of the GH100 based on the Hopper architecture will reach 140 billion, which is almost 2.5 times that of the current Ampere-based GA100 (54.2 billion) or AMD’s CDNA 2-based Instinct MI200 series (58 billion). The chip size of the GH100 is close to 900mm², and it is equipped with 288 SMs, which can provide three times the performance of the current A100 computing card. Although it is smaller than the rumored 1000mm² in the past, it is larger than the GA100 (862mm²) and the Instinct MI200 series (about 790mm²).
In addition, NVIDIA may also prepare a COPA solution based on the Hopper architecture. Nvidia researchers have published an article on multi-chip design solutions, and there may be two chips based on the same architecture in the future, designed for high-performance computing (HPC) and deep learning (DL) market segments.