Nvidia’s next-generation GPUs use TSMC’s 5nm process
Samsung recently started a large-scale capacity expansion and technology research and development plan, intending to catch up with TSMC in the field of wafer manufacturing. Samsung has previously announced that it will invest $151.5 billion in the construction of wafer fabs in the next ten years, and the goal is to triple the current production capacity by 2026. At the same time, it plans to mass-produce the 3nm GAA process in the first half of 2022, and mass-produce the second-generation 3nm process in 2023. In addition, Samsung has also stepped up the development work of the chip and introduced AMD’s graphics technology in the Exynos series of SoCs. Recently, it is also committed to the development of automotive chips suitable for the new generation of cars, and news of the acquisition of related companies has also been reported.
Recently, it has been reported in the industry that Samsung hopes to introduce Qualcomm and AMD into its advanced process plan. AMD has a deep cooperation relationship with TSMC. The success of the 7nm process node has put AMD in the fast lane of development. As a more powerful competitor, facing Samsung’s aggressive offensive, TSMC will certainly not wait to die. Since Samsung can fight for AMD, TSMC can also compete for Nvidia’s GPU orders.
According to DigiTimes, Nvidia will use TSMC’s 5nm process to manufacture the next-generation Hopper architecture H100 GPU and the Ada Lovelace architecture GeForce RTX 40 series GPU. H100 GPU is NVIDIA’s product for data centers and AI computing, using CoWoS advanced packaging. Since the previous generation of A100 also used TSMC, it is not surprising that the same foundry will continue to be used. GeForce RTX 40 series GPU is a new generation of game graphics cards for ordinary consumers. It still retains the traditional design. It will not use MCM multi-chip packaging like AMD’s RDNA 3 architecture-based Navi 31. The full shift to TSMC will have a greater impact on capacity allocation in the future.
Although Samsung and Nvidia have good cooperation on the GeForce RTX 30 series. In particular, Samsung provided a very favorable price for this large order from Nvidia. However, its 8nm process is not so satisfactory in terms of performance, production capacity, yield, and other aspects, which has caused problems in GPU supply. TSMC may not give Nvidia a good price, but there should be better guarantees in the semiconductor process.
Although Samsung and Nvidia have good cooperation on the GeForce RTX 30 series. In particular, Samsung provided a very favorable price for this large order from Nvidia. However, its 8nm process is not so satisfactory in terms of performance, production capacity, yield, and other aspects, which has caused problems in GPU supply. TSMC may not give Nvidia a good price, but there should be better guarantees in the semiconductor process.
