Skip to content
  • by
  • News
  • 2 min read

Nvidia to Adopt TSMC’s 3nm-Class Process Technology for Next-Generation GPUs

Nvidia is reportedly planning to embrace TSMC's advanced 3nm-class process technology in the coming year, according to a report by DigiTimes. The company is expected to utilize this technology for the production of its upcoming GB100 compute graphics processor. Furthermore, it is likely that Nvidia will adopt the same fabrication technology for all its Blackwell GPUs.

The GB100 is projected to be launched in 2024, following Nvidia's typical two-year pattern of introducing new GPU architectures. This next-generation compute GPU is specifically designed for artificial intelligence (AI) and high-performance computing (HPC) applications. Rumors indicate that the GB100 will incorporate a multi-chiplet design to significantly enhance performance compared to the current GH100 product, which is based on the Hopper architecture.

It remains uncertain which of TSMC's 3nm-class process technologies Nvidia will opt for. TSMC offers various options, including performance-enhanced N3P and HPC-oriented N3X. Nvidia has previously utilized customized fabrication technologies for its Ada Lovelace, Hopper, and Ampere GPUs. Therefore, it is possible that the company will continue this trend by using a custom node for its Blackwell graphics processors as well.

Nvidia will not be the only player to adopt TSMC's N3 technology in the near future. AMD, Intel, MediaTek, and Qualcomm are all slated to adopt one of TSMC's 3nm-class nodes between 2024 and 2025. In fact, MediaTek has already completed the tape-out process for its first N3E design with TSMC.

Currently, Apple is the sole company using TSMC's N3B (1st generation N3) technology for its A17 Pro system-on-chip in smartphones. However, this technology is expected to be adopted for other system-on-chips, including the M3, M3 Pro, M3 Max, and M3 Ultra, which will be featured in Mac personal computers.

– DigiTimes