TSMC’s gearing up to build Nvidia’s next-gen GPUs on 5nm

Nvidia’s prepping to build Lovelace and Hopper series GPUs using TSMC’s 5nm technology

According to a new report from DigiTimes (Paywalled), Nvidia’s preparing to build new RTX 40 series and H100 series graphics cards using their 5 nm lithography node.

Currently, TSMC is using most of its 5 nm manufacturing capacity to produce chips for Apple, and AMD are known to be using TSMC’s 5 nm technology to create their Zen 4 series processors. With AMD and Nvidia using the same lithography technology, it is likely that TSMC’s 5 nm manufacturing capacity will not be large enough to meet consumer demand for the latest tech.

Nvidia’s RTX 40 series graphics card will reportedly utilise a new GPU architecture called Lovelace, named after the famed mathematician Ada Lovelace. Nvidia’s future H100 series workstation chips should succeed Nvidia’s Ampere A100 series, using their new Hopper MCM architecture.

If the latest rumours are correct, Nvidia’s Lovelace architecture will target gamers while their Hopper architecture will target enterprise customers with a focus on data centre workloads and AI acceleration.  

  
Nvidia’s Lovelace and Hopper series GPUs and accelerators are due to launch in the second half of 2022, with many expecting Nvidia to launch refreshed RTX 30 series products in the first half of the year. 

Hopper should launch as Nvidia’s first MCM (Multi-Chip-Module) graphics card, using TSMC’s CoWoS technology to connect multiple chips together to create a larger, more powerful GPU/accelerator. Nvidia is reportedly not using an MCM design for its high-end Lovelace series gaming GPUs in 2022. 

You can join the discussion on Nvidia’s plans to use TSMC’s 5nm node for Lovelace on the OC3D Forums.