Rumor: Nvidia’s upcoming Lovelace GPUs will receive up to 18,432 CUDA cores

Nvidia is rumored to come with a new GPU architecture called Lovelace. GPUs based on this architecture are likely to be produced at 5nm and will receive a maximum of 18,432 CUDA cores.

Twitter user kopite7kimi, which more often shares details about unannounced components, reports that the Lovelace-based AD102 GPU may get twelve graphics processing clusters, while the current Ampere architecture has seven such clusters. With that, the Lovelace chip would have a maximum of 72 tpc’s, 144 streaming multiprocessors and 18,432 CUDA cores. That is an increase of 71 percent compared to the current GA102 GPU, writes VideoCardz. It is not yet clear when the first Lovelace GPUs will appear on the market.

The first reports of an Nvidia Lovelace architecture appeared earlier this month. It was already reported that the GPUs are made on a 5nm process. It is not yet clear who will produce the GPUs; TSMC and Samsung both already make chips on their own 5nm nodes. Samsung currently produces Nvidia’s current RTX 30 graphics cards, but the previous RTX 20 GPUs were made using TSMC’s 12nm process.

Lovelace would succeed the current Ampere architecture. Initially it was expected that Hopper would be the next GPU series from the chip manufacturer, but according to rumors, this architecture is being postponed. Hopper is said to be the first Nvidia GPU to use ‘multi-chip modules’. It is not yet clear whether that architecture will eventually end up in consumer GPUs.

* Based on unconfirmed rumors