Rumor mill: Nvidia’s Ada Lovelace GPUs are still a pretty well-guarded secret, but industry insiders believe the company will rely on TSMC’s N5 node to manufacture them. These are expected to be monstrous GPUs with 50 percent more gaming performance and higher power consumption than RTX 3000 series GPUs, so dropping Samsung’s 8nm node could help on both fronts.
Nvidia RTX 4000 graphics cards could fail as early as next summer, and it is also possible that we will see an update to the RTX 3000 series before that. The next generation of cards from Team Green is codenamed Ada Lovelace on the inside, but little is known about the new architecture and how what she will bring. The only opinion among industry insiders seems to be that net performance will double over the current generation, while power consumption will be north of the 400W range.
It is rumored that Nvidia will use TSMC’s 5nm technology node, which is believed to be significantly more expensive than its 7nm technology node that AMD uses for its RX 6000 series GPUs. report Digitimes appears to be confirming these expectations, citing supply chain industry sources.
The publication also notes that Nvidia plans to use TSMC’s chip-on-wafer-on-substrate (CoWoS) packaging technology for its upcoming H100 GPU. This model will be based on Nvidia’s compute-centric Hopper architecture and is expected to be the company’s first multi-chip GPU module. Ada Lovelace GPUs will likely use FC-BGA packaging provided by ASE technology as well as traditional GDDR6X memory.
The RTX 3000 series GPUs are manufactured using Samsung’s relatively mature 8nm technology node, so moving to TSMC’s 5nm node is risky as this is also something other companies are going to use for their future products. Apple currently manufactures its A14, M1 Pro and M1 Max chipsets on the same node, and AMD plans to use it for its Zen 4 processors. At the same time, Bitmain provided some of TSMC’s 5nm manufacturing capacity for the mining hardware that will presented in the first quarter of 2022.
Nvidia may have decided to use TSMC for both Hopper and Ada Lovelace, which may have allowed it to provide more storage. Naturally, the company will not comment on the news, so you have to wait and see.