NVIDIA, NVIDIA spends billions to secure TSMC capacity for RTX 40 graphics cards build at 5nm,

NVIDIA spends billions to secure TSMC capacity for RTX 40 graphics cards build at 5nm

NVIDIA’s next-generation GPUs include Hopper and Ada Lovelace architectures (RTX 40), both are based on TSMC’s 5nm process. According to media reports, they prepaid $1.64 billion in Q3 and will pay another $1.79 billion in the future, while the overall long-term order prepayment is a whopping $6.9 billion, much higher than last year’s level.

The $6.9 billion order is not sure about the ratio of Samsung to TSMC, but from the process upgrade direction, it is clear that finding TSMC to buy 5nm capacity accounts for most of it, so it is safe to say that for next year’s RTX 40 series graphics cards, NVIDIA is paying real money for this wave.

In the second half of 2022, NVIDIA is planning to bring RTX 40-series graphics cards to market that are based on TSMC’s 5nm process. According to reports, the company is investing a considerable investment to make this possible.

NVIDIA is adopting TSMC’s 5nm production for its RTX 40 series

According to recent reports, NVIDIA plans to launch its next-generation Hopper and Ada Lovelace graphics cards in 2022. Both GPU families will be based on TSMC’s 5nm process. Hopper targets data centers, artificial intelligence, cloud computing, metaverse, and other massive computing areas. It would be based on MCM designs and use TSMC’s CoWos.

Ada Lovelace will be NVIDIA’s gaming architecture and will be part of the RTX 40 Series family. The most powerful silicon would be the AD120 with up to 18,432 CUDA cores. The number of CUDA cores is increased by 71% and frequencies of up to 2.2GHz can be achieved. According to rumors, FP32 could reach 81TFlops, which is a 1.25x increase over the RTX 3090.

NVIDIA’s RTX 40 series will be based on TSMC’s 5nm process, although it won’t be a preferred customer. Apple would be the main customer, followed by MediaTek, AMD, and other manufacturers.

This is because Apple, MediaTek, and AMD have priority and have paid for that priority. Demand is assured, and they didn’t have to spend a lot of money for this priority. For this reason, NVIDIA had to give a big advance to secure 5nm production.

NVIDIA would have paid $1.64 billion in the third quarter, with $1.79 billion remaining in the near future. The total amount to be paid by NVIDIA is $6.9 billion, which will be split between Samsung and TSMC. It is not clear what share each manufacturer will have. However, NVIDIA wants most of TSMC’s chip production to be under the 5nm node.