Nvidia Allegedly Shifts RTX 4090 Production Over to H100 Hopper GPUs
Nvidia has reportedly shifted some orders from the GeForce RTX 4090 to the H100 compute processor based on the Hopper architecture. while the information comes my driver There are factors in support of the report that we were unable to confirm at press time. This could be a result of US sanctions on China’s supercomputing sector.
Nvidia’s latest gaming and high-performance computing (HPC) chips are TSMC’s called 4N, as opposed to the commonly used N4, a 4nm-class node belonging to TSMC’s N5 family of manufacturing processes. Manufactured using customized process technology. Note that placing an order with TSMC won’t have an immediate impact given the time it takes to create the latest processors and how orders are being placed. On the other hand, the information has not been directly or indirectly confirmed, so please take it with a grain of salt.
Nvidia now uses TSMC to host a host of GPUs of varying complexity, including the latest GH100 (80 billion transistors, 814mm^2), AD102 (76.3 billion transistors, 608mm^2), AD103 and AD104 processors. is generating At TSMC it may sound like it wouldn’t be too difficult for him to reallocate from one product, but there are nuances. His AD102 GPU from Nvidia powers a GeForce RTX 4090, which tops the list of best graphics cards for gaming as of October 2022.
From TSMC’s point of view, it doesn’t matter what they produce on their N5-ready production line. Since both GH (Hopper) and GA (Ada Lovelace) are manufactured in the same fab, foundries have (almost) no problem shifting production within a given quota. Meanwhile, Nvidia is also using TSMC’s advanced packaging capabilities to create hopper GPUs for computation (i.e. using TSMC’s 3D Fabric 2.5D technology to connect HBM3 memory to the chip itself). are available).
From Nvidia’s point of view, things are quite different. Changing the TSMC allocation is difficult to begin with, but it’s one thing. Chip makers are prioritizing products that sell in the thousands (Nvidia’s H100 offerings cost $10,000 a piece, while his GeForce RTX 4090’s recommended price for consumers is $1,499). . Although the AD102 and GH100 are comparable in complexity, the profits and margins that Nvidia gets with these products are quite different, ultimately impacting the company’s R&D spending.
We all know the demand for graphics cards from gamers is high, but the shift in production could be a direct result of US sanctions on China’s supercomputer sector. It could cost Nvidia around $400 million if it doesn’t waive its right to ship computing GPUs. Chinese entity.