Unlike AMD’s move to 7nm in 2019, NVIDIA is conservative in GPU processes, but this year it seems different, with the exception of 7nm amperes, the 5nm GPU. NVIDIA’s current Turing GPU is using TSMC 12nm process, considering that 12nm is only an improved version of 16nm, so NVIDIA uses 16/12nm nodes for at least 4 years, which is really rare in GPU history, in the past is more than 2 years or so to upgrade the process.
Of course, in NVIDIA’s view, 16/12nm process GPU performance, energy efficiency are very powerful, ahead of the opponent for a year or two, why take risks, spend more money to upgrade the GPU process?
Earlier news also reported that NVIDIA this year will use the lower cost of Samsung 8nm or even 10nm process, but these days the wind changes, in addition to the 7nm amper GPU, NVIDIA’s next generation of 5nm GPU, rumored Hopper also frequent exposure.
Yesterday’s news that 7nm ampere GPU is mainly TSMC contract, 5nm Hope GPU is Samsung 5nm process contract, but Digitimes reported that TSMC also received NVIDIA 7nm and 5nm GPU orders, this meaning is different, divided into two foundry means nVIDIA orders will not be small, suggesting that NVIDIA in the new generation GPU bet not small, to know that this year AMD will only upgrade to the GPU The 5nm GPU is gone.
NVIDIA’s 7nm, 5nm GPU has not yet been hammered, a more reasonable analysis that 7nm ampere GPU is for the data center market, 5nm Hope GPU is for the game card.
If this is the case, then NVIDIA this year is probably a wave of outbreaks, 12nm GPU have performance, energy efficiency advantages, 7nm, 5nm not to mention, Lao Huang this year this is to open up directly, all this will be announced in the GTC 2020 speech on May 14.