it almost feels like TSMC are doing the hardwork for Nvidia, and they just come along with their cookie cutter premade architecture, slap it on the new wafer for increased gains, and cash in.
I know there is more to it than that, but without TSMC or similar, Nvidia would surely stall
it almost feels like TSMC are doing the hardwork for Nvidia, and they just come along with their cookie cutter premade architecture, slap it on the new wafer for increased gains, and cash in.
I know there is more to it than that, but without TSMC or similar, Nvidia would surely stall
Well yeah of course. They have to rely heavily on TSMC to keep moving their designs forward. Edit in. This is why Intel got so stuck for so long on 14nm. ++plusplus etc.
That is why they should seriously consider setting up their own fabs. Because, as we saw with Ampere to this gen? they are over a barrel. Which kinda amazes me why they have never done it before.
Like if AI really is all that, and they really have the billions it is said they do? they need to do it. It is one area Intel has covered, and can absolutely nail Nvidia when it comes to cost. They have no middleman.
And if their GPUs get better, which I suspect they will? yeah, Nvidia could be in trouble. They need to forward think, which is something they never really bothered to do.
But yeah, in theory Nvidia tape out a design and the success or failure comes down to the fab. Hence why Ampere was worse than Fermi with its generational leap, but it was cheap. Personally I think they should have stuck it out with Samsung, but I guess they knew AMD would eventually overtake them by a huge margin if they did so.
To be fair despite Intel slowing down it was only a year or two ago where their actual limitations compared to competition was their fabs and even then they had more dense chips being produced. It was only until TSMC N7+ did TSMC take over in density. Intel 4 is more than likely going to surpass that and match TSMC 4nm(despite 4 for Intel being akin to 7nm)
Nvidia and AMD rely on others to do 50% of the work. It's still cheaper than using their own fab. That's why AMD sold off Global Foundries all those years ago and also why Intel is allowing other companies to use their fabs. It's simply to expensive to operate solo.