Hasn't Intel indicated that their GPU will probably be on 14nm? Maybe that's just one version of it?
Oh I see people are speculating because Raja visited Samsung. But the rumors there are talking Samsung fabbing Intel's GPU on 5nm next year? I'm very skeptical of that happening unless its basically just sampling it (kinda like AMD did with Vega 20 last year) while they wait for production to really ramp up.
I also have doubts about Samsung's process, in that I have a hunch they're gaming the naming to try and act like they're keeping up with TSMC.
Agreed! Regarding NVIDIA's slow move to 7nm, lets be honest, it's only because there is little competition and if they aren't in a rush that points to current NVIDIA offerings being more than capable of combating NAVI with reduced pricing. Looking forward to ampere, hoping the 3080TI is a beast.
I disagree. I don't think Nvidia had much of a choice. Apple probably got almost all the initial 7nm production, and AMD seems to have bought a large amount of the following (partly because they had a heads up that GF was going to stop their 7nm so they were sure to get ahead of things there; and they were aggressive on securing 7nm specifically to gain advantage over Intel and Nvidia), and then Qualcomm will have been fabbing the 855 there as well. Its entirely possible that Nvidia would've been very constrained to get 7nm any sooner than they are by going with Samsung. Heck, this seems kinda late to be making such an announcement even, so I really wonder how soon we'll get much 7nm products from Nvidia.
Samsung's 7nm is EUV and thus was like a year late (maybe longer, Samsung is saying they'll have EUV lines up and running 2020) compared to TSMC's 7nm. And also because of EUV, which is limited by the number of the EUV machines needed for it, means it might also be constrained some even once its up and running. Its possible that Nvidia is Samsung's only 7nm customer (outside of themselves), which is why Samsung is probably desperate to get customers.
I wonder if Apple is going to to 7nm+, as 6nm isn't ready yet. Which that would probably mean Nvidia wouldn't have been able to go TSMC 7nm+ this year either. Which, it seems like TSMC's 5nm (EUV) might be ahead of their 6nm (it starts risk production early next year)?
Nvidia likely had not a whole lot of choice. They'd have to pay TSMC more because of competition (because most were going to TSMC as they were the only ones with 7nm ready).
I definitely wouldn't read that this means Samsung's 7nm is going great. Heck, without Samsung bending over backwards to win them, I'm not sure we'd see Nvidia go with them even. We'll see, but I view this more as Samsung desperate for customers. I wonder if this might also mean that Intel passed on Samsung (so Nvidia really had to basically settle for them).
I do agree that Nvidia isn't terribly desperate to get 7nm stuff out because AMD isn't pushing them that hard (we'll see if Navi changes that any). But I'm not sure that's the actual limiting factor at play here. I keep trying to get people to realize that designing, engineering, and manufacturing GPUs is becoming a problem. It costs a LOT of money, it takes longer, and they get less gains from process (and trying to fab such large transistor dense chips on these new processes is presenting issues above that). On top of that, they're having to try and appease multiple markets with these chips, and its inevitable that its going to lead to things not being what gamers want. There's a reason why Nvidia talked about how much Pascal took to develop. Without mining, I think we might even have seen the dGPU market collapse (not totally, but such that it would cause a lot of concern; heck we kinda saw that even with mining with Softbank dropping their Nvidia investment), due to high prices and lack of real tangible performance improvements. I think that's why Nvidia felt ray-tracing was needed to try and reinvigorate the market for dGPUs.