By using this site, you agree to our Privacy Policy and our Terms of Use. Close
Pemalite said:
Bofferbrauer2 said:

In 7nm, that chip shouldn't be any bigger than the X1 in the Switch, which is produced in 28nm, which means production costs would be about the same.

A chip built at 7nm and a chip built at 14/16nm that is the exact same physical size? It will be cheaper to manufacture on 14/16nm.
Wafer costs go up every time you shrink
.

Bofferbrauer2 said:

Keep in mind I said 2021, by then the 7nm process should have matured enough, though there's still the possibility of a 10/12nm process (16/14nm would just be too outdated by then even by Nintendo standards imo).

I think you might have fallen for the advertising ploy that is "nm" naming these days.
Whilst 12nm does have inherent advantages over 14nm at say... Global Foundries... 12nm is just a refined 14nm process, which in turn is a refined 20nm process
.

At-least Samsung and TSMC changed their BEOL.
You are looking at probbal 10-15% gains going from 14nm to 12nm at most... And I think I am being generous there.

Bofferbrauer2 said:

RAM prices are falling again, and by 2021, DDR5 should be arriving.

DDR5 doesn't equate to LPDDR5.
Ram prices may also increase by 2021... You only need a couple of factories to be taken offline from a weather event
.

Bofferbrauer2 said:

 So I doubt 8GiB DDR4 by then would be more expensive than 4GiB when the Switch launched. While I didn't precise it here (I did so before in another thread), that upgraded Switch+ would be released at the same 299$ pricetag as the original Switch in my book.

I was more or less pointing at the memory speed on a 128bit bus in conjunction with a doubling of DRAM.

You can have the Ram run at the same clockrate/Mhz as the current Switch and you would still more than double your bandwidth, so it doesn't make sense to drive up the memory controller which consumes power when you have already made massive bandwidth gains.

Or, Nintendo might opt for higher clocked memory but a smaller bus which brings with it a ton of cost-benefits like a simper memory controller, less PCB traces and thus layers, simpler power delivery... You know. The usual.

@bolded: That's why I said about the same, not exactly the same. Sure, the 7nm will be more expensive, but not by much.

@italic: Nope, more like the original meanings, like 16nm and 12nm being the full nodes while 14nm and 10nm being half nodes. GF calling their upgraded 14nm a 12nm process was pretty awkward for me in that regard.

That said, I don't remember seeing any ARM chip being produced in any 12nm process, they all seemed to go straight to a 10nm process. I doubt anything else than Ryzen+ will be produced in that process either.

@underscored: I don't expect DDR5 in any Switch revision. Just wanted to point out with this that when CPU transition to another RAM standard, the old standard generally drops markedly in price. Hence why I followed it with saying that by then 8GiB DDR4 shouldn't be more expensive then compared to the 4GiB when the Switch launched.

@bolded and italic: I have a feeling that the bandwith is already a bottleneck on the Switch, hence why I wanted to widen it more than the increase in GPU processing power. Switch corrently is only using LPDDR4-1600, resulting in measly 25.6 GB/s bandwith. I agree that going all the way to LPDDR4-2666 while also doubling the channel with to 128 bit is probably overkill (LPDDR4-2133 would probably suffice), but 83 GB/s would certainly be a safer bet for no bandwith choking than just going 128bit and keeping the LPDDR4-1600