Maxwell and Pascal aren't too different. Pascal is slightly different on a smaller process (since Maxwell is generally on 28nm and Pascal is on 16nm, that is a significant die shrink).
Maxwell and Pascal aren't too different. Pascal is slightly different on a smaller process (since Maxwell is generally on 28nm and Pascal is on 16nm, that is a significant die shrink).
| WolfpackN64 said: Maxwell and Pascal aren't too different. Pascal is slightly different on a smaller process (since Maxwell is generally on 28nm and Pascal is on 16nm, that is a significant die shrink). |
The rumour is that Switch is based on 20nm technology. Which seems likely a) because that fits in with the timeline when the Switch was being designed with Nintendo not being the fastest company when it comes to designing and bringing a product to market plus b) cheaper
Also the claimed 1ghz cpu speed etc fits in well with the 20nm thermal design. Considering a portable console can't go into a lower performance mode when it gets too hot as that would impact gameplay, they are normally designed to a lower performance level that can be stable at all times. I think 16nm would have allowed them to clock the Switch higher.

curl-6 said:
It's 30%, not 40. |
No it's 40%.
I believe it was around this: 157GFlops undocked, 393GFlops docked.
(39.94%)
Or quoting from the Eurogamer article. "That's not a typo: it really is 307.2MHz - meaning that in portable mode, Switch runs at exactly 40 per cent of the clock-speed of the fully docked device."
bonzobanana said:
The rumour is that Switch is based on 20nm technology. Which seems likely a) because that fits in with the timeline when the Switch was being designed with Nintendo not being the fastest company when it comes to designing and bringing a product to market plus b) cheaper Also the claimed 1ghz cpu speed etc fits in well with the 20nm thermal design. Considering a portable console can't go into a lower performance mode when it gets too hot as that would impact gameplay, they are normally designed to a lower performance level that can be stable at all times. I think 16nm would have allowed them to clock the Switch higher. |
Possibly, that would mean it's a slightly more efficiënt Maxwell GPU and a slightly less efficiënt CPU. But rumors are still rumors. Even if it's true, I wouldn't be surprised to see a die shrink down the line.
WolfpackN64 said:
Possibly, that would mean it's a slightly more efficiënt Maxwell GPU and a slightly less efficiënt CPU. But rumors are still rumors. Even if it's true, I wouldn't be surprised to see a die shrink down the line. |
Die shrink down the line definitely if it manages to sell well enough to justify it unlike the wii u situation.

bonzobanana said:
Die shrink down the line definitely if it manages to sell well enough to justify it unlike the wii u situation. |
I also believe the CPU in the Wii U was at that point really too old to justify any die shrink anyway. There are much more capable PowerPC chips they could have chosen. They shot themselves in the foot there.
| WolfpackN64 said: Maxwell and Pascal aren't too different. Pascal is slightly different on a smaller process (since Maxwell is generally on 28nm and Pascal is on 16nm, that is a significant die shrink). |
The difference that matter is -according to nVidia - Maxwell tops at 1TFLOPS and Pascal is 1.8TFLOPS.
That is almost twice.
OT: Reports claim that Pascal cost significantly more than Maxwell. Nintendo care about expense so all roads point to Maxwell. Modern Nintendo care nothing about power (nor apparently their fans) so why go Pascal?
All these rumours are counterproductive as they build up the console to be more than it is before reality and the web shoots it down.
justinian said:
The difference that matter is -according to nVidia - Maxwell tops at 1TFLOPS and Pascal is 1.8TFLOPS. That is almost twice. OT: Reports claim that Pascal cost significantly more than Maxwell. Nintendo care about expense so all roads point to Maxwell. Modern Nintendo care nothing about power (nor apparently their fans) so why go Pascal? All these rumours are counterproductive as they build up the console to be more than it is before reality and the web shoots it down.
|
You mean mobile Maxwell and Pascal. The top Maxwell GPU has around 6.1 Tflops of single precision processing power.


Barkley said:
No it's 40%. I believe it was around this: 157GFlops undocked, 393GFlops docked. (39.94%) Or quoting from the Eurogamer article. "That's not a typo: it really is 307.2MHz - meaning that in portable mode, Switch runs at exactly 40 per cent of the clock-speed of the fully docked device." |
It's 40% of docked mode, and 30% of a fully clocked Tegra X1. For some reason I thought we were talking about the latter, my mistake.
| bonzobanana said: Remembering that it's described as a custom chip its possible its basically Maxwell but may have some minor improvements that are also in Pascal. If they have stripped out the big/little arm cpu design and just left the big arm cpu's they have space to put something else, I suspect they may well increase cache to give full frame buffer size. Then how do you describe it when the chip is both improved on both Maxwell and Pascal in that regard. Let's say effectively you have a gpu architecture that is 2/3rds Maxwell and 1/3rd Pascal how would you describe it? If your honest you would say Maxwell but then in doing so you negate the Pascal improvements. It's a custom chip probably designed between the mid-gen refreshe's of Nvidia's 2 gpu architectures. What if Nintendo have designed it with full 3DS game compatibility for downloaded games not cartridges. Perhaps then there is a pica200 gpu in there too or part of the pica200 that not only helps run 3ds games but assists the main maxwell/pascal architecture for Switch games too. When chipworks do a scan of the Switch main chip we may find it looks absolutely nothing like the Nvidia reference designs. |
Well, it depends on the manufacture procedure.
Maxwell (in X1) is produced in 20 nm, while pascal is almost the same arquitecture as maxwell, with minor improvments here and there, but manufactured in 16nm.
If it was Maxwell but shunk to 16nm, it would be too similar to pascal, so id consider it pascal, even without the minor pascal improvments.
If it was has the minor improvments of pascal, but at 20nm, id consider it maxwell.
The die size is the important thing here, IMO.