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

Forums - Nintendo Discussion - VGC: Switch 2 Was Shown At Gamescom Running Matrix Awakens UE5 Demo

numberwang said:

From the pastebin: https://pastebin.com/V5nTeh4h

Tegra X1 on TSMC 20nm is 116mm2, while the Tegra X1+ on TSMC 16nm FinFet is 100mm2. Nvidia Orin (full T234) is 455mm2. If we assume that T239 (= Switch2) is about 50% the area, we end up with a die size of 227mm2 on Samsung 8N.

That would be a big boi die. From what I can find on the interwebs, Steam Deck's APU is 163mm² and a desktop 4070 is 294,5mm².

I think they have to keep it small but i think they got such a good deal on these 8nm that they will probably launch a pro in much smaller form factor 2 years later.



Around the Network

Prediction time:

Switch 2 mobile:
Max 10W for the whole system, max 7W for the SOC
GPU 3xx MHz
~1 TFP
Heavy 3rd party games will run at 360p DLSS-uped to 1080p


Switch 2 docked:
Max 13W whole system, 12W for SOC
GPU 7xx MHz
1.6 TFP
3rd party games: 540p DLSS-uped to 1440p (maybe the last step from 1440p to 4K using a less intensive method)

Compare 11W and 9W max for Switch 1 when docked and mobile. We will need slightly better cooling this time.
https://www.anandtech.com/show/11181/a-look-at-nintendo-switch-power-consumption/2



numberwang said:

From the pastebin: https://pastebin.com/V5nTeh4h

Tegra X1 on TSMC 20nm is 116mm2, while the Tegra X1+ on TSMC 16nm FinFet is 100mm2. Nvidia Orin (full T234) is 455mm2. If we assume that T239 (= Switch2) is about 50% the area, we end up with a die size of 227mm2 on Samsung 8N.

That would be a big boi die. From what I can find on the interwebs, Steam Deck's APU is 163mm² and a desktop 4070 is 294,5mm².

227mm2 ... is an absurdly large die size. As stated there it's practically double the size of a Tegra X1 (20nm), the Switch 1 launch processor.  

Even the Steam Deck APU is only 161mm2 I believe, the bulky Legion Go's APU is not even that large ... 227 mm2 would be so big I'd almost want to see it because it would be huge. There's no way they could keep the Switch's form factor it would have be significantly larger than the OG Switch. 

With due respect to Kopite, he got the name correct, but by his own admission he was wrong on the codename (Dane is not Drake), the CUDA cores wrong, the Ada/Lovelace thing wrong ... I mean basically he got everything wrong outside of the name if you look at the Nvidia leak. I dunno, it just doesn't seem to make a lot of practical sense for Nintendo to release a honking Steam Deck sized portable. 

Last edited by Soundwave - on 14 September 2023

The size of the device is not determined by the die size but the power consumption of it. It's just a bit of a waste to run a big chip at very low clocks and power.



numberwang said:

Prediction time:

Switch 2 mobile:
Max 10W for the whole system, max 7W for the SOC
GPU 3xx MHz
~1 TFP
Heavy 3rd party games will run at 360p DLSS-uped to 1080p


Switch 2 docked:
Max 13W whole system, 12W for SOC
GPU 7xx MHz
1.6 TFP
3rd party games: 540p DLSS-uped to 1440p (maybe the last step from 1440p to 4K using a less intensive method)

Compare 11W and 9W max for Switch 1 when docked and mobile. We will need slightly better cooling this time.
https://www.anandtech.com/show/11181/a-look-at-nintendo-switch-power-consumption/2

LOL this exactly what i expected, it's the Nintendo way



Around the Network

The large die size doesn't make sense for economic reasons rather than performance ones.  

Larger die -> more defects in the manufacturing process-> fewer chips per wafer -> higher long-term manufacturing costs (even if the wafer costs less, the chips won't necessarily, due to the lower yield rate.)

That is without considering that the yield rate of 4/5nm is about 30% more per wafer than 7/8nm, for the same die size.

Nintendo and Nvidia aren't idiots. They're not going to pay more money for a weaker chip. 



sc94597 said:

The large die size doesn't make sense for economic reasons rather than performance ones.  

Larger die -> more defects in the manufacturing process-> fewer chips per wafer -> higher long-term manufacturing costs (even if the wafer costs less, the chips won't necessarily, due to the lower yield rate.)

That is without considering that the yield rate of 4/5nm is about 30% more per wafer than 7/8nm, for the same die size.

Nintendo and Nvidia aren't idiots. They're not going to pay more money for a weaker chip. 

You have no idea what deal they made or when it was made.



zeldaring said:
sc94597 said:

The large die size doesn't make sense for economic reasons rather than performance ones.  

Larger die -> more defects in the manufacturing process-> fewer chips per wafer -> higher long-term manufacturing costs (even if the wafer costs less, the chips won't necessarily, due to the lower yield rate.)

That is without considering that the yield rate of 4/5nm is about 30% more per wafer than 7/8nm, for the same die size.

Nintendo and Nvidia aren't idiots. They're not going to pay more money for a weaker chip. 

You have no idea what deal they made or when it was made.

Deal with whom? Nvidia isn't going to lose money paying too much to TSMC (or Samsung) if they don't have to.

Your delusion that Nintendo, a very conservative company when it comes to risk-assessment and investments, and which has gone through cyclic sales shocks, is stockpiling hardware for years is much more speculative. Especially, as you noted, there was a chip shortage between 2020-2023 (which is likely why they canceled the Switch Pro.) 

Edit: Plus there would be evidence of this in the financial reports of all of the companies involved.

Last edited by sc94597 - on 14 September 2023

sc94597 said:
zeldaring said:

You have no idea what deal they made or when it was made.

Deal with whom? Nvidia isn't going to lose money paying too much to TSMC (or Samsung) if they don't have to.

Your delusion that Nintendo, a very conservative company when it comes to risk-assessment and investments, and which has gone through cyclic sales shocks, is stockpiling hardware for years is much more speculative. Especially, as you noted, there was a chip shortage between 2020-2023 (which is likely why they canceled the Switch Pro.) 

Edit: Plus there would be evidence of this in the financial reports of all of the companies involved.

Well they been saying they wanna make enough consoles where they won't have shortages. Nvda especially can't catch up because ai chip demand so it would make sense for them to make a deal and start manufacturing in 2021. this guy would not be confirming 8nm if he didn't have a reliable source and he's also ppointing all the things he got wrong but well have to wait and see, my bet is he is right.



zeldaring said:
sc94597 said:

Deal with whom? Nvidia isn't going to lose money paying too much to TSMC (or Samsung) if they don't have to.

Your delusion that Nintendo, a very conservative company when it comes to risk-assessment and investments, and which has gone through cyclic sales shocks, is stockpiling hardware for years is much more speculative. Especially, as you noted, there was a chip shortage between 2020-2023 (which is likely why they canceled the Switch Pro.) 

Edit: Plus there would be evidence of this in the financial reports of all of the companies involved.

Well they been saying they wanna make enough consoles where they won't have shortages. Nvda especially can't catch up because ai chip demand so it would make sense for them to make a deal and start manufacturing in 2021. this guy would not be confirming 8nm if he didn't have a reliable source and he's also ppointing all the things he got wrong but well have to wait and see, my bet is he is right.

That's not how this works. You don't guarantee  stock by overpaying during the peak of a silicon shortage (that was in 2021) three years before you plan to release the platform in question.

You wait to invest when prices drop and yield rates increase through a new node process. 

And if you are Nintendo, a company that has gone from selling 105 million consoles to 13 million and then back up to >130 million, you don't start to manufacture your next platform 3 years early when sales projections are far less precise. Again, especially during the peak of a silicon shortage.