Switch 2 Speculation

The "8nm" thing appears to be more leaker bullshit, going from an actual looking leak of a LinkeIn profile whole thing is just an unsurprising Ada + whatever ARM cores on some sort of 4nm, don't know if it's Samsung or TSMC.

Regardless upscaling is already super prevalent down onto the Series S. So all DLSS is going to do is be somewhat better than current FSR in motion. It'll still be a "60fps on Series = 30fps on Switch 2" kinda deal, at least in mobile/undocked mode.

A complete redesign of the SMs would cost much more than the small bits the tensor cores need. It makes no sense to leave them out, especially if it's based on Orin. They will have the same Tensor Core ratio as the other GPUs of the hardware gen they choose. Nvidia can probably scale DLSS Inference requirements down. A bit worse quality, but less time needed. Especially for 30 FPS thats not a problem, 60 FPS and higher will be more of a problematic. but i can't imagine Nintendo targeting high framerates.

Even on the smallest GPU utilizing the first gen tensor cores(RTX2060), DLSS is perfectly viable with near enough the same cost in performance as any other RTX part. I think it shows that DLSS on its own is really not that demanding from the tensor cores.

And if you're choosing an affordable old process like Samsung 8nm, going with a larger die size in general for a bigger GPU should be viable.

I can see it all making sense. I still dont like the idea of it being on such an old process, but Nintendo doesn't tend to go too ambitious on hardware so definitely think it's possible.
Ok but what about RT cores? How many would it have? To dedicate that much die space to RT & Tensor cores on a mobile GPU seems kind of suspect. If we look at the steam deck’s apu, it only has 8 cu’s. I don’t think we can expect many sm’s in the switch 2 apu. If die space is dedicated to RT cores and sm’s, it comes at the expense of the base GPU performance no? Furthermore, does it make sense to put only a few RT cores on the switch GPU? It’ll basically useless in the grand scheme of things? Certainly it won’t be enough to do RT reflections/gi/path tracing?
 
Ok but what about RT cores? How many would it have? To dedicate that much die space to RT & Tensor cores on a mobile GPU seems kind of suspect. If we look at the steam deck’s apu, it only has 8 cu’s. I don’t think we can expect many sm’s in the switch 2 apu. If die space is dedicated to RT cores and sm’s, it comes at the expense of the base GPU performance no? Furthermore, does it make sense to put only a few RT cores on the switch GPU? It’ll basically useless in the grand scheme of things? Certainly it won’t be enough to do RT reflections/gi/path tracing?
12 RT cores. they are inherent to the architecture. removing them would cost more money and wouldn't net you any perceivable gains.

personally, I don't think this is 8nm. if they'd have to clock it so low, they wouldn't have made a 12SM device in the first place
 
12 RT cores. they are inherent to the architecture. removing them would cost more money and wouldn't net you any perceivable gains.

personally, I don't think this is 8nm. if they'd have to clock it so low, they wouldn't have made a 12SM device in the first place
Depends what the architecture is, I guess. 1650/1660 series Turing didn't have RT cores, and neither have any of their 100 series HPC parts. But yea, I do agree that if anything required a full rework of the architectural layout of the SM, it probably wouldn't be worth doing. Just keep it in there, and they could still boast about having 'hardware accelerated ray tracing' even though nobody would use it, much like with some recent phone processors.

As for it being 12SM's, we dont know that yet. Obviously we dont know anything yet really, but I think it's a mistake to expect Nintendo to care too much about pushing high performance hardware. Switch was 'barely' good enough, and they'd probably be fine with that yet again, even if it's a low clocked 12SM GPU.
 
Depends what the architecture is, I guess. 1650/1660 series Turing didn't have RT cores, and neither have any of their 100 series HPC parts. But yea, I do agree that if anything required a full rework of the architectural layout of the SM, it probably wouldn't be worth doing. Just keep it in there, and they could still boast about having 'hardware accelerated ray tracing' even though nobody would use it, much like with some recent phone processors.

As for it being 12SM's, we dont know that yet. Obviously we dont know anything yet really, but I think it's a mistake to expect Nintendo to care too much about pushing high performance hardware. Switch was 'barely' good enough, and they'd probably be fine with that yet again, even if it's a low clocked 12SM GPU.
the only way for it to not have 12SM is if Nintendo isn't using the T239. the Nvidia leak to the Nvidia Github commits show it's the T239. the insider leaks from Eurogamer and Video Game Chronicle point to the T239 by association as well.

Nintendo might not care about high performance, but the T239 could fit with that as well. as we see with the myriad of AMD handhelds, Nintendo can go larger and higher power if they really wanted to push the limit.
 
the only way for it to not have 12SM is if Nintendo isn't using the T239. the Nvidia leak to the Nvidia Github commits show it's the T239. the insider leaks from Eurogamer and Video Game Chronicle point to the T239 by association as well.

Nintendo might not care about high performance, but the T239 could fit with that as well. as we see with the myriad of AMD handhelds, Nintendo can go larger and higher power if they really wanted to push the limit.
There were a number of misleads as to what hardware the Switch 1 had as well. Just saying we dont know yet. After the successful partnership with Nvidia, it wouldn't be out of the question if Nintendo asked for a custom chip spec of some sort.

Not doubting the 12SM spec, just saying there's options. Could even be the T239 was being used as an earlier test chip or something. Most console manufacturers will use some kind of existing tech for basic early target testing.
 
Most console manufacturers will use some kind of existing tech for basic early target testing.
Nintendo is literally the only one who could have done anything like that. Sony nor Microsoft ever had target spec hardware or even close to before they had to decide what they want.
 
the only way for it to not have 12SM is if Nintendo isn't using the T239. the Nvidia leak to the Nvidia Github commits show it's the T239. the insider leaks from Eurogamer and Video Game Chronicle point to the T239 by association as well.

Nintendo might not care about high performance, but the T239 could fit with that as well. as we see with the myriad of AMD handhelds, Nintendo can go larger and higher power if they really wanted to push the limit.

The T239 story is still pretty strange and there must be more going on like a revised launch shedule for switch 2 or whatever.
T239, while based on orin, is a custom SoC never used anywhere else. First it was mentioned in 2021, the SoC should be available for mass production in 2022 or at latest early 2023. Developers could've received samples in 2022 also.

Now we're talking about late 2024 release. But why should Nvidia develop a custom SoC and then just put it aside for 1 year? This makes no sense, you use your development ressources with a project shedule which fits to the planned release date. T239 makes only sense if Switch 2 release was planned for 2023, but was shifted 1 year.

But if Switch 2 release was shifted, why did they do it?
Only economic reasons? Has the planned hardware not changed? There could also be an innovation from Nintendo which is late in development and forced Nintendo to postpone the Switch-Next 1 year.

In addition we had many Switch Pro rumours, which were also based on the T239 and the pro should've been canceled. But there shouldn't ve ever been plans for a switch pro, if T239 was planned for Switch-Next. For sure Nintendo didn't just take the T239 and planted it in a Switch-Next.

Things don't add up for me.
 
Last edited:
The T239 story is still pretty strange and there must be more going on like a revised launch shedule for switch 2 or whatever.
T239, while based on orin, is a custom SoC never used anywhere else. First it was mentioned in 2021, the SoC should be available for mass production in 2022 or at latest early 2023. Developers could've received samples in 2022 also.

Now we're talking about late 2024 release. But why should Nvidia develop a custom SoC and then just put it aside for 1 year? This makes no sense, you use your development ressources with a project shedule which fits to the planned release date. T239 makes only sense if Switch 2 release was planned for 2023, but was shifted 1 year.

But if Switch 2 release was shifted, why did they do it?
Only economic reasons? Has the planned hardware not changed? There could also be an innovation from Nintendo which is late in development and forced Nintendo to postpone the Switch-Next 1 year.

In addition we had many Switch Pro rumours, which were also based on the T239 and the pro should've been canceled. But there shouldn't ve ever been plans for a switch pro, if T239 was planned for Switch-Next. For sure Nintendo didn't just take the T239 and planted it in a Switch-Next.

Things don't add up for me.
the recent Xbox slides show the planned lifetime of a console-oriented chip design, showing (an expected) 4 years from design to release. if we assume the start was 2021 (or earlier), then this wouldn't be much different from other vendors

Next-Gen-Xbox-Development-Roadmap-2.jpg
 
Last rumor is release a year from now.
Two models: digital only for 400$, with card reader for 450$
That sounds quite expensive
 
the recent Xbox slides show the planned lifetime of a console-oriented chip design, showing (an expected) 4 years from design to release. if we assume the start was 2021 (or earlier), then this wouldn't be much different from other vendors

View attachment 9773
Tapeout is only about two years before launch, though. That's still a lot of time to work with the actual chip design beforehand, especially when they're often just kind of using off-the-shelf IP for the most part nowadays.
 
Last rumor is release a year from now.
Two models: digital only for 400$, with card reader for 450$
That sounds quite expensive
That's very expensive for a Nintendo product. But they might think they can capitalize on Switch's incredible popularity anyways. Thing is - Switch is already such an incredibly rich platform as is, so they are gonna have a harder time pushing out 'must have' games that really entice people to make the upgrade. And I'm not sure 3rd party AAA titles or anything will be enough for that, given you could buy a damn PS5 for the same price.

$50 to be able to carry over your Switch physical library and to buy physical games is still gonna be a no-brainer for many, though. And again brings up the question of what they're doing for storage and how they handle that situation with expansion and whatnot. Might be a little messy.
 
That's very expensive for a Nintendo product. But they might think they can capitalize on Switch's incredible popularity anyways. Thing is - Switch is already such an incredibly rich platform as is, so they are gonna have a harder time pushing out 'must have' games that really entice people to make the upgrade. And I'm not sure 3rd party AAA titles or anything will be enough for that, given you could buy a damn PS5 for the same price.

$50 to be able to carry over your Switch physical library and to buy physical games is still gonna be a no-brainer for many, though. And again brings up the question of what they're doing for storage and how they handle that situation with expansion and whatnot. Might be a little messy.

And they can't do the "cheap" trick like wiiu games ported to switch, as launch titles.

As switch is very popular unlike wiiu.
 
Nintendo is literally the only one who could have done anything like that. Sony nor Microsoft ever had target spec hardware or even close to before they had to decide what they want.

i think thats incorrect.
 
Rumor ...
In a new podcast shared today by Nate the Hate, who proved to be reliable in the past regarding Nintendo rumors, it was confirmed that the next Nintendo console will support NVIDIA DLSS 3.5 AI-powered Ray Reconstruction, which improves the quality of ray tracing considerably.

Speaking of the Nintendo Switch 2 feature set, Nate the Hate also commented on which feature that is part of NVIDIA DLSS will be supported by the console. While he says that he believes Frame Generation won't be supported, the insider considers the console extremely future-proof, as both Nintendo and NVIDIA are working to make the console as advanced as possible in the realm of mobile technology with the feature set that makes the most sense for the system.
...
While Nate the Hate proved to be reliable in the past, we have to take what was revealed today with a grain of salt. Still, given how positive the reports talking about how good the Matrix Awakens Unreal Engine 5 demo ran on the console, it's extremely likely the console will indeed support Ray Reconstruction.
 
Last edited by a moderator:
And then they will name it Nintendo Switch ON or something and people gets confused it's a new console or a light switch accessory for switch...

Seriously tho, with how fg needs 60fps+ to be looking good, seems it'll be pretty much useless for Nintendo. As they probably cheapening on the screen with just 60hz screen
 
And then they will name it Nintendo Switch ON or something and people gets confused it's a new console or a light switch accessory for switch...

Seriously tho, with how fg needs 60fps+ to be looking good, seems it'll be pretty much useless for Nintendo. As they probably cheapening on the screen with just 60hz screen
Most Nintendo games are already 60FPS 🤷🏼‍♂️
 
Back
Top