Switch 2 Speculation

Scanline effect on 3DS is due 2x horizontal pixel density in 2D mode. I have TN screens.

Then it's different to what I refer to as it's only visible in IPS screen.

Btw do you have a switch? It's IPS panel also have scanlines but very few people able to see it.

Its most visible on the clouds in zelda botw and around link

I suspect Nintendo will keep having whatever screen cheapest and have blemishes that only visible to small amounts of population, for switch 2.

But I hope by the time Nintendo have switch 2, the cheapest screen available already good. No weird blemishes.

Nowadays if they use cheapest oled, it could have weird tint and or flicker in lowest brightness, black smear, weird color balance... Among other things (been a few months since I hunt cheap tablet / phone with oled screen).

But I think they will go with RGB oled so they could have the best resolution x performance.
 
Main reason for xavier would be DLSS + ray tracing. Developers could simply take the pc version of the games and port them to the switch 2. At 720p you'd have decent ray tracing performance , i'd think on par with the consoles doing 4k or maybe even 1440p and you can use dlss to upscale to native screen res.

There's no RT hardware in Xavier, it's a Volta GPU.
Also, I think only a small portion of Xavier is dedicated to CPU and GPU, as the chip has an enormous focus on IO and video decoding for dozens of cameras. It's way too focused on automotive and industrial edge computing.


The first Switch gathered enough of a market to justify ordering an actually custom SoC, but Nintendo being Nintendo I wonder if they'll pay up nvidia prices for such a chip.
 
Btw would be pretty interesting if switch 2 gonna be whole new soc that's incompatible with switch 1, but Nintendo put some of switch 1 hardware inside switch 2 to offer BC

Then use soc from mediatek...

But seriously, Nintendo have a good track record for BC, so switch 2 should be able to play switch 1. Maybe that's why the rumor was switch 4k.

For first year or two, there will be no exclusive switch 2 games. But switch games that runs in switch 2 will be rendered at higher resolution and or quality, performance.

Or it'll be like new nintendo 3ds and only 1 or 2 retail games comes exclusive for the new hardware for the whole lifespan of the console...
 
Btw would be pretty interesting if switch 2 gonna be whole new soc that's incompatible with switch 1, but Nintendo put some of switch 1 hardware inside switch 2 to offer BC

Then use soc from mediatek...

If Nintendo ever dropped nvidia for their next SoC then my bet would be for Samsung+RDN.
Though unless they're cutting off backwards compatibility, I don't think nvidia would let them emulate their Maxwell GPU anytime soon.
 
The first Switch gathered enough of a market to justify ordering an actually custom SoC, but Nintendo being Nintendo I wonder if they'll pay up nvidia prices for such a chip.

Yep, but I don't see Nvidia requiring a bunch of cash to reconfigure an existing SOC to meet the needs of a Switch successor. Stripping down the Xavier chip for example wouldnt require the same budget as a truly custom from the ground up SOC. Xavier stripped of the extra automotive hardware would probably fit the bill. Nintendo is the only partner Nvidia has left in this space. So while Nvidia may not be positioned to offer the best performance per dollar, that may not matter to Nintendo, the support from Nvidia on SDK would be more important. Nintendo may not source high end parts for their hardware, but that doesnt mean they don't spend money on R&D. When you look at what Nintendo paid IBM and AMD for the Wii and Wii U processors, forking out some cash to Nvidia to strip down the Xavier chip seems very reasonable to me.
 
When you look at what Nintendo paid IBM and AMD for the Wii and Wii U processors, forking out some cash to Nvidia to strip down the Xavier chip seems very reasonable to me.
As answered above, nvidia is likely going to ask for a lot of money nonetheless, as the company positions itself as the premium maker of GPUs.

TX1 was the exception here, as it had been an ultimately failed chip from 2015, by the time it entered the Switch 2 years later.
 
As answered above, nvidia is likely going to ask for a lot of money nonetheless, as the company positions itself as the premium maker of GPUs.

TX1 was the exception here, as it had been an ultimately failed chip from 2015, by the time it entered the Switch 2 years later.

You could be right, but if they push for too much, they run the risk of Nintendo moving on from them. It seems foolish to potentially screw up a deal where tens of millions, or even a hundred million plus processors could be sold. I know Nvidia is a pretty proud company, but I think AMD dominating the console space hurt their ego a bit, and thats why they went looking for a partner. Time will tell, but if I were a betting man, I would bet Nvidia and Nintendo stay partners for at least one more Nintendo console.
 
Problem is nvidia is doing the switch api too. I guess not having nvidia for S2 could complicate the BC ?

If not I hope they go PowerVR Series B :eek:
 
If they're using an older Nvidia SOC, how would that compare to the latest SnapDragon or some other SOC?

Obviously SnapDragon may be too costly, unless Qualcomm would rip out the baseband and give them a lower price but still the best GPU performance available in the SD line.
 
Problem is nvidia is doing the switch api too. I guess not having nvidia for S2 could complicate the BC ?

If not I hope they go PowerVR Series B :eek:

That's an interesting question. ARM is now owned by Nvidia, at least hypothetically (regulators still deciding approval right?) so oops on CPU compatibility, which is more important than GPU compatibility most often. Not that rich enough companies care, Apple can emulate x86 all they want as they proved. But Nintendo is in the business of making money while erring on the side of spending as little as they can. Thus Nvidia has the challenge of wringing as much money out of Nintendo as possible while cutting just short of whatever line would make more sense for Nintendo to go and do something more expensive for BC while going to, assumedly, AMD.

That being said, the IP licensing of the ARM ISA is weird. I could take a stab that there'd be a way around it by going to a licensee I suppose. Heck maybe another integrated SOC makre, it would make sense for battery life and such. ARM GPUs are more power efficient, if of course less scaleable, than Nvidia's own.
 
That's an interesting question. ARM is now owned by Nvidia, at least hypothetically (regulators still deciding approval right?) so oops on CPU compatibility, which is more important than GPU compatibility most often. Not that rich enough companies care, Apple can emulate x86 all they want as they proved. But Nintendo is in the business of making money while erring on the side of spending as little as they can. Thus Nvidia has the challenge of wringing as much money out of Nintendo as possible while cutting just short of whatever line would make more sense for Nintendo to go and do something more expensive for BC while going to, assumedly, AMD.

Having a compatible GPU architecture is just as important as having a compatible CPU architecture for the purposes of backwards compatibility ...

NVN isn't going to work on just any Nvidia hardware. It's specifically designed to expose raw 2nd gen Maxwell shaders and the fixed function pipeline as well so there's virtually no chance that NVN will be compatible with Volta, Turing, & etc ...

At best, what one could hope for in retaining backwards compatibility while adding new features is designing a hypothetical "3rd gen" Maxwell where they'll have to potentially reimplement things like ray tracing by 'extending' their older Maxwell architecture to include new functionality instead of basing their initial design around more recent architectures which is already built-in their hardware ...

I don't imagine that Nvidia will be all that interested in pushing out a new generation of Maxwell to just appease one customer's requirement for backwards compatibility ...
 
NVN isn't going to work on just any Nvidia hardware. It's specifically designed to expose raw 2nd gen Maxwell shaders and the fixed function pipeline as well so there's virtually no chance that NVN will be compatible with Volta, Turing, & etc ...

Assuming you are correct, and the Switch API would not be compatible moving to something like Volta, then sticking with Maxwell/Pascal is certainly possible. Combine 512 Cuda cores running at 1.5 Ghz with four A7* Arm CPU cores and a 8GB of memory fitted to a 128bit memory bus with whatever the most current LPDDR memory is certainly an option. Is there anything stopping Nvidia from being able to implement some Tensor cores so they can use DLSS? You would end up with a console in the form factor of the Switch that approaches PS4 levels of performance but has the benefit of DLSS.
 
Having a compatible GPU architecture is just as important as having a compatible CPU architecture for the purposes of backwards compatibility ...

NVN isn't going to work on just any Nvidia hardware. It's specifically designed to expose raw 2nd gen Maxwell shaders and the fixed function pipeline as well so there's virtually no chance that NVN will be compatible with Volta, Turing, & etc ...

At best, what one could hope for in retaining backwards compatibility while adding new features is designing a hypothetical "3rd gen" Maxwell where they'll have to potentially reimplement things like ray tracing by 'extending' their older Maxwell architecture to include new functionality instead of basing their initial design around more recent architectures which is already built-in their hardware ...

I don't imagine that Nvidia will be all that interested in pushing out a new generation of Maxwell to just appease one customer's requirement for backwards compatibility ...

Emulating other ISAs is generally far, faaar more taxing than emulating GPU architectures. If no enhancements to the old Switch games were needed (no docked mode in portable mode for example) then it's fully concievable that a 2021 mobile GPU could emulate a 2015 mobile GPU under the right circumstances. The GPU often needs specific instruction emulation tricks to run things correctly over speed. EG you can run some games like Bayonetta 2 perfectly on the current, reverse engineered Switch emulator Yuzu on an Mx250, a rather underpowered mobile GPU that a reasonable Switch successor should probably best.

But as stated, once the CPU ISA is native it shouldn't be a problem. Which is why I don't see GPU BC as nearly as concerning.
 
Assuming you are correct, and the Switch API would not be compatible moving to something like Volta, then sticking with Maxwell/Pascal is certainly possible. Combine 512 Cuda cores running at 1.5 Ghz with four A7* Arm CPU cores and a 8GB of memory fitted to a 128bit memory bus with whatever the most current LPDDR memory is certainly an option. Is there anything stopping Nvidia from being able to implement some Tensor cores so they can use DLSS? You would end up with a console in the form factor of the Switch that approaches PS4 levels of performance but has the benefit of DLSS.

There's theoretically nothing stopping Nvidia from reimplementing tensor cores on the 2nd gen Maxwell architecture or any of the other new functionality ...

Emulating other ISAs is generally far, faaar more taxing than emulating GPU architectures. If no enhancements to the old Switch games were needed (no docked mode in portable mode for example) then it's fully concievable that a 2021 mobile GPU could emulate a 2015 mobile GPU under the right circumstances. The GPU often needs specific instruction emulation tricks to run things correctly over speed. EG you can run some games like Bayonetta 2 perfectly on the current, reverse engineered Switch emulator Yuzu on an Mx250, a rather underpowered mobile GPU that a reasonable Switch successor should probably best.

But as stated, once the CPU ISA is native it shouldn't be a problem. Which is why I don't see GPU BC as nearly as concerning.

Technically, GPUs do have ISAs. Just because they don't run your standard C++ or Python programs like we see on CPUs doesn't mean they don't have instruction sets. I think you underestimate how much overhead emulation poses. Not even a GTX 1650 can emulate NVN perfectly since some Switch games are still seeing slowdowns on hardware far more powerful than it is and we have yet to achieve pixel correct accuracy which will add more overhead to emulation but the worst part here is that these are results which are using proprietary Nvidia API extensions so that even low level access on the same hardware vendor isn't going to make emulation much easier on a console ...

GPUs are just as important for backwards compatibility if not maybe more so since they have massive black boxes that is the fixed function pipeline which can very easily change in both it's functionality or behaviour from generation to generation and they can also be removed overtime as well further complicating emulation ...

Emulation as it is isn't all that realistic of a solution when you take look at the current best implementation's software compatibility. Many games on the Switch can't be completed even with the best emulator ...
 
There's no RT hardware in Xavier, it's a Volta GPU.
Also, I think only a small portion of Xavier is dedicated to CPU and GPU, as the chip has an enormous focus on IO and video decoding for dozens of cameras. It's way too focused on automotive and industrial edge computing.


The first Switch gathered enough of a market to justify ordering an actually custom SoC, but Nintendo being Nintendo I wonder if they'll pay up nvidia prices for such a chip.

Wasn't aware that Xavier didn't have RT hardware , I thought it was tensor tech for the deep learning part.

Then I hope they wait and go with an RT solution. I think they can easily stay with the switch as it is until 2022. So it could be possible.


The first switch justifies them using another off the shelf SoC in the second one. It worked really well for them so why invest a ton of money ?

In all seriousness who are they competing with ? There is no other dedicated portable from any large player. You have that windows portable that is $700 bucks and then a bunch of emulator devices that are sub $100 which are great for what they are but aren't competing with new games. So if you were Nintendo why would you dump money into something custom ?

If I was nintendo I would want the ability to play all switch games at 1080p 60fps which should be easy for them , maybe 1440p in docked mode if they do docks again. I'd want to get in 4-5 hours of battery life as that seems to be the sweet spot for most people. I would also think they'd want to support raytracing and dlss.

Like I've said the xbox series s is a great system for Nintendo. If they can create a mobile system that can simply run the series s games with the same features at 720p using dlss to 1080p or even 1440p in handheld they will be fine. Heck looking at control and dlss they could even go to 540p and scale to 1080p or 1440p just fine.

That is what is what I'd love to see out of a 2022/23 switch 2.

But to me I wont buy a switch 2 if my switch games don't carry over. At that point I'd be hoping for a xbox series s portable instead.
 
Wasn't aware that Xavier didn't have RT hardware , I thought it was tensor tech for the deep learning part.

Then I hope they wait and go with an RT solution. I think they can easily stay with the switch as it is until 2022. So it could be possible.


The first switch justifies them using another off the shelf SoC in the second one. It worked really well for them so why invest a ton of money ?

In all seriousness who are they competing with ? There is no other dedicated portable from any large player. You have that windows portable that is $700 bucks and then a bunch of emulator devices that are sub $100 which are great for what they are but aren't competing with new games. So if you were Nintendo why would you dump money into something custom ?

If I was nintendo I would want the ability to play all switch games at 1080p 60fps which should be easy for them , maybe 1440p in docked mode if they do docks again. I'd want to get in 4-5 hours of battery life as that seems to be the sweet spot for most people. I would also think they'd want to support raytracing and dlss.

Like I've said the xbox series s is a great system for Nintendo. If they can create a mobile system that can simply run the series s games with the same features at 720p using dlss to 1080p or even 1440p in handheld they will be fine. Heck looking at control and dlss they could even go to 540p and scale to 1080p or 1440p just fine.

That is what is what I'd love to see out of a 2022/23 switch 2.

But to me I wont buy a switch 2 if my switch games don't carry over. At that point I'd be hoping for a xbox series s portable instead.

Useable hardware RT in a mobile isn't happening anytime soon. At today's gpu efficiencies it'll be a good showing if it hits an xbox one's performance.

As for emulation... yes, cpu emulation is hard. See Rosetta 2's performance versus Microsoft's own x86 emulation.

But in terms of performance, gpu emulation isn't hard. You need detailed knowledge of the gpu to emulate some of its behaviors, which Nintendo might have. Which is why I see gpu emulation for bc as possible. What I don't see is how saying yuzu's failure to run many games at all, not due to performance issues in any way but due to emulation crashes thanks to unreplicated behaviors, is an indication that gpu emulation is impossible. It's a volunteer built, reverse engineered, free piece of software and they're still making progress.

The bigger issue I see is Nintendo itself. You have to practically force tech down their throats as they go so unwillingly. So going with and underpowered Nvidia gpu, fully modern, that Nvidia does all the bc stuff for themselves, seems the most likely option. AMD did the same thing for the PS5/XS and RDNA2 is hardly a [somewhat modified Sea Islands], but BC is close to one hundred percent on both consoles.
 
Useable hardware RT in a mobile isn't happening anytime soon. At today's gpu efficiencies it'll be a good showing if it hits an xbox one's performance.

As for emulation... yes, cpu emulation is hard. See Rosetta 2's performance versus Microsoft's own x86 emulation.

But in terms of performance, gpu emulation isn't hard. You need detailed knowledge of the gpu to emulate some of its behaviors, which Nintendo might have. Which is why I see gpu emulation for bc as possible. What I don't see is how saying yuzu's failure to run many games at all, not due to performance issues in any way but due to emulation crashes thanks to unreplicated behaviors, is an indication that gpu emulation is impossible. It's a volunteer built, reverse engineered, free piece of software and they're still making progress.

The bigger issue I see is Nintendo itself. You have to practically force tech down their throats as they go so unwillingly. So going with and underpowered Nvidia gpu, fully modern, that Nvidia does all the bc stuff for themselves, seems the most likely option. AMD did the same thing for the PS5/XS and RDNA2 is hardly a [somewhat modified Sea Islands], but BC is close to one hundred percent on both consoles.

nintendo is going to be awfully embarrassed if they show up to the mobile party without RT
 
Back
Top