Switch 2 Speculation

Think people expecting much more than similar raw performance to a 7-10W TDP steam deck in its handheld mode (SoC power not total system power, excludes screen etc) is setting themselves up for disappointment. It's likely still going to be a relatively small, relatively cheap device (if it's >$400USD I'd be surprised) from a company that likes to make a profit on system sales or with a game/accessory, using <10W total system power in handheld mode that needs hours of battery life. The switch dock itself is limited to 18W too although the max in "normal" conditions in games for most situations is going to be less, would Nintendo want a bigger, bulkier device with more cooling that'd cost more? Would Nintendo spring for a more advanced node to get more performance/efficiency over saving money on a good generational improvement? They don't seem like "Nintendo" things to do to me, they could've gone for Parker which would've been more efficient but didn't. Even with DLSS integrated, it's not going to be magic. It will last and age better than contemporary handheld PCs that's for sure (Steam deck, ROG ally etc) but expecting a big generational leap over their raw specs/perf at similar power budgets is unlikely (Ampere being RDNA2 contemporary and all).

Simplest thing - imagine what a 5-15W Nvidia mobile SoC from about 2 years ago would be capable of, what level of performance do you have in mind? Great because that's what it is, expect that. Better than PS4 fidelity in docked mode although not massively so, DLSS'd to 4k with bits of RT sprinkled on top for a little flourish with handheld mode being scaled down appropriately. That's the most realistic expectation to me, given the physical and financial limitations
 
just imagine what a closed hardware with fully controlled DLSS could do for a future handheld. PS5 performance level on a handheld?

"Gaming at Ultra Low Resolutions with DLSS - 240p and beyond".


That all looked pretty good on my 11 inch tablet. Perhaps what you really want is to frame cap it to a 30/60 fps and then see how dlss fares.
 
"It's a customized version of a 5 (6 by the time it releases) year old SOC!" rumor again.

Orin is indeed the last time Nvidia put out a mobile SOC. And I've gotta admit, that sounds Nintendo as hell, and a perfectly Nintendo reason to stick with Samsung 8nm. Not because it makes any financial sense for a new SOC, but simply because Nintendo would once again not bother with a new SOC at all.

Of course this 100% contradicts the "Matrix Awakens!" supposed leaks from Gamescon. But let's face it, it's Nintendo, that sounds questionable to begin with.
 
"It's a customized version of a 5 (6 by the time it releases) year old SOC!" rumor again.

Orin is indeed the last time Nvidia put out a mobile SOC. And I've gotta admit, that sounds Nintendo as hell, and a perfectly Nintendo reason to stick with Samsung 8nm. Not because it makes any financial sense for a new SOC, but simply because Nintendo would once again not bother with a new SOC at all.

Of course this 100% contradicts the "Matrix Awakens!" supposed leaks from Gamescon. But let's face it, it's Nintendo, that sounds questionable to begin with.
Yeah, we get it. Don't buy into the hype. Nintendo will do as Nintendo does.

We are just setting ourselves up for disappointment 😂

Although my son would sure like Minecraft and Fortnite not taking forever to load.
 
"It's a customized version of a 5 (6 by the time it releases) year old SOC!" rumor again.

Orin is indeed the last time Nvidia put out a mobile SOC. And I've gotta admit, that sounds Nintendo as hell, and a perfectly Nintendo reason to stick with Samsung 8nm. Not because it makes any financial sense for a new SOC, but simply because Nintendo would once again not bother with a new SOC at all.

Of course this 100% contradicts the "Matrix Awakens!" supposed leaks from Gamescon. But let's face it, it's Nintendo, that sounds questionable to begin with.

How's it contradictory? The original leak only mentioned vague terms about comparable visuals to current gen with some advanced effects. No concrete details about framerate, resolution, etc. Demo could have easily been running sub 540p/30 fps with DLSS upscaling and most gaming journalists wouldn't have been able to tell the difference. I wouldn't put any stock into rumors that don't have hard numbers with them.

Samsung 8nm is probably the most likely process because it will be the cheapest with very high yields for a small mobile chip. And frankly at the low voltage and low frequency that I think Nintendo will run, I wouldn't expect that big of a difference between nodes. Orin Nano/Nx has a TDP right where Nintendo would target and still deliver 3-4x perf over Switch with more modern features.
 
Yeah, we get it. Don't buy into the hype. Nintendo will do as Nintendo does.

We are just setting ourselves up for disappointment 😂

Although my son would sure like Minecraft and Fortnite not taking forever to load.
I mean, SSDs and flash memory in general are cheap enough now even for Nintendo (heck I just bought a 980pro 1tb for $30) so maybe there's hope on that front

The supposed LinkedIn leak mentioning 2.6ghz clockspeed isn't Ampere at all. No Samsung 8nm ever ran anywhere close to 2.6ghz, that Ada and TSMC only clockspeeds. In fact the 4050 is basically exactly that clockspeeds. Which also take up way too much power, the 4050 draws 150w at those clockspeeds. Meaning this T239 wouldn't be in a 10w part at all.

So these "leaks" don't seem to add up to anything coherent at the moment.
 
Last edited:
Taking the various T239 leaks at face value (I don't see why not at this point), I think there are still a few interesting questions. Chief among them for me is storage - will Nintendo stick with MicroSD, perhaps limiting compatibility to high speed cards (120MB/s reads to 170 MB/s reads perhaps)? Or will all storage be internal with multiple SKUs? Perhaps needing to support low-cost cartridges limits the ability of games to access fast storage... or maybe cartridges will require an install to the internal storage. I have no particular insight on this point, I just think it is a big question mark at the moment that has big implications for how the system will function.

Beyond any speculation about the T239 and its level of performance, I think the broader Nintendo-Nvidia technical partnership is probably a bigger deal. Nintendo has been very resistant to adopting TAA in their titles, mostly to their detriment in my view. DLSS being available on the system level will improve image quality so, so much, as long as titles can present the necessary inputs. And it's entirely possible that future Nvidia technologies will also find their way onto the console as time goes on.

As far as the game leaks go, it's very hard to evaluate without seeing those games in action. The demos could have various cutbacks and concessions that make them a lot more realistic than they seem on the surface.
 
Taking the various T239 leaks at face value (I don't see why not at this point), I think there are still a few interesting questions. Chief among them for me is storage - will Nintendo stick with MicroSD, perhaps limiting compatibility to high speed cards (120MB/s reads to 170 MB/s reads perhaps)? Or will all storage be internal with multiple SKUs? Perhaps needing to support low-cost cartridges limits the ability of games to access fast storage... or maybe cartridges will require an install to the internal storage. I have no particular insight on this point, I just think it is a big question mark at the moment that has big implications for how the system will function.

Beyond any speculation about the T239 and its level of performance, I think the broader Nintendo-Nvidia technical partnership is probably a bigger deal. Nintendo has been very resistant to adopting TAA in their titles, mostly to their detriment in my view. DLSS being available on the system level will improve image quality so, so much, as long as titles can present the necessary inputs. And it's entirely possible that future Nvidia technologies will also find their way onto the console as time goes on.

As far as the game leaks go, it's very hard to evaluate without seeing those games in action. The demos could have various cutbacks and concessions that make them a lot more realistic than they seem on the surface.
I think would leaked specs irs 4.3 tflops or 5 at full power.. it should do 1440p native or with dlss.. xbox one x performance
 
Taking the various T239 leaks at face value (I don't see why not at this point), I think there are still a few interesting questions. Chief among them for me is storage - will Nintendo stick with MicroSD, perhaps limiting compatibility to high speed cards (120MB/s reads to 170 MB/s reads perhaps)? Or will all storage be internal with multiple SKUs? Perhaps needing to support low-cost cartridges limits the ability of games to access fast storage... or maybe cartridges will require an install to the internal storage. I have no particular insight on this point, I just think it is a big question mark at the moment that has big implications for how the system will function.
What I think Nintendo will do:
  • Keep using Micro SD cards - allows them to sell their own branded/partnered SD cards for crazy prices on their own store which a handful of people will buy. They like accessory sales like everyone else
  • 128GB/256GB eMMC storage because it's cheap, one SKU unless they're digital only. Hoping for UFS 2.x because it's a lot faster but is more expensive, Nintendo thinking "why spend more for the same capacity" is on brand
  • Carts could go either way. They're relatively expensive and slow, but do they have the infrastructure to go digital only? Gut instinct says no, maybe they use a higher capacity, up to 128GB cart with minor improvements
If they did go digital only it could give them the chance to use multiple SKUs which would be increased margins, more chance to sell SD cards, possibly raise online prices and save on physical sales which would be big but again, "it's Nintendo"
 
  • 128GB/256GB eMMC storage because it's cheap, one SKU unless they're digital only. Hoping for UFS 2.x because it's a lot faster but is more expensive, Nintendo thinking "why spend more for the same capacity" is on brand
according to some insiders, dev kits have 512GB of storage. as far as I can find, 512GB of emmc doesn't exist and emmc is dying rapidly now that practically every phone uses UFS
 
according to some insiders, dev kits have 512GB of storage. as far as I can find, 512GB of emmc doesn't exist and emmc is dying rapidly now that practically every phone uses UFS

Yeah it'll be UFS, fortunately that's cheap. 512gb is also the high end of UFS, so sure why not.

Note: DLSS isn't magic for Nintendo titles either, they'd still have to implement motion vectors to begin with, which they don't, so... I don't know, it's Nintendo. Go ask them what it is they're doing.

Now if T239 were an Ada based chip. Let's say it's 768 cores in mobile config. Clockspeed @1.2ghz for mobile, maybe 2.2ghz for docked. Around the Steamdeck, with better CPU for mobile, maybe @12w. Docked I can see it running The Matrix Awakens heavily upscaled, say 540p to 1080p. But that's a bit of a useless demo, games would need to run in mobile mode.
 
Note: DLSS isn't magic for Nintendo titles either, they'd still have to implement motion vectors to begin with, which they don't, so... I don't know, it's Nintendo. Go ask them what it is they're doing.

I wondered about this at first when Ada was released for PC, whether Nvidia might consider a DLSS mode that could be applied to older titles, specifically ones without motion vectors.

The OFA block is programmable and can be used in multiple modes, with like a 10:1 spread in frame time cost, depending on how much accuracy you need. It seems to work well enough even in what has to be one of the 'faster' modes to generate an intermediate frame for FG. On older 'legacy' Switch titles that would be a cake walk on the new GPU, why not burn a bunch more frame time on letting the OFA block run wild in one of its most accurate modes, and then just DLSS the whole business up to native res of the internal display, or 4k in docked mode? UI and other items might be problematic, but since Nintendo and NVidia know their API in and out, who knows, with a few title specific patches they might be able to inject it at just the right spot in the pipeline to mitigate the worst of the artifacts.

Edit: doesn't the driver level DLDSR sort of do this already? I should double check to see what its limitations are.
 
I wondered about this at first when Ada was released for PC, whether Nvidia might consider a DLSS mode that could be applied to older titles, specifically ones without motion vectors.

The OFA block is programmable and can be used in multiple modes, with like a 10:1 spread in frame time cost, depending on how much accuracy you need. It seems to work well enough even in what has to be one of the 'faster' modes to generate an intermediate frame for FG. On older 'legacy' Switch titles that would be a cake walk on the new GPU, why not burn a bunch more frame time on letting the OFA block run wild in one of its most accurate modes, and then just DLSS the whole business up to native res of the internal display, or 4k in docked mode? UI and other items might be problematic, but since Nintendo and NVidia know their API in and out, who knows, with a few title specific patches they might be able to inject it at just the right spot in the pipeline to mitigate the worst of the artifacts.

Edit: doesn't the driver level DLDSR sort of do this already? I should double check to see what its limitations are.

I feel there's a strategic consideration here on the PC side given the still relative infancy stage of the technology in terms adoption.

A driver and software agnostic implementation will always be at a significant quality/performance deficit over an implementation with software support (eg. with the supplied motion vectors). But the problem here is we know the software side overall will likely prefer the path of least resistance essentially just to tick features, which would as a by product hamper software side DLSS adoption.

On the user side there's also the similarly an issue with association. There's a risk here of the user (and I guess media coverage) then associating DLSS with the lower quality/less performant driver implementation versus the better software supported implementation. Especially given Nvidia's overarching marketing strategy is they do same to favor cultivating a "quality" vs "quantity" approach in terms of perception.

Edit: doesn't the driver level DLDSR sort of do this already? I should double check to see what its limitations are.

I'm not sure if Nvidia's ever gone deep into the specifics of what DLDSR is actually doing. But from what I remember (think?) it's not reconstructing the image in the same sense DLSS does. Rather it replaces the gaussian filter step in DSR with a ML approach and doesn't require any temporal data?
 
USF would be good vs SDD for power consumption and probably cost?

Carts/SD cards being able to keep up with internal storage might be an issue. Insist on installs from carts and offer a USF card storage expansion.

Like Xbox's SSD cards but even poorer value. ;)
 
Back
Top