StarTopic Future Nintendo Hardware & Technology Speculation & Discussion |ST| (Read the staff posts before commenting!)

[Ghostsonplanets]

i see,Breath of the Wild on Switch sucessor is possibly demoed at 900p upcalled at 4K, is not a native 4K, no way a handheld/hybrid console is able to run in this resolution

900p is already what it runs on current Switch...
BoTW Demo was probably running with DLSS 4K. But the base resolution was probably raised significantly. Besides, we don't know if Nintendo applied more bell and whistles to the demo version. We already know it featured instant loading.

 

[Giancarlo]

If you haven't watched the DF video, go ahead and do it now. Don't worry about tech stuff if you're not sure what they mean, you should still get some ideas from looking at visuals alone.

Keep in mind that is the floor of what Switch 2 can do (meaning it's most likely going to be better than what you see in DF video).

If anyone were disappointed, it's probably because they set themselves up with unrealistic expectations. Switch 2 because of its docked/handheld form factor, isn't meant to be competitive with PS5 and XSX's raw power.

if you expecting a hybrid console could be more powerful then the current gen console, Switch sucessor if i understand for my lack of techinal/graphical perfomance understanding, Switch sucessor will be a handheld PS4 with a few modern wrinkles

 

[Hermii]

900p is already what it runs on current Switch...
BoTW Demo was probably running with DLSS 4K. But the base resolution was probably raised significantly. Besides, we don't know if Nintendo applied more bell and whistles to the demo version. We already know it featured instant loading.

I mean if it was running 4k at all, it's more likely to be without dlss imo. Or some trickery like having fsr take the image from the dlss resolution to 4k. I don't see dlss on Drake being fast enough for 4K 60 to make sense.

 

[Giancarlo]

dont you all feel, a sort of depression feeling, now that Nintendo Switch is approaching it end, dont you all hoped Nintendo Switch would live until 2030 and then Switch sucessor, everywhere i go, i see people with this trough, why Nintendo is killing Switch now, cant they keep it alive for 5 more years

 

[darthdiablo]

dont you all hoped Nintendo Switch would live until 2030

I actually don't, not that long. Switch (like every other Nintendo console I've owned) will hold a special place in my memory banks, but I'm ready for something new and exciting now.

 

[Hermii]

dont you all feel, a sort of depression feeling, now that Nintendo Switch is approaching it end, dont you all hoped Nintendo Switch would live until 2030 and then Switch sucessor, everywhere i go, i see people with this trough, why Nintendo is killing Switch now, cant they keep it alive for 5 more years

You Are talking to the wrong community lol.

 

[Kenka]

900p is already what it runs on current Switch...
BoTW Demo was probably running with DLSS 4K. But the base resolution was probably raised significantly. Besides, we don't know if Nintendo applied more bell and whistles to the demo version. We already know it featured instant loading.

Well, frame times are a function of the output resolution. So if the RTX 2050 takes 18 ms to output a 4K frame, it would take always this time no matter the input material.

If the TotK demo was indeed running at 4K 60 FPS using DLSS then it can only mean that the tensor cores take less time to go through the upscale workload than the downclocked RTX 2050.

Significantly so I should add.

 

[Ghostsonplanets]

I mean if it was running 4k at all, it's more likely to be without dlss imo. Or some trickery like having fsr take the image from the dlss resolution to 4k. I don't see dlss on Drake being fast enough for 4K 60 to make sense.

Natively it's impossible. It would need a ~12x perforrmance jump to go from 810-900p30 to 4K60. If we take DF DLSS frametime costs at facevalue, it is indeed impossible. But I'm not really keen into taking that numbers as it seems too strange.

That being said, if we assume DF DLSS 4K frametime costs are right, then a 1440p DLSS + simple upscaling/spatial pass to 4K is feasible, yes.

Well, frame times are a function of the output resolution. So if the RTX 2050 takes 18 ms to output a 4K frame, it would take always this time no matter the input material.

If the TotK demo was indeed running at 4K 60 FPS using DLSS then it can only mean that the tensor cores take less time to go through the upscale workload than the downclocked RTX 2050.

Significantly so I should add.

Right, but as I said, that's if we take DF numbers at facevalue. Their frametime cost seems weird and I would really want some more testing.

 

[ItWasMeantToBe19]

DLSS frametimes for 720p>1440p for like the 3070 (based on NVIDIA documentation and nsight timing) have the tensor core process resolving in like 1 ms so I'm not sure what is happening with these DF results unless there's multiple steps that occur on the CUDA cores afterwards that NVIDIA doesn't mention.

(though the 3070 has 4x the tensor cores and likely 1.5x the clock speed of the Switch 2 so that would lead to 6-8 ms of DLSS frametime if it's a linear decline so maybe idk)

(Frame deferring makes sense on paper, hopefully it works in practice)

 

[Stinky Horse]

There's various post processing steps (game dependent) that need to be done after the DLSS upscale. Naturally the cost of those will scale with your pixel count.

I mean if it was running 4k at all, it's more likely to be without dlss imo. Or some trickery like having fsr take the image from the dlss resolution to 4k. I don't see dlss on Drake being fast enough for 4K 60 to make sense.

Yeah I don't trust the reporting on that. It COULD be using a variety AI upscaling as an alternative to TotK's FSR 1, however.


Related to that I'm actually curious how good a 1080p DLSS screen could look through a 4K AI upscaling pass!

 

[Deleted member 3433]

dont you all feel, a sort of depression feeling, now that Nintendo Switch is approaching it end, dont you all hoped Nintendo Switch would live until 2030 and then Switch sucessor, everywhere i go, i see people with this trough, why Nintendo is killing Switch now, cant they keep it alive for 5 more years

There's a bittersweet feeling here, knowing that one of my favorite platforms of all time has its best days behind, but that is kinda balanced by optimism and knowledge about its successor. Assuming that backwards compatibility is a thing, there's always the games that are still with us.

 

[Adulfzen]

dont you all feel, a sort of depression feeling, now that Nintendo Switch is approaching it end, dont you all hoped Nintendo Switch would live until 2030 and then Switch sucessor, everywhere i go, i see people with this trough, why Nintendo is killing Switch now, cant they keep it alive for 5 more years

From Nintendo themselves I could see software support going until 2027 tho it'd get more and more limited as time went on while by 2030 I'd expect only third party developers to really bother especially if that means they can just count on whatever backwards compatible solution Nintendo came up with.

Nintendo might enforce a policy where past a certain date, developers have to make sure their software runs well on Switch 2 hardware even if it's just through backwards compability, Sony did something of the sort for PS5.

 

[lemonfresh]

i see,Breath of the Wild on Switch sucessor is possibly demoed at 900p upcalled at 4K, is not a native 4K, no way a handheld/hybrid console is able to run in this resolution

My pc is not that powerful its like over 5 years old and I can run botw at 4k and it runs pretty good

 

[Apenknul]

There's a bittersweet feeling here, knowing that one of my favorite platforms of all time has its best days behind, but that is kinda balanced by optimism and knowledge about its successor. Assuming that backwards compatibility is a thing, there's always the games that are still with us.

This, but also, how would Nintendo keeping the Switch alive, if the sales aren't booming as much as some years back?

 

[gamerguy]

Natively it's impossible. It would need a ~12x perforrmance jump to go from 810-900p30 to 4K60. If we take DF DLSS frametime costs at facevalue, it is indeed impossible. But I'm not really keen into taking that numbers as it seems too strange.

That being said, if we assume DF DLSS 4K frametime costs are right, then a 1440p DLSS + simple upscaling/spatial pass to 4K is feasible, yes.

Right, but as I said, that's if we take DF numbers at facevalue. Their frametime cost seems weird and I would really want some more testing.

DF are so close to 16ms in that test, it seems silly to think nvidia didn't do the math, or design the chip to reach that resolution.

750mHz was as low as Rich could go, but he also ran with that speed to equalize the flops between the two chips. The 2050 has more cores, and he was after parity.

Do we know what those clocks do to the tensor cores?... maybe the math works out, but it's possible that at 1gHz, or whatever the Switch2's docked clock speed is, the tensor cores can do the maths in 16ms.

I doubt nvidia designed the chip to get that close to 4k dlss, then threw their hands up and walked away...

...it might give some insight into the docked frequency...

 

[Giancarlo]

From Nintendo themselves I could see software support going until 2027 tho it'd get more and more limited as time went on while by 2030 I'd expect only third party developers to really bother especially if that means they can just count on whatever backwards compatible solution Nintendo came up with.

Nintendo might enforce a policy where past a certain date, developers have to make sure their software runs well on Switch 2 hardware even if it's just through backwards compability, Sony did something of the sort for PS5.

well i expect at least 4/5 more years of suport for Switch, in 2029 been the first year fully focused on Switch sucessor, kinda like Sony did with PS5

 

[Kise Ryota]

Do we know what those clocks do to the tensor cores?... maybe the math works out, but it's possible that at 1gHz, or whatever the Switch2's docked clock speed is, the tensor cores can do the maths in 16ms.

I don't know how it works exactly, but at 1GHz switch 2 would have 33% more clock, but the 2050 has 33% more tensor cores. So I don't know if that just equals everything.

 

[Giancarlo]

This, but also, how would Nintendo keeping the Switch alive, if the sales aren't booming as much as some years back?

easy release all the games that was planned on it next hardware to Switch instead, new 3D Mario, new Animal Crossing, new Mario Kart all for Switch, is a big maybe if Switch sucessor actually launch next year

 

[Ghostsonplanets]

Do we know what those clocks do to the tensor cores?... maybe the math works out, but it's possible that at 1gHz, or whatever the Switch2's docked clock speed is, the tensor cores can do the maths in 16ms.

Higher clocks = faster math calculations, assuming no other bottleneck. But DF tested with the RTX 2050, which has 33% higher amount of Tensor Cores (64 x 48)....

If only DF did more tests. A second round of this video would be very welcome.

 

[Deleted member 3433]

This, but also, how would Nintendo keeping the Switch alive, if the sales aren't booming as much as some years back?

Yeah, sales have been slowing down for quite some time now. Nintendo had to move on, and so do we.

 

[Ghostsonplanets]

Yeah, sales have been slowing down for quite some time now. Nintendo had to move on, and so do we.

More important than hardware sales, specially for an already massive seller device, is active userbase. If users are still active and buying content, it's still worth to release games on it. But if the Switch userbase quickly moves on, like 3DS one did, then the cross-gen period will be over quickly.

 

[Adulfzen]

well i expect at least 4/5 more years of suport for Switch, in 2029 been the first year fully focused on Switch sucessor, kinda like Sony did with PS5

again depends how you define the support, I could see NSO updates still happening in 2030 even but full on games, even ports and smaller remasters seem less likely past a certain point.

 

[Giancarlo]

again depends how you define the support, I could see NSO updates still happening in 2030 even but full on games, even ports and smaller remasters seem less likely past a certain point.

i define suport by games releases, the more games released on Switch(both first-party/third party), the better for Nintendo/us

 

[JoshuaJSlone]

dont you all feel, a sort of depression feeling, now that Nintendo Switch is approaching it end, dont you all hoped Nintendo Switch would live until 2030 and then Switch sucessor, everywhere i go, i see people with this trough, why Nintendo is killing Switch now, cant they keep it alive for 5 more years

Oh god no.

 

[LiC]

Just my brain raising the alarm due to experience with previous gens Nintendo HW threads I'm not saying it's impossible though! But as i said, i'd happily eat that crow.

Previous gen hardware threads never had leaks directly from the chipmaker, public source code commits from the chipmaker, or shipping records from the chipmaker and a factory. We have all of those things and they all show the same thing.

 

[robertman2]

dont you all feel, a sort of depression feeling, now that Nintendo Switch is approaching it end, dont you all hoped Nintendo Switch would live until 2030 and then Switch sucessor, everywhere i go, i see people with this trough, why Nintendo is killing Switch now, cant they keep it alive for 5 more years

Me to the Switch 1 when 2 drops

 

[gamerguy]

I don't know how it works exactly, but at 1GHz switch 2 would have 33% more clock, but the 2050 has 33% more tensor cores. So I don't know if that just equals everything.

ahh... maybe it kinda equals out then

it still might give insight to frequency though, yeah?... how much faster would those cores need to run to gain that 1.3ms

it's possible 750mHz was just too slow to test the capabilities of the Switch 2... from what I can tell in the video, that frequency was used because he couldn't go lower

but what flop number was he trying to match?... power efficiency matters less in docked mode, and we might also be looking at higher flops too

 

[Machrider]

For me it's not whether or not its strictly better than the Steam Deck that mattered. It's the fact that I was hoping the third party games would be mostly 60fps 1080p in handheld mode and instead it's looking like the more likely result for third party games is going to be 1080p 30fps in handheld even with the benefits of DLSS.

No offense but 1080p 60 fps for modern 3rd party games running off battery power is kind of a wild ask. The PS5 barely runs some of these 3rd party games at 60 fps/1440p.

 

[ItWasMeantToBe19]

These tests do raise some significant questions about the viability of Ray Reconstruction on the Switch 2 to add on to prior questions about RR.

RR seemingly works by selecting rays from frames n-1, n-2, ..., that can be reused for frame n.

Things we don't know about RR currently

-How much additional frametime does it cost relative to DLSS itself?
-Where in the pipeline does RR happen? Does it happen at the same time as Super Resolution or at a different time? What GPU work is still needed after RR happens? This may not work well with deferred frames if significant GPU work has to be done after RR.
-Is RR compatible with denoisers other than ReBLUR or ReLAX? ReLAX is 1000% not viable on Switch 2 and ReBLUR seems unlikely to work well on Switch 2.

 

[Kingpin Rogers]

dont you all feel, a sort of depression feeling, now that Nintendo Switch is approaching it end, dont you all hoped Nintendo Switch would live until 2030 and then Switch sucessor, everywhere i go, i see people with this trough, why Nintendo is killing Switch now, cant they keep it alive for 5 more years

That would be my own personal hell. I’ve been ready for the Switch 2 since 2021. The sooner we move on from Switch 1 the better.

 

[Kise Ryota]

but what flop number was he trying to match?

Rich was trying to match a possible docked mode at 1GHz. It's a safe number to bet IMO. Switch 2 in docked mode could have 1.1GHz too, which isn't a big diference of course, or it could reach 1.33GHz, which would be at the same raw performance of Rich's 1GHz seen here
But it's really hard to try to guess Switch 2's clocks while docked. We could miss it by a 33% margin

 

[Ghostsonplanets]

Oh god no.

Oh shoot, Switch is the longest generation without new hardware from Nintendo, isn't.

but what flop number was he trying to match?... power efficiency matters less in docked mode, and we might also be looking at higher flops too

3 TFLOPs. Which is basically the same TFLOPs as T239/Switch 2 Docked running at 1GHz. It's why he said that his testing was trying to be like to like of Switch 2 Docked potential performance.

 

[Kise Ryota]

After Rich's tests, what can we extrapolate for handheld mode in terms of DLSS? The base clock used for testing (750MHz) was a plausible docked mode at 1GHz.

Having, for example, half of that clock (375MHz), would DLSS quality or balanced still be ok for 60fps@1080p games on handheld? Or even 30fps

 

[gamerguy]

Oh shoot, Switch is the longest generation without new hardware from Nintendo, isn't.

3 TFLOPs. Which is basically the same TFLOPs as T239/Switch 2 Docked running at 1GHz. It's why he said that his testing was trying to be like to like of Switch 2 Docked potential performance.

So kind of an educated guess. That extra 1.3ms doesn't seem like a lot in that context.

Loved the video... just curious about the BotW rumors.

edit: you could almost reverse engineer the docked frequency by what clock the tensor cores would have to run to achieve 4k60

 

[pelusilla6]

Waiting for the vietnamese people to add September shippments, there will be more juicy stuff for sure

 

[ItWasMeantToBe19]

After Rich's tests, what can we extrapolate for handheld mode in terms of DLSS? The base clock used for testing (750MHz) was a plausible docked mode at 1GHz.

Having, for example, half of that clock, would be DLSS quality or balanced still ok for 60fps games on handheld?

It depends on how DLSS scales down and whether the DLSS step can happen on a deferred frame.

540p>1080p DLSS from the video (it was 720p>1080p in the video, but this doesn't matter for DLSS speed) was around 3.55 ms which by a naive doubling would be 7.1 ms. It would be possible to do 60 FPS here without adding on a frame of input lag, but it would eat up 42.6% of the frametime for every frame.

 

[Ghostsonplanets]

After Rich's tests, what can we extrapolate for handheld mode in terms of DLSS? The base clock used for testing (750MHz) was a plausible docked mode at 1GHz.

Having, for example, half of that clock, would be DLSS quality or balanced still ok for 60fps games on handheld?

Rich thinks, based on T234 power curve, that 540MHz is a viable clock candidate for HH mode. Which also match Thraktor expectations for T239 clock...on TSMC 4N.

Oldpuck said he ran some testings at 550MHz before an bios update locked his ability to set the clocks. Would be interesting to see if he was able to get some DLSS data from this limited testing.

 

[ILikeFeet]

These tests do raise some significant questions about the viability of Ray Reconstruction on the Switch 2 to add on to prior questions about RR.

RR seemingly works by selecting rays from frames n-1, n-2, ..., that can be reused for frame n.

Things we don't know about RR currently

-How much additional frametime does it cost relative to DLSS itself?
-Where in the pipeline does RR happen? Does it happen at the same time as Super Resolution or at a different time? What GPU work is still needed after RR happens? This may not work well with deferred frames if significant GPU work has to be done after RR.
-Is RR compatible with denoisers other than ReBLUR or ReLAX? ReLAX is 1000% not viable on Switch 2 and ReBLUR seems unlikely to work well on Switch 2.

• depends on how many denoisers you're using. if you're using few denoisers, then RR will be more expensive. if you're using many due to multiple RT effects, then RR will be as fast if not faster
• literally in the same spot as Super Resolution. the usual GPU work is needed afterwards like post processing
• RR is most likely just those two denoisers. if Nvidia wanted to add another they could
• what proof you have of ReLAX not working working. you're still making claims you're not backing up with evidence

 

[Thraktor]

I don't think displaced micromesh will really have much of a use. it feels like it adds too many steps to the process. devs can make a nanite-like solution with lods and then fill in the blanks with normal maps and not change much of their pipeline

If it were supported on Switch 2, it would be useful. It's solving a slightly different problem than nanite. Even with nanite or mesh shader based dynamic LOD solutions, the BVH used for ray tracing is typically much lower fidelity than the geometry used for rasterised steps, as ray tracing geometry isn't limited by the bottlenecks that nanite and mesh shaders are designed to overcome. It's limited by storage space, traversal time and building time, all of which are proportional to the size of the BVH, and including lots of fine detail in the BVH is impractical due to the size it can grow to.

Displacement micro meshes attempt to solve this problem by replacing fine geometry in the BVH with micro meshes. Instead of having a subtree with a few hundred triangles, you can replace it with a single displacement micro mesh which can be consumed and processed natively by the RT hardware. There's no real way you can replicate it on hardware that can only perform ray tracing against triangles in a BVH, other than just adding all the triangles to the BVH directly and taking the performance hit.

With only a relatively small percentage of PC GPUs (for now) supporting it, I'm not sure how much use it will get. My guess is that any PC games which do support it will simply disable it on non-Ada hardware. Just treat the triangles with displaced micro-meshes as flat triangles for the purposes of ray tracing. Which probably wouldn't be any different to the quality of ray tracing geometry that's currently in use.

 

[Skittzo]

Oh god no.

This graph gives me instant Elastomania flashbacks.


Kinda crazy how long it's really been though, yeah.

 

[ILikeFeet]

If it were supported on Switch 2, it would be useful. It's solving a slightly different problem than nanite. Even with nanite or mesh shader based dynamic LOD solutions, the BVH used for ray tracing is typically much lower fidelity than the geometry used for rasterised steps, as ray tracing geometry isn't limited by the bottlenecks that nanite and mesh shaders are designed to overcome. It's limited by storage space, traversal time and building time, all of which are proportional to the size of the BVH, and including lots of fine detail in the BVH is impractical due to the size it can grow to.

Displacement micro meshes attempt to solve this problem by replacing fine geometry in the BVH with micro meshes. Instead of having a subtree with a few hundred triangles, you can replace it with a single displacement micro mesh which can be consumed and processed natively by the RT hardware. There's no real way you can replicate it on hardware that can only perform ray tracing against triangles in a BVH, other than just adding all the triangles to the BVH directly and taking the performance hit.

With only a relatively small percentage of PC GPUs (for now) supporting it, I'm not sure how much use it will get. My guess is that any PC games which do support it will simply disable it on non-Ada hardware. Just treat the triangles with displaced micro-meshes as flat triangles for the purposes of ray tracing. Which probably wouldn't be any different to the quality of ray tracing geometry that's currently in use.

well, Nvidia already added compute fallback, so maybe it's not going to be as limited. even more, the compute fallback might run better if there's sufficient tessellation

 

[Hosermess]

* Hidden text: cannot be quoted. *

Now I'm just picturing them having their own chat groups filled with "they're so back"/"they're joever"/"holy shit they're on page 2023" posts that annoy some % of their coworkers.

 

[nyaako]

Oh shoot, Switch is the longest generation without new hardware from Nintendo, isn't.

Makes sense since they no longer make a second handheld device like Game Boy or DS to release along the Switch.

 

[Apenknul]

easy release all the games that was planned on it next hardware to Switch instead, new 3D Mario, new Animal Crossing, new Mario Kart all for Switch, is a big maybe if Switch sucessor actually launch next year

I hate to break it to ya, friend, but this industry has to be flowing like a creek.
The Switch is due to its 7th year, and if you keep the system as it is now, people aren't gonna be interested in it anymore, since they want something new. In 5 years, the tech of the Switch will be more outdated than it is now, let stand in 2030. Nobody would be interested in a new Animal Crossing or Mario Kart, since it wouldn't that fresh as it would on a completely new system. The system sold enourmously, but new games in 5 years won't do that. It'll only sell some games, but not many more systems, and that won't do much good for Nintendo. With outdated tech, big third party won't be interested in it anymore either, and by then Nintendo is all on it's own, a la Wii U.

They have to keep being fresh and improve themselves to be more interested by the public, third party's, investors and such things to keep the creek flowing. Because, if a creek ain't flowing, and keep being still, there won't be much live in it and can get infectious.
After some time, Nintendo and every system developer have to make a new and better water wheel for that flowing creek.

 

[Anatole]

• depends on how many denoisers you're using. if you're using few denoisers, then RR will be more expensive. if you're using many due to multiple RT effects, then RR will be as fast if not faster
• literally in the same spot as Super Resolution. the usual GPU work is needed afterwards like post processing
• RR is most likely just those two denoisers. if Nvidia wanted to add another they could
• what proof you have of ReLAX not working working. you're still making claims you're not backing up with evidence

Is the Cyberpunk modder you posted the only source for RR being ReLAX/ReBLUR plus a neural network for temporal rejection of rays? That just doesn’t seem credible to me. The claim in all of Nvidia’s messaging was that the neural network in RR is doing the actual simultaneous denoising via reconstruction and supersampling, like in the Intel paper. Using ReLAX or ReBLUR would be in direct contradiction to that.

 

[ItWasMeantToBe19]

• depends on how many denoisers you're using. if you're using few denoisers, then RR will be more expensive. if you're using many due to multiple RT effects, then RR will be as fast if not faster
• literally in the same spot as Super Resolution. the usual GPU work is needed afterwards like post processing
• RR is most likely just those two denoisers. if Nvidia wanted to add another they could
• what proof you have of ReLAX not working working. you're still making claims you're not backing up with evidence

ReLAX is just very heavy, whatever, I don't want to fight about this one.

My question is not about relative denoiser cost, but just total additional frametime cost over SR alone.

Because if SR is eating up 7.5 ms of frametime on the tensor cores, then RR eating up another 9.2 ms (or more) of frametime on the tensor cores would make RR completely unviable.

 

[ItWasMeantToBe19]

Is the Cyberpunk modder you posted the only source for RR being ReLAX/ReBLUR plus a neural network for temporal rejection of rays? That just doesn’t seem credible to me. The claim in all of Nvidia’s messaging was that the neural network in RR is doing the actual simultaneous denoising via reconstruction and supersampling, like in the Intel paper. Using ReLAX or ReBLUR would be in direct contradiction to that.

So I'm reading through a lot of ReSTIR papers and I think NVIDIA might be unwilling to spend resources training a neural network to denoise ReSTIR because ReSTIR introduces too much noise correlation they need to get rid of first (and this noise correlation would probably fuck up a neural network) and this is apparently a very hard problem.

But if the modder is correct, it does make NVIDIA's messaging extremely deceptive.

 

[darthdiablo]

Waiting for the vietnamese people to add September shippments, there will be more juicy stuff for sure

Me too bud. I’ve been checking everyday to see if new Vietnam data came in.

 

[Machrider]

I think the Switch 2 specs are quite solid, being able to run games like A Plague Tale Requiem and Cyberpunk 2077 at PS5 settings is no joke, the resolution I wish could be a little better in docked mode, but still.

I think a Switch 2 Pro could really help, and hopefully this time around with no COVID type break (and component shortages), Nintendo will consider a Switch 2 Pro.

It doesn't have "to split the userbase!", just run the same games with a little boost in frame rate and resolution, maybe ray tracing also turned on. It's really not that different from New 3DS to the OG 3DS, here we see Monster Hunter 4 running at like closer to 50 fps than 30 fps with higher resolution textures, that's a pretty huge boost for New 3DS over the original 3DS.



I think high end games becoming likely more widespread on Apple iPhone/iPad and Android devices in the next 4-5 years should push Nintendo to also stay on top of performance and not rest on the status quo so much. At some point you're probably also going to see Steam Deck 2.

 

[TomNookYankees]

I‘m talking raw performance. All these things listed are because DLSS helping performance.

• Cyberpunk is not that much higher than base PS4 performance WITH DLSS ENABLED.
• Fortnite runs at 30fps with RT WITH DLSS ENABLED.
• All the game listed are running using DLSS. Rich really should have tested those games WITHOUT the use of DLSS as well.

When Fortnite comes to Switch 2, I highly doubt epic would use RT for a 30fps performance again on Switch. It would be no RT at 60fps which is more or less the base PS4 version.

Did you not watch the video? Rich tests both with and without DLSS.