DF - Nintendo Switch 2 Confirmed Specs: CPU, GPU, Memory, System Reservation + More

So is it confirmed Ampere then? Because there was a thread a few days back with folk claiming it was Ada
It is because of a chinese video where they bought a dev unit motherboard and xrayed the SOC. Its Ampere but its laid out like Ada. So people took that comment in the video and ran that it was some sort of hybrid gpu that was going to push the switch2 to absurd levels of performance.

Long story short it's Ampere, but taking the lessons Nvidia has learned making Ada, applied it to Nintendo's custom solution.
 
I hope hardware decompression is used by most. It can be hit or miss with 3rd parties in that regard for current gen consoles. Not sure how often direct storage is used by PC games.
 
It is because of a chinese video where they bought a dev unit motherboard and xrayed the SOC. Its Ampere but its laid out like Ada. So people took that comment in the video and ran that it was some sort of hybrid gpu that was going to push the switch2 to absurd levels of performance.

Long story short it's Ampere, but taking the lessons Nvidia has learned making Ada, applied it to Nintendo's custom solution.

Its SMs are 22% smaller than T234 "Ampere" too in the layout similar to Ada, so its really custom. Peoples were thinking before they were picking a T234 and just ripping out things but leaving SMs as is.

Ada in terms of raster pipeline is pretty much Ampere with better gate clocking for efficiency. Which is what T239 allegedly went for. Ada performs so good because of higher clocks, mainly allowed by TSMC 4nm node, but clock for clock it would be similar. RT got huge boost with cache, SER, RT new gen, etc, but that's not really what Switch 2 would be chasing or focus its design on.

I hope hardware decompression is used by most. It can be hit or miss with 3rd parties in that regard for current gen consoles. Not sure how often direct storage is used by PC games.

Since its hardware I would imagine that its not gonna be the situation like software solutions on PC with directX.
NVapi and the SDK would most likely automate this.
 
Last edited:
Its SMs are 22% smaller than T234 "Ampere" too in the layout similar to Ada, so its really custom. Peoples were thinking before they were picking a T234 and just ripping out things but leaving SMs as is.

Ada in terms of raster pipeline is pretty much Ampere with better gate clocking for efficiency. Which is what T239 allegedly went for. Ada performs so good because of higher clocks, mainly allowed by TSMC 4nm node, but clock for clock it would be similar. RT got huge boost with cache, SER, RT new gen, etc, but that's not really what Switch 2 would be chasing or focus its design on.



Since its hardware I would imagine that its not gonna be the situation like software solutions on PC with directX.
NVapi and the SDK would most likely automate this.
Well I was just commenting on the comment context of on going conversations I have seen on here since the leaks of that motherboard were posted in early April. As per the OG quote, some takeawys from lurking threads was it was an ADA gpu in a handheld. A lot of comments in those threads were like, "4+ Tflops and DLSS3.2 with Ada gpu! OMG swoon"
The gymnastic leaps were getting absurd. I think I saw a couple that were like this will chase PS5 on game quality etc etc.
I believe it was around the time that DF was posting their weekly chat video about re-adjusting the pixel counts for Cyberpunk to be 360p-540p upscaled. At least around the same time frame.
 
Since its hardware I would imagine that its not gonna be the situation like software solutions on PC with directX.
NVapi and the SDK would most likely automate this.

PS5, Series S/X, and PCs all have hardware decompression (if you have a pc with modern hardware). None of it's automated on any of them. Not ever the PS5 with it's dev friendly SDK.
 
If it's not there it's a BIG miss for Nintendo. No VRR on 40fps games is a recipe for disaster.
Typically 40fps mode is capped and locked at 40fps/120Hz for even frame pacing. There's no need for VRR. Some games with 40fps modes will let you uncap the frame rate in that mode and use VRR. IIRC R&C Rift Apart and Hogwarts can do this. Seems like that would make for inconsistent input latency, but I've never tested it.
 
It's really not and these insane notions will die rapidly as soon as launch day. The first thing is that the switch 2 is weak af. It's officially weaker than the steam deck Oled but with access to DLSS and RT cores. The first thing is ampere flops need to be divided by 2 due to the dual issue flopflation trick that delivered almost 0 performance in games.

That leaves the switch 2 at just 1.5tfs when docked with memory bandwidth only matching the steam deck Oled at 102gb/s. When in handheld mode, things are significantly worse. With reduced memory bandwidth and reduced performance, it's so far behind pc handhelds. Embarrassing for a system releasing in 2025 and I say that as someone who has a switch 2 Mario kart bundle preorder at $700 CAD. Well behind the series S, behind the ps4 and realistically around the Xbox one S but with a better storage architecture and a dedicated file decompression block.

Honestly the switch 2 prices become more ridiculous the more you examine it. The king of misers are at it again with one of the worst unambitious systems ever released. This should have been on tsmc 5nm or at least match the 2020 next gen consoles at tsmc 7nm. At least then, it would likely run at higher clocks with better battery life….
Can you explain how the 3080 is around 30% faster than the 2080 Ti, despite having the same number of SMs and almost the exact same clockspeed? (So the 2080 Ti is essentially half the TFs)

It's also funny that Turing was faster than RDNA 2 on a per teraflop basis (~8.6 TF 2070 competes with ~9.7 TF 5700 XT), RDNA 2 is 1.25X GCN (Polaris) and GCN (Polaris) was ~1.1X GCN (Hawaii). So even a 1.5 TF Turing part would be expected to be significantly faster than the base PS4. You're taking DF's "flopflation" narrative, and completely ignoring their actual point that different architectures perform differently at a given TF level.

And the bandwidth is now suddenly a big problem, when it keeps the same ratio as the Ampere PC parts, and when we have already seen the Rog Ally scale significantly beyond the Steam Deck, despite having the same 102 GB/s.

Edit:

I mean, we already know the switch 2 is weaker than the rtx 2050 mobile. DF already confirmed it and the specs were leaked out so we know it's a derivative. We know how that level of hardware performs. The only ones who are surprised are nintendo clowns.
Yes, DF's video showed a cut down 2050 mobile with 3 TF doing significantly better than the last gen consoles, and somewhat closing the gap to the Series S with DLSS enabled.
 
Last edited:
Typically 40fps mode is capped and locked at 40fps/120Hz for even frame pacing. There's no need for VRR. Some games with 40fps modes will let you uncap the frame rate in that mode and use VRR. IIRC R&C Rift Apart and Hogwarts can do this. Seems like that would make for inconsistent input latency, but I've never tested it.
With LFC you could allow games to drop to 28 or in Xbox's case to 20 FPS for a few moments and avoid stuttering (although it is not a silver bullet).
I think that Nintendo cheapened out on HDMI 2.0 and the dock so we should not really make excuses for that trying to justify the lack of VRR where it would be most useful.
 
Ada in terms of raster pipeline is pretty much Ampere with better gate clocking for efficiency. Which is what T239 allegedly went for. Ada performs so good because of higher clocks, mainly allowed by TSMC 4nm node, but clock for clock it would be similar. RT got huge boost with cache, SER, RT new gen, etc, but that's not really what Switch 2 would be chasing or focus its design on.
Yep, that is what they focused on, but why? Is the OLED or micro LED update and maybe an interim revision with lower power consumption something they want to sell and thus holding back to a design apparently taped out in 2021 worth it?

I do not expect Nintendo to chase path tracing on mobile but the RT advancements from Ada to me mean that the HW they have there has more chance to be used if you can extract more performance per Watt.
Similarly, going from a modified 10 nm process (Samsung's process is between 10nm and 8nm for many of its characteristics) to TSMC 4nm would mean somewhat lower power consumption and higher frequency (which as you said Ada was optimised to reach) and considerably lower power consumption (thus longer battery life).

Maybe and Ada based 4nm SoC being smaller enough might be cheaper to manufacture too (then again the Samsung process might not be in huge demand). I do not think it is unfair to state Nintendo should have asked for an Ada based semi-custom design. Switch 2 is a design that will be around for almost a decade.
 
Last edited:
Smartphones have the actual oposite approach in their hardware design, they have way higher clocks because they need to boost tasks to quickly idle and save battery, contrary to consoles they are not asked to sustain high performance for hours or else they'd be using low power/low clocks mode, just like handheld PCs if I'm not mistaken.
That's true, but 1GHZ seems very low. Docked mode could really benefit from the higher clocks and wouldn't have to factor in heat and battery consumption the same way.
 
So 2.1 to 2.3 KiloGameCubes

I vote for 0.0021-0.0023 "MegaCubes".

---

Society when we reach One MegaCube:

1*n2GqadKH-755OSaLSR3k3Q.jpeg
 
Last edited:
You all got distracted with the TF talk. What about the mystery of the GRs? Is Switch 2 a path tracing machine or someone forgot to carry the one.
 
That's true, but 1GHZ seems very low. Docked mode could really benefit from the higher clocks and wouldn't have to factor in heat and battery consumption the same way.
Wouldn't that affect game logic tho? That's why they kept the same frequencies in both modes in Switch 1 afaik
 
With LFC you could allow games to drop to 28 or in Xbox's case to 20 FPS for a few moments and avoid stuttering (although it is not a silver bullet).
I think that Nintendo cheapened out on HDMI 2.0 and the dock so we should not really make excuses for that trying to justify the lack of VRR where it would be most useful.
I don't disagree, but 2 things can be true at the same time. 40fps modes don't typically rely on VRR, and Nintendo should have done better with the dock.
 
Wouldn't that affect game logic tho? That's why they kept the same frequencies in both modes in Switch 1 afaik
Yes, that's why they use CPU boost speeds only at loadings, In some cases game logics like physics are tethered to the CPU frequency on consoles… DF did a good video explaining this for the Switch 1.
 
Nintendo *seems* to have listened to us out there who wanted better performance on many of the Switch games and it's a decent power bump. I'm happy.

We'll have plenty of time for the eventual Super Switch 2(Pro version) and OLED version.
 
Yes, that's why they use CPU boost speeds only at loadings, In some cases game logics like physics are tethered to the CPU frequency on consoles… DF did a good video explaining this for the Switch 1.

uh... we aren't living in 1995 anymore. this isn't really a thing in any modern game and can easily be demonstrated to not be an issue by overclocking the Switch 1.

the real reasons the clock speeds are the same are 1: power draw in handheld mode, 2 heat related issues in docked mode, and 3: consistency between modes.
the first two make sense, the third one doesn't really.
if they are limiting it due to the third reason, I think they want development to be simpler with (practically) only 1 CPU setting. but especially on Switch 2 a higher CPU clock in docked mode could open up the possibility of 40/60fps modes that aren't possible in handheld mode due to CPU limits.

so I hope the limit is in place for one of the first two reasons... because the third one would be misguided imo.
 
uh... we aren't living in 1995 anymore. this isn't really a thing in any modern game and can easily be demonstrated to not be an issue by overclocking the Switch 1.

the real reasons the clock speeds are the same are 1: power draw in handheld mode, 2 heat related issues in docked mode, and 3: consistency between modes.
the first two make sense, the third one doesn't really.
if they are limiting it due to the third reason, I think they want development to be simpler with (practically) only 1 CPU setting. but especially on Switch 2 a higher CPU clock in docked mode could open up the possibility of 40/60fps modes that aren't possible in handheld mode due to CPU limits.

so I hope the limit is in place for one of the first two reasons... because the third one would be misguided imo.
*Some games… I heard this from DF a long time ago.
 
*Some games… I heard this from DF a long time ago.

sure, but if some very niche case like this would happen, probably with some weird proprietary japanese engine, there's alway the possibility of offering different modes to Devs, like they already do with the GPU modes anyway
 
Last edited:
sure, but if some very niche case like this would happen, probably with some weird proprietary japanese engine, there's alway the possibility of offering different modes to Devs, like they already do with the GPU modes anyway
I think that's going to happen eventually.
 
Nintendo *seems* to have listened to us out there who wanted better performance on many of the Switch games and it's a decent power bump. I'm happy.

We'll have plenty of time for the eventual Super Switch 2(Pro version) and OLED version.
My wish is for Nintendo and third parties go back and spruce up Switch titles to play even better on Switch 2 or even natively if possible.
 
It's really not and these insane notions will die rapidly as soon as launch day. The first thing is that the switch 2 is weak af. It's officially weaker than the steam deck Oled but with access to DLSS and RT cores. The first thing is ampere flops need to be divided by 2 due to the dual issue flopflation trick that delivered almost 0 performance in games.

That leaves the switch 2 at just 1.5tfs when docked with memory bandwidth only matching the steam deck Oled at 102gb/s. When in handheld mode, things are significantly worse. With reduced memory bandwidth and reduced performance, it's so far behind pc handhelds. Embarrassing for a system releasing in 2025 and I say that as someone who has a switch 2 Mario kart bundle preorder at $700 CAD. Well behind the series S, behind the ps4 and realistically around the Xbox one S but with a better storage architecture and a dedicated file decompression block.

Honestly the switch 2 prices become more ridiculous the more you examine it. The king of misers are at it again with one of the worst unambitious systems ever released. This should have been on tsmc 5nm or at least match the 2020 next gen consoles at tsmc 7nm. At least then, it would likely run at higher clocks with better battery life….
it's weak, embarassing, ridiculous and you preordered it. what are you?
 
it's weak, embarassing, ridiculous and you preordered it. what are you?
Curious. I plan to run a lot of tests on the machine, punish a review, etc. All of these will earn me a small amount of money to offset the cost of the device.

You thought you had something there…. Sorry to rain on your parade.
 
Top Bottom