Nintendo Switch 2 specs: 1080p 120Hz display, 4K dock, mouse mode, and more

[-]

NutsackEuphoria@reddit

How would the hardware compare to the Steamdeck's?

Is it slightly better or slightly worse?

[-]

Out of a smaller battery 5200MAh and likely 3.7V, yeah. Not compared to the old switch, but the steam deck (similar MAh rating but 7.7v).

[-]

nmkd@reddit

You're confusing efficiency and "good battery life".

I don't care if it would run off of a single AAA if it only lasts 1h.

[-]

Vince789@reddit

If the rumors of 8x A78C CPU are true, then it should have a decently more powerful CPU (A78 is roughly on par with Zen2 in IPC, and the Steam Deck only has 4x)

Unless it's heavily power constrained due to 8nm & focusing on battery life

[-]

marcost2@reddit

They will need to be heavily power constrained, if at all used.

For example, when using an Orin AGX with the 15w power profile the GPU is downclocked to 420Mhz, you lose 8 cores (the only publicly available table is for the BIG BOY orin agx with 12 cores) and they are limited to 1.1Ghz

Orin wasn't designed with low power in mind, in contrast with the X1

I do wonder what power profiles we will see in handheld/docked mode, if in docked mode it goes up to 40W then it would be slightly more powerful/on par with the deck

[-]

theQuandary@reddit

All-core is 1.1GHz, but max frequency for T234 is listed as 2.2GHz.

[-]

marcost2@reddit

A frequency which they only reach at 60W(40W for the 32gb version) according to their power plan modes. Now, could nintendo do load balancing and load-frequency scaling on a per-core basis? Yeah, they could, but it would be a big undertaking considering how simplistic the frequency scaling is on the Switch 1.

[-]

theQuandary@reddit

A78 isn't some unknown CPU design. We've seen LOTS of chips with it and have a VERY good idea about what it can do.

T234 uses A78AE. It's designed for critical systems and is very conservative about everything and allows a lot of extra redundancy (and lockstepping cores) which increase power consumption and reduces performance. Looking around at other manufacturers, it seems like none of them go over 2.2GHz with the a78ae leading me to believe that's the max allowed for certification.

https://www.arm.com/products/silicon-ip-cpu/cortex-a/cortex-a78ae

T239 uses A78c. This core allows higher max clockspeeds (3.3GHz), 8mb of L3 instead of 4mb, 8 big cores instead of 4 big + 4 little, and backported some 8.3-8.6 security features that Nintendo no doubt wants to make jailbreaking harder.

Nintendo announced Civilization 7 coming to the console. If you look around for Civ 6 reviews on Switch, they generally don't care for the experience being so slow. the entire simulation genre (among others) is very affected by CPU performance.

t seems reasonable to allow devs to downclock the GPU and boost clock a core (or maybe a core complex) to a higher clockspeed on request.

[-]

marcost2@reddit

Yes, we also have specific frequency curves straight from nvidia for different power consumptions.
Also, you keep mentioning a die shrink but your only evidence for it is "it's smaller", despite the fact that even going 5nm would require entirely retooling for a questionable density increase, and some power savings (also the fact that 5LPP/5LPE is insanely more expensive than 8NM which is dirt cheap by now)

On that front, the A78AE only has a limit of 4 per cluster, which does not mean that it has "little cores", as per Nvidia's diagram they just have more clusters instead. Now this has the disadvantage of splitting up the L3 (only 2mb per core, versus the maximum 4mb per cluster) but it allows them to shut down the entire cluster if it's not being used. On the other hand the A78C might be slightly more compact (Only things duplicated for lockstep seems to be DSU locks and bridges which aren't that big and are quite dense) does have the possibility of having full access to all the L3 cache, which we still have no idea what size it is (as it can have as 512kb per ARM spec)

The A78C does however have fewer clock domains (if i'm reading this RTL right, which i might not be, i'm a computer scientist with electronical engineering experience) which would actually increase idle power consumption.

I agree, it sounds reasonable, my comment was mostly due to the fact that on the original switch nintendo exposed very little granularity, with them boosting all the cores at once and having very little in the way of frequency steps (this might have been fixed on the 16nm revision, i didn't touch the device after helping port android to it). To that point, the X1 was supposed to have a 2.2Ghz max clock on their A57 cores, but the switch never saw more than 1Ghz (outside of the "boost" mode, which only increased it to 1.7Ghz, that was only added via a firmware update)

[-]

theQuandary@reddit

Also, you keep mentioning a die shrink but your only evidence for it is "it's smaller",

https://imgur.com/W4ohTUz

https://imgur.com/a8vrnHJ

One of those is the leaked PCB and the other is the RAM dimensions for comparison.

You aren't putting a 455mm2 chip there. You aren't putting a 350mm2 chip there either. You could eliminate everything except the GPU, CPU, and DRAM and still only barely be capable of fitting it inside that space (but it's a SoC and needs all that other stuff).

despite the fact that even going 5nm would require entirely retooling for a questionable density increase

Switching from 8nm (actually more like TSMC 10nm process) to 5nm (somewhere between TSMC N6 and N5) is a 2x increase in max transistor density. That's a big change.

also the fact that 5LPP/5LPE is insanely more expensive than 8NM which is dirt cheap by now

Samsung 8nm is only slightly more dense than TSMC 10nm and is rumored to cost around $5000/wafer. Samsung 5LPE was rumored to cost $11-12k/wafer 4 years ago and has almost certainly dropped in price as N7 and N5 availability has opened up.

Doubling the cost per wafer doesn't double the cost of the chip if you're getting 70% more chips per wafer AND getting way better yields due to the chip being smaller AND getting lower power consumption too.

as it can have as 512kb per ARM spec

AMD shows just how valuable that L3 cache is for performance. 6 cores with 6mb of L3 would almost certainly be better for gaming, area, and power consumption compared to 8 cores with 4mb of L3. The interesting question to me is whether they went with 256kb or 512kb of L2. If they did a high-performance and low-performance split, they may have actually done both.

The A78C does however have fewer clock domains which would actually increase idle power consumption.

This isn't necessarily true. A78AE no doubt need more varied clocks because of their need to synchronize and the weird timing effects that can have. In that case, A78AE could have worse power consumption because of all the extra wiring and flipflops needed.

A78AE no doubt has terrible latency between those clusters which A78C shouldn't have. Running cores in lockstep adds all kinds of complexities throughout the core and especially in the last pipeline stages where everything has to be validated and synched.

the original switch nintendo exposed very little granularity, with them boosting all the cores at once and having very little in the way of frequency steps

As I recall, Tegra 4 had the same issue with the 4 P-cores locked together and only the 5th low-power core being independently clockable. That may have cut it for 4 cores, but just isn't going to work with 8 cores because most games can't use all those cores and want to turn them off most of the time.

Unfortunately, none of the Tegra line were particularly great products. X1 seemed like a stopgap product while they were pivoting their Transmeta stuff from x86 to ARM only for them to be fairly meh designs too. It's now 10 years later and Nvidia seems to be moving harder into CPUs. I'd guess things have gotten better as they couldn't get much worse.

[-]

marcost2@reddit

You aren't putting a 455mm2 chip there.

You can also keep that stuff and cut out some tensor/rt cores, the ISP unit, the DLA units, the PVA units, the ethernet controller. You know, all that stuff that's useful for the automotive dev board but not so useful for a portable console

is a 2x increase in max transistor density. That's a big change.

You are right, i got confused by the 6nm comment and was thinking about 7LPP which is "not great"

Samsung 8nm is only slightly more dense than TSMC 10nm

8LPP might have costed that much in 2018 but it sure as hell isn't costing that now, and 5LPE was having high defect issues well into 2022 so i don't know how well that scales (you can find some qualcomm investors press releases from the 888 era talking about the yield issues)

The interesting question to me is whether they went with 256kb or 512kb of L2. If they did a high-performance and low-performance split, they may have actually done both.

I absolutely agree, my comment was more as a response to the fact that the A78C can have 8MB of L3 cache. What ARM allows isn't what manufacturers follow, and with how poorly Samsung nodes have SRAM scaling, cutting on cache isn't out of the picture.

I would love to see a 512Kb of L2, but seeing how Orin had 256Kb i'm afraid to be hopeful (even in lower power chips, more cache is more better, waiting for data from dram is expensive on energy)

This isn't necessarily true.

My point was that it seemed from my RTL reading that the A78AE was able to clock gate multiple parts of its core independently to reduce power consumption opportunistically, and i'm not seeing the same logic in the A78C, now again, not a electronic engineer but.

A78AE no doubt has terrible latency between those clusters

It weirdly doesn't? It's mostly hidden by cache access latency and the crossbar design is very clever, like yeah in a vacuum the A78C might be faster if you can keep your test code inside the L3 at all times but if you are spilling into DRAM they are quite competitive with each other (clock frequency differences aside)

As I recall, Tegra 4

No no, this is a different issue. The X1 has 10 (11? i don't recall precisely) frequency domains for the CPU and at least 6 for the GPU, and all cores can be addressed independently. How do we know this? The pixel tablet of course!(you forgot that existed right? me too) That thing has been ported to Linux 4.19 and mainline and can even park cores completely independently. Also the switch on Android inmediately exhibits the same behaviour

Unfortunately, none of the Tegra

Transmeta? Did i miss something? Even the earliest tegra i can find is ARM.

And yeah the X1 was a huge fucking disappointment, specially after the blunder of the A53 cores. And yeah, the newer tegras are actually quite decent, however they don't seem to be moving in the direction Nintendo might want, the newer Tegras seem to be moving in the direction of bigger and more power for AI usage, Thor is rumored to be a giant chip with a cTDP of up to 100W. In a way things have gotten better, just maybe not for Nintendo? I wonder how bad it would be to move to something like a Snapdragon chip, Adreno has shot up in perf/W in these last few years

[-]

theQuandary@reddit

You can also keep that stuff and cut out some tensor/rt cores, the ISP unit, the DLA units, the PVA units, the ethernet controller. You know, all that stuff that's useful for the automotive dev board but not so useful for a portable console

Tensor/RT cores are definitely still included and may even be enhanced. PVA/ISP is likely needed to run their new camera system they showed off. They say they are using DLSS, DLA is probably needed if they have a hope to run it at anything approaching acceptable levels.

I didn't mention this, but Nvidia got ticked off with TSMC over Blackwell and has gravitated toward Samsung. The rtx 40 series was made on Samsung 5nm. Likewise, the cancelled Atlan and probably the upcoming Thor also target Samsung 5nm. This implies that Nvidia already has to port a bunch of stuff to the new node anyway. There's even a case that Nvidia would love to test it in the Switch 2 SoC before rolling it out to their industrial partners.

8LPP might have costed that much in 2018 but it sure as hell isn't costing that now

Do you have any sources? Wafer costs dropped a bit then spiked with inflation for TSMC. The only thing 8nm has going for it is that it didn't include the EUV price jump.

5LPE was having high defect issues well into 2022

DeepX AI was reportedly getting 90% yields on Samsung 5nm which isn't that bad (source).

i'm not seeing the same logic in the A78C

ARM is very aggressive about optimizing for power. From what I can tell, A78C released after A78AE. It's hard to believe that they'd leave out anything that would save power. After all, A78AE is generally plugged in to something while A78C is not. Do you have a source? All I could find was additional power domains for the DSU stuff which doesn't exist in A78C.

It weirdly doesn't?

Do you mean between cores on a complex or between core complexes? Every chip I've ever seen the numbers for shows several times higher latency going between core complexes.

Transmeta? Did i miss something? Even the earliest tegra i can find is ARM.

Yes, probably the most interesting CPU design that never really needed to exist.

Nvidia bought rights to all the Transmeta IP in 2008. They went on a hiring spree of Transmeta employees shortly after. Rumors spiked that they were working on a new generation of Transmeta CPU.

Shortly after, Nvidia and Intel got embroiled in an IP lawsuit. Project Denver started some time around here. Intel and Nvidia get involved in a lawsuit over GPU tech and Intel winds up paying some $1.5B. The details of the settlement aren't known, but sources allege that Intel specifically forced Nvidia to agree not to make an x86 CPU.

Nvidia was still nervous (and is today). They felt that both Intel and AMD having iGPUs would drive them out of the market (which it did). Shortly after the settlement, Nvidia unveiled Project Denver as an ARM project, but according to a rather recent article, the former Transmeta CEO states in no uncertain terms that x86 was the original goal.

The first Transmeta-style CPU was Tegra K1 which was used for their Jetson line and also the Google Nexus 9 - anandtech and sampled in 2014 (I didn't forget about the Pixel C btw, but the device got really poor reviews and hardly sold).

X1 you know about, but it was a stopgap product without enough R&D (as evidenced by the issues with the design) and X2 went back to Denver 2 (the only consumer product I know of was the terrible Magic Leap).

The successor to X2 was Xavier which was based on Caramel which was basically Denver 3. I can only assume that Nvidia wasn't getting what they wanted because this released in 2019 and they moved to buy out ARM in 2020 and Denver was dropped in favor of Neoverse designs.

I wonder how bad it would be to move to something like a Snapdragon chip, Adreno has shot up in perf/W in these last few years

I don't know about the capabilities of 8 Elite's new GPU, but X Elite's GPU was certainly lacking in capability.

Nvidia has Nintendo over a barrel. Losing all backward compatibility is a hard sell to consumers. Rewriting is hard enough with first-party titles and simply won't happen with a lot of third-party titles. Emulation incurs a large enough penalty that it's not a viable option unless the new system were even more powerful than the current Switch 2. I suspect that Adreno still needs at least one more GPU generation to catch up in features and maybe another generation to optimize the new hardware (not to mention the software).

[-]

marcost2@reddit

Tensor/RT cores

PVA is probably not needed, and the ISP is overkill. Those two blocks are designed for automotive cameras systems and are way overkill in a console.

Also DLA isn't just their AI accelerator? I thought DLSS run only on tensor cores, though if it's a custom impl they might leverage that

EDIT: as noted in this video

If you could please point me to said commit just for my curiosity, i'll take your word for it but i really don't want to scrape a 1 hour video for it

I didn't mention this, but Nvidia got ticked off with TSMC over Blackwell

Source? The way i heard it, it was the other way. What with Ampere Datacenter still being TSMC and getting huge boosts in efficiency due to that, and rubin still being slated for TSMC Source

Do you have any sources?

Legally no, i can't give a source without a lawyer knocking on my fucking door, but i can say it's gotten a lot cheaper to fab on 8nm

DeepX AI was reportedly getting 90% yields

Oh, they figured it out then, for our TC in sept 2022 we were seeing like a low 80s high 70s yield rate which was quite awful for the size of our TC

ARM is very aggressive about optimizing for power.

Source is me reading the developer-exclusive RTL (again, lawyers) i'm not a electronics engineer so i might be seeing it wrong, but one possible explanation might have to do with it's intended use? From all i can read the 78C is designed for higher power portable device, like say a laptop; so ARM might have decided that they could simplify for the average case here and not lose much, afterall what laptop would need to shut down like 6 cores?

Do you mean between cores on a complex

I mean between cores on different complexes. Like having data ping-pong between two cores on different complexes gives very near the same result as doing it in the same complex. My test might be flawed though

Yes, probably the most interesting CPU design that never really needed to exist.

Oh no yeah i know transmeta i thought they had made an actual transmeta-based cpu for a tegra and missed it. I did not know that about denver though! Thanks for the rabbit hole.

Nvidia has Nintendo over a barrel.

Yeah i know, i called it back when the original switch launched that Nvidia is not a generous partner and it seems they are paying the price now. I still do wonder how bad would it be to generate the translation layer from nvapi to whatever adreno uses

Also the adreno on the 8 elite is like 3.3TFlops with a supposed max tdp for the chip of 8.2W, that's why i was interested in it (yeah yeah 2nm i know but still interesting little bugger)

[-]

theQuandary@reddit

The way i heard it, it was the other way.

Maybe I spoke my opinion too forcefully. It could be either way. Rubin was already deep into design for TSMC before Blackwell got messed up (TSMC says it's a faulty Nvidia design and Nvidia says it's due to faulty TSMC packaging), so it being on TSMC isn't really an indication of very much. We'll know in a couple of years.

i thought they had made an actual transmeta-based cpu for a tegra

They did. As the CEO said, it was suppose to be the next-gen Transmeta design based on the previous IP Nvidia had bought and created by the same team. All they did is change out the microcode from x86 to ARM64 (though saying "all they did" is understating things a bit).

I still do wonder how bad would it be to generate the translation layer from nvapi to whatever adreno uses

Qualcomm isn't a generous partner either. A good GPU and access to all the source code would still probably suffer a 2-3x penalty, so they'd be trying for a 10 TFLOPS chip in another few years.

If I were to make a controversial recommendation, it would be for Apple to buy out Nintendo, move them to Apple Silicon, then leave them to do their thing. That would expand Nintendo's market massively from 150M devices to somewhere close to 1500M devices. They would gain access to far better hardware and at far cheaper prices due to Apple's vertical integration. Apple would gain instant marketshare with the platformer gamers which probably have the most overlap with Apple anyway.

[-]

marcost2@reddit

Current implementations do

Sounds like a nightmare to coordinate, specially since the tensor cores and the DLA cores probably communicate over a common bus, so can't imagine it's great for latency.

edit: Here's one of their commits

Yeah i found that one too, but it just add support for the t239 to cpufreq which doesn't tell much. Of the other 18, 10 are unrelated (GPG keys causing them to get pulled up), 4 are messages related to that cpufreq patch and the other 4 are from a patch updating the bpmp (NVIDIA Tegra Boot and Power Management Processor (BPMP)) firmware, so again i have no clue where people get this idea that they are porting ada lovelace stuff to the t239 (to be clear i've read this before, just never seem to be able to find an actual source more than "linux commits")

Maybe I spoke my opinion too forcefully.

Yeah, it's probably wait and see right now, the problem being that TSMC has a near monopoly on high end nodes, and they continue to just run away from the competition

They did.

Well damn, i'm gonna have to go dig up one of those boards then, transmeta cpu's are fun to mess around with, you can get pretty crazy IPC's if you are dumb enough

Qualcomm isn't a generous partner either.

Fair enough, and Apple would probably be the right partner (if they ever cared for it). I do however think you are overestimating how much overhead a translation layer for just gpu instructions costs, and even then they could just make it so that having a cartridge gave you access to a digital version of the ported game. It's still all ARM64, just changing the graphics API

[-]

theQuandary@reddit

I haven't looked into it too far, but it's looking like they may actually be talking about Nvidia's custom Jetson Linux stuff and most isn't actually upstreamed.

https://developer.nvidia.com/embedded/jetson-linux

https://docs.nvidia.com/jetson/archives/r35.4.1/DeveloperGuide/text/SD/WorkingWithSources.html

I do however think you are overestimating how much overhead a translation layer for just gpu instructions cost

not really CPU wise the cores used have the same IPC as Zen2, just lower clocked

[-]

Johnny_Oro@reddit

Not sure how 8 core 8 thread will stack up against 4 core 8 thread. The power fed to the CPU will be much lower though I'm sure, so I think it's going to be largely meaningless anyway.

The steam deck CPU is so slow that the RAM's bandwidth ends up going to waste, it showed worse bandwidth than haswell despite the much faster RAM in Chip's test. That bandwidth will exist to feed the GPU mostly.

I think switch 2 has more powerful GPU with 4K capability, and that will need a lot more bandwidth, starving the CPU further.

[-]

chaddledee@reddit

Yeah, really hard to say.

Switch 2 CPU and GPU will be more efficient than the Steam Deck, but will probably have a lower power limit. On top of that, the Switch 2 GPU is beefier, so more of the power budget will go to the GPU than CPU compared to Steam Deck. 8c8t is way better than 4c8t, especially for gaming.

If I had to guess, it'll have lower single core perf, higher multicore perf, and much higher GPU perf. The fact that there's a fan in the dock now gives me hope that the the Switch 2 power limit increases much more significantly when docked than the Switch 1 did.

[-]

marcost2@reddit

Orin is not an efficient chip, if you look up it's counterparts it's really not designed for low-power applications, with it's typical cTDP being 15-50W with a "MAX" mode of 60W, and if you look up the frequency tables the scaling is quite horrendous

[-]

theQuandary@reddit

Orin AGX (t234) seems to be a 455mm2 chip. PCB leaks prove T239 isn't bigger than 210mm2 or so at the biggest despite only losing 4 SMs (16 -> 12) and 4 CPU cores (12 -> 8). That's right in line with a die shrink from 8nm (61.2MTr/mm2) to 6nm (126.5MTr/mm2).

https://www.nvidia.com/content/dam/en-zz/Solutions/gtcf21/jetson-orin/nvidia-jetson-agx-orin-technical-brief.pdf

Max power for the 32gb variant is 40w. A die shrunk variant with lower clocks and fewer GPU/CPU constantly leaking power would probably max out at 20-25w and minimum power would probably drop to 8-10w and that's assuming they didn't also add some extra power gating to the new die at Nintendo's request.

[-]

marcost2@reddit

AFAIK 8nm->6nm is not a simple die shrink even keeping with samsung, it requires all new libraries(with 8nm being a derivative of their 10nm node and 6nm being a derivative of their 6nm) which would require a new test chip and all new validation, i sincerely doubt nvidia is doing all that for a tiny client as is Nintendo. Also IIRC samsung 6nm didn't have that density? I might be wrong i only ever saw number for 6LPP but it wasn't a doubling of density, more like a 1.2x

Back on topic, from die shots the majority of die space is the SM and cache, it could be either simplified SM (so no RT cores? i don't recall if those are used for dlss) or reduced cache. Do also take into account that the big orin agx use 256bit interaces while the T239 uses a 128bit

Also also, the big orin chip is only 64GB and max 60W, the Orin AGX 32GB version has 2 fewer SMs and 4 fewer cores, and that's the one that has a 40W max cTDP (this is also on that technical brief you linked)

All in all, this is all speculation, but i would be incredibly surprised if nintendo ponied up the money for a new tapeout

[-]

theQuandary@reddit

The change from A78AE to A78C already necessitates a new layout.

Nintendo sold 150M switches. That's hardly a tiny client.

The fact is that you can't fit a 455mm2 chip in the space shown.

from die shots the majority of die space is the SM and cache

here's a die shot of t234.

Only around HALF of the die is the SM + CPU cores which is normal for any SoC. If you remove 4 SMs and 4 CPU cores, it's only around 10% of the die area. Adding in half of the DRAM controllers still doesn't get you anywhere close to the 250mm2 reduction needed.

no RT cores? i don't recall if those are used for dlss

Raytracing and DLSS are confirmed for the Switch 2, so if they weren't present on the t234, you'd expect the size per SM to increase. That said, Wikipedia claims that the t234 die (GA10B) has 8 RT cores which makes sense because removing them completely would require significant hardware and software changes and would make a minimal impact on the die size or price to customers.

Also also, the big orin chip is only 64GB and max 60W, the Orin AGX 32GB version has 2 fewer SMs and 4 fewer cores, and that's the one that has a 40W max cTDP (this is also on that technical brief you linked)

I knew that. I mention the 32GB version because with fewer CPU cores, fewer SMs and lower GPU clocks, it is much closer to t239.

All in all, this is all speculation, but i would be incredibly surprised if nintendo ponied up the money for a new tapeout

Nintendo paid for a die shrink of the original Switch CPU which not only shrunk the chip, but removed the useless A53 cores making it a redesign too.

[-]

marcost2@reddit

I wrote 6nm, but meant 5nm.

Oh gotcha gotcha, yeah that's a bigger jump, but even worse from a redesign perspective

The change from A78AE to A78C and reduction in SMs and DRAM interface already necessitates a new layout

New phyiscal design? Sure, but you can get away with not revalidating timings and plls which are a big expense and take a fuckton of time, plus ARM guarantes both the A78C and A78AE as "drop in" replacements in terms of RTL so they have to be roughtly similar, the cut SM do require a new layout.

Nintendo sold 150M switches. That's hardly a tiny client.

And in 2018 they were 10% of Nvidia's revenue (while only selling 17M unit, that's pretty impressive tbh), but that was a way different climate, datacenter was only 20% and workstation was also 10%. Now's a different picture, datacenter was 78% of Nvidia's 2024 revenue, with gaming falling to 10% of the revenue (oh how the turntable), automotives and tegra are rounding errors at least than 1%

The fact is that you can't fit a 455mm2 chip in the space shown.

Oh yeah i agree

here's a die shot of t234.

Only around HALF of the die is the SM + CPU cores which is normal for any SoC. If you remove 4 SMs and 4 CPU cores, it's only around 10% of the die area. Adding in half of the DRAM controllers still doesn't get you anywhere close to the 250mm2 reduction needed.

I disagree that it's only around 10%, the 4SM (including their crossbars and whatnot), between that and the core, plus the ddr interfaces it should drop are by around 100mm2. Yes i'm still far from the 250mm2, but i'm doing this based off of a poorly annotated die shot with only some passing knowledge on identifying structures (for example i'm pretty sure the top border has a bunch of interfaces that could be cut, to my untrained eye that looks at least like some pcie lanes?).
I'm not discussing the die shrink from a size perspective tbh, i'm doing it from an economics perspective (as we know neither Nintendo nor Nvidia work for free)

Raytracing and DLSS are confirmed for the Switch 2, so if they weren't present on the t234, you'd expect the size per SM to increase. That said, Wikipedia claims that the t234 die (GA10B) has 8 RT cores which makes sense because removing them completely would require significant hardware and software changes and would make a minimal impact on the die size or price to customers.

Oh i completely forgot that raytracing was confirmed, so no simplified cores then :/ (although they could cut some rt/tensor, maybe a 12sm, 6 tensor, 1/2 rt configuration? that should save some more space and power)

I knew that. I mention the 32GB version because with fewer CPU cores, fewer SMs and lower GPU clocks, it is much closer to t239.

It is still however, GA10B, just a defective one. My point of bringing the big boy is that we have the power modes nvidia ships it with on linux
https://developer.ridgerun.com/wiki/index.php/NVIDIA_Jetson_Orin/JetPack_5.0.2/Performance_Tuning/Tuning_Power
How valuable is this? Not very, but it's a nice theorizing point given it's the closest data points we have

Nintendo paid for a die shrink of the original Switch CPU which not only shrunk the chip, but removed the useless A53 cores making it a redesign too.

Yes and no and kinda? TSMC16NM IS a die shrink, meaning same libraries (unlike 5nm which has all-new libraries) meaning while a new physical layout was required, clock validation and timing was still valid. ALSO, they didn't pay for it, not on their own anyways, Nvidia paid for part of it for use in their Shield 2019(was it 2019?) refresh.

Cutting out parts of a chip is surprisingly a lot easier than moving to a new node with new libraries (at my company a 5nm->5nm TSMC redesign took 12 months from planning to tapeout and production, the 5nm->3nm port however has been ongoing now for 23 months)

Anyways this is a nice convo :)

[-]

DerpSenpai@reddit

>4 core 8 thread

8 threads doesn't mean it equals 8 cores, in MT, full parallel compute scenario like rendering, it gives less stalls in each core and gives like 30% better perf, so a 4c/8T might have scenarios where it can beat a 6c/6T core, but it real world scenarios, you will always prefer the higher core count. always

[-]

Johnny_Oro@reddit

It's going to depend whether they're really using A78C (8 p-cores) or A78 (4p+4e hybrid). I remember there were some speculations that it's going to be using the former, but the latter, a hybrid architecture, would be more beneficial to Switch's battery life. There's strong evidence that it's going to be an A78 variant though.

The only teardown article I found says it's using A78 derivative with 1 "big", 3 "medium", and 4 "small" cores.

Ampere had a big jump because they increased the amount of cuda cores by that much.

No, they didn't. If they had, that would have affected gaming performance. If you compare the 2080 to the 3080, the 3080 has only 48% more SMs (46 vs 68), but 195% higher TFLOPS (10.1 vs 29.8).

Ampere had that big of a jump in TFLOPS because it had double the number of FP32 units per SM. But that 1) has no effect on other operations like FP64 or any form of INT, and 2) the gains are only realized in some specific (mostly workstation) tasks because those doubled FP32 units were still sharing the same number of dispatch units, registers, caches and interconnects as Turing. That's why when you look at actual gaming performance you see the 3080 is about as much faster than the 2080 (50%) as it has more SMs (48%), and not 3 times faster as the misleading TFLOPS spec would imply. You only see the benefit of those doubled FP32 units on the very specific tasks that can fill those units through the same unchanged amount of dispatches/registers per SM, and on 3D rendering/gaming that is pretty much never.

Moreover, an architecture's performance per flop is not linear, at around 9 TFLOPS, Turin is ahead of Pascal by 42% (and of Ampere by 50%) but at around 2 TFLOPS they are tied.

This is a perfect illustration of how you have no clue what you're talking about.

You can't just say "at 2 TFLOPS". How are you getting Turing and Pascal chips down to 2 TFLOPS? Are you taking a large chip and running it at very low clocks? Are you taking a very small chip and running it at high clocks? Because those two different cases would perform drastically differently.

GPUs aren't just CUDA cores, they'll have different amounts of other structures like ROPs, geometry/tesselation engines, memory buses and bandwidth, and completely different manufacturing processes with different power efficiency curves, which make a comparison like this completely meaningless.

RDNA3's changes on the other hand, are more egregious in regard to the flops discussion

No, they weren't. What AMD did in practice achieves the same results as what Nvidia did with Turing, but through a different method. They implemented dual-issue instructions, meaning each FP32 unit in RDNA 3 can execute up to 2 different instructions per cycle so long as both instructions fit in the register and there's enough time left in a cycle to execute the second one. In simplified terms, it's somewhat like hyper-threading for GPU cores. The approach is different (double the FP units per SM sharing the same supporting structures for Nvidia, vs FP units that can execute up to 2 instructions per cycle for AMD), but the result is the exact same, a theoretical doubling of FP32 performance, which can be achieved in some niche workstation workloads, but pretty much never in gaming. Neither of them are any more or less "egregious" than the other, and in practice the result is identical.

instead you now double one variable in the flops formula that had been consistent for all GPUs for the past 20 years or so

Which is exactly the same thing Nvidia did, doubling the number of FP32 units without doubling the number of supporting structures, meaning you never benefit from this doubling in gaming and the "doubled" FLOPS numbers are completely worthless for games.

[-]

Siats@reddit

You could have looked a little more at the graph. You certainly have a better grasp of the details than I do but you are missing the forest for the trees. Yes, at the level of the 3080, Ampere completely falls flat on its face, exactly as you described but we are not dealing with a 3080-equivalent chip.

Does Ampere have half the performance/compute of Turin across the stack? It doesn't appear to be the case, check the graph, that answers your question about the GPUs used, all stock (except for the 2TFLOPS ones, those are 1630 and the 1050 Ti). Of course, there's more to a GPU than just cores but we are already simplifying things, using flops to gauge performance. If you reject this kind of comparison outright then I have nothing, you can go your merry way.

I do thank you for correcting me on the details I got wrong, but this comparison started with the claim of the 780m being 3 times as powerful as what's inside the SW2 and while I can accept that Ampere has inflated compute in a very similar way to RDN3 itself, this particular implementation (780m) is clearly far less capable for its compute than the closest Ampere approximations we have available.

[-]

Nerina23@reddit

Pretty telling that the SoC details are still under wraps.

Steam Deck and PS5 for example were shouted from the rooftops "Look at all the awesome hardware inside".

How much space do we have?

We roughly know the die size from the PCB images and they indicate 190-210mm2 based on the memory sitting next to it.

https://imgur.com/W4ohTUz

https://imgur.com/a8vrnHJ

Well, it turns out that if you double the density, then you shrink from 400mm2 to around 200mm2.

Maybe you should research instead of accusing people of making stuff up.

[-]

Kepler_L2@reddit

Overall, that puts the frames/FLOPS as favoring Ampere over RDNA2 a bit.

What? RDNA2 has way higher FPS/TFlop.

Switch 2 is 5nm Samsung

More significant, most likely, is cost. Even with how rare and low volume microSD Express cards are right now, they're running at around $0.23/GB. CFexpress A cards, for all of their traction, are around double that.

Is it possible to make an adapter to turn microSD Express into full size like you can with normal SD cards?

[-]

Constellation16@reddit

Yea, SDe is the only thing I really care about with this console. Proliferating of fast pluggable NVMe storage. Hopefully they will quickly become the standard in SBCs, etc.

[-]

Vb_33@reddit

Im in love with the fact that it has the holy Trinity. People were expecting only at most 1 of these features but Nintendo included all 3: VRR, 120hz and HDR. And it's all available even in handheld mode.

One of my wish list compromises was that Nintendo would have support for 1 or more of these features but only docked to your television not the actual handheld screen itself. IDC that it's OLED I'd rather have an ok LCD screen with HDR, 120hz and VRR than a 60hz HDR OLED screen.

[-]

Soggy_Association491@reddit

I doubt it will do any HDR with an IPS screen.

[-]

nmkd@reddit

It has HDR.

[-]

theQuandary@reddit

That's why I mentioned the uarch difference. Newer generations generally have a higher real-world utilization rate.

[-]

Azzcrakbandit@reddit

I mean, amd more than doubled their teraflops in the high end cards going from rdna 2 to 3, yet that's not reflected when looking at it's performance improvement in games.

[-]

You are the one moving the goalposts here.

You are arguing that it is NEVER reliable while I am arguing that it is MOSTLY reliable.

You COULD get terrible scaling, but as a matter of practice, you generally DON'T within any given company.

I generalized that over several generations, the FLOPS scaling correlation is generally true for any specific manufacturer (turns out that it is). You jumped in to claim with ZERO evidence that it couldn't possibly be true. You then pulled out an outlier in an attempt to disprove a generality (by the way, if you sample from GCN1 through RDNA4 instead of a single outlier, the FLOPS scaling trend holds true).

[-]

Azzcrakbandit@reddit

Hey man, I'm still waiting for the new goal postto be installed.

[-]

Azzcrakbandit@reddit

Are you doing better yet?

[-]

Azzcrakbandit@reddit

What are you talking about? I never said there wasn't a correlation. I provided modern evidence. I provided more than one modern example to show it's not an outlier.

It's funny how you try to undermine my argument by claiming I have no proof when I've provided several examples, and yet you have failed to provide any at all. Please do better.

[-]

ResponsibleJudge3172@reddit

AMD per flop with RDNA3 is MUCH lower than Ampere because their implementation is entirely different

[-]

Hailgod@reddit

flops are only useful to compare within the same generation.

[-]

theQuandary@reddit

I edited my post further up. As it shows, the FLOPS comparison is reasonably close. Certainly close enough to show that 4K of existing Switch games is seems possible.

[-]

Vb_33@reddit

It's an aspect of it, fp32 does matter for gaming performance but as you say it is not the only factor.

[-]

Toojara@reddit

3.1 TFLOPS is 6.2x more than 0.5 TFLOPS and the newer uarch is much more efficient

That's the fun part, it's actually not. Ampere had a massive FLOP/FLOP performance drop compared to Turing due to ALU changes, though as a result the FLOPS skyrocketed. But even Blackwell needs roughly 50% more FLOPS than Maxwell to achieve the same level of performance.

[-]

theQuandary@reddit

Ampere's per-flop performance is predicated on sharing the int ports with floats (not too different from RDNA3, but with less hardware/ports). We'll have to see how much they can optimize performance here. If the 8-core CPU rumors are correct, then the shared memory arch means a decent percentage of those

We'll see what careful optimization can do to balance int/float operations and what the final specs are. Nintendo generally doesn't overstate things, but maybe they're heavily relying on DLSS for those numbers.

And just for reference, a 3060 is 12 TFLOPS and that's realistically a 1440p GPU. Quarter of that is 1080p native.

What does that have to do with anything? I saw someone running Half Life 2 at 16k resolution (that's 4x 8k or 16x 4k) with >60 FPS.

It comes from knowing what one is talking about, but here's an example if you don't want to put in the effort to test it yourself.

TPU lists the 5090 as averaging 246.2fps at 1080p, 108.8fps at 4K.

Trying to run games at native 4K would be a huge waste of processing power. There's really no point doing it on a much more powerful hardware, let alone on Switch. But that's not the point.

Calling a game "4K" when it's just upscaled to 4K is stupid because it doesn't mean anything.

Deep Learning Super Sampling 4 supersampling?

DLSS 4 has a substantial frametime cost across the board. It's still heavy on ampere gpus running at hundreds of watts. Not sure why you think the performance cost is tiny.

If you watch the video I linked, you will see an ampere card struggling with DLSS 3. 4 is heavier than 3.

[-]

frostygrin@reddit

Deep Learning Super Sampling 4 supersampling?

It's Nvidia who decided to market frame generation and ray reconstruction as "DLSS". :) So yes, Deep Learning Super Sampling 4 supersampling... As opposed to ray reconstruction that really is significantly more demanding on older cards.

DLSS 4 has a substantial frametime cost across the board. It's still heavy on ampere gpus running at hundreds of watts. Not sure why you think the performance cost is tiny.

It's relatively tiny, compared to DLSS 3 and to running demanding games without DLSS. Especially as DLSS 4 makes it possible to run games at lower resolution with better fidelity. I have the 2060 - and DLSS 4 is a game changer. I don't run it at 4K though - but then games are going to be a lot more demanding at 4K too.

[-]

conquer69@reddit

It's relatively tiny, compared to DLSS 3

DLSS 4 is heavier than DLSS 3. Not sure why you think DLSS 4 is lighter than DLSS 3. https://imgur.com/vsL6PMS

Notice how switching from DLSS 3 CNN to DLSS 4 Transformer incurs a substantial performance hit.

If it can be that heavy on a 5070 ti that pulls 280w, the performance hit of running it on a 10w mobile gpu won't be tiny. Even DLSS 3 is heavy for it.

The nintendo switch isn't looking to run games at 4K. Even upscaling them to 4K with DLSS will be problematic. Seriously, watch the video I linked.

[-]

frostygrin@reddit

DLSS 4 is heavier than DLSS 3. Not sure why you think DLSS 4 is lighter than DLSS 3. https://imgur.com/vsL6PMS

Which game is this? This kind of performance reduction is what you'd see in games with ray reconstruction. Or maybe it's the other way around and the performance hit is so noticeable because the card is so fast that normally this bottleneck is hidden and irrelevant on slower cards. You just end up switching from DLSS Quality with CNN to DLSS Balanced with Transformers - and get a significant boost compared to native rendering.

The nintendo switch isn't looking to run games at 4K. Even upscaling them to 4K with DLSS will be problematic.

Azzcrakbandit@reddit

Words don't matter, just the results. I couldn't care less what they call it as long as it looks good.

[-]

Herby20@reddit

I don't think it is out of the question that Nvidia would stand to make quite a bit of money making a custom solution with DLSS 4 esque abilities for the Switch 2.

[-]

conquer69@reddit

Yeah it will have some form of DLSS but it might not look as good as DLSS 3.

[-]

That doesn't mean that Nvidia would make less money per unit than for sales to Nintendo which was the original claim.

[-]

Sarin10@reddit

yup. Apple TV 4ks top out at $150.

Oafah@reddit

He's right. Consoles themselves are usually loss-leaders, with only recent iterations actually turning a profit. You're much better off just being a vendor of one of the key components - especially when you're the only game in town.

[-]

Johnny_Oro@reddit

Nvidia Shield was $200 in 2014 and barely selling. Nintendo Switch was $300 in 2017 and selling like hot cakes. It's insane how much more money they made off the Switch.

[-]

theQuandary@reddit

I don't know that they made that much money. Nvidia had managed to tick off every mobile partner they worked with. If that weren't enough, Tegra X1 had 4 defective A53 cores making it a complete no-go for most device makers.

X1 was released in 2015 and when the Switch released in 2017, the only customer for the chip was Nvidia theirself with Jetson IoT and Shield TV.

Tegra X1 was a 118mm2 chip. Apple's A9 from the same year was 96mm2 and went into a $650-750 device. There's no way Nvidia could get big margins on a chip of that size while pricing it low enough to fit into a $300 Switch (BOM was estimated at $257 before assembly/shipping costs)

I believe the truth is more like Nintendo didn't care too much about the missing E-cores because their device would be running P-cores the whole time anyway. Nintendo got a sweetheart deal on Nvidia's stockpile of otherwise unsellable chips and Nvidia got to cover their costs and turn a small profit.

[-]

Johnny_Oro@reddit

The BOM was estimated at $257? I'm quite sure it's way lower than that. The $200 Tegra was sure as hell sold at a profit, no matter how low it was, which I'm sure not low because that's the launch price tag. And by the time the switch came out it was a 3 year old chip on an old node.

Nintendo switch had the additions of a screen, not a very good quality one, 3500 MAh battery, and a couple of controllers with gyroscope, not very good quality either. I don't think those additions were worth $100. Lenovo P2 had 5100MAh battery and it was only $200 in 2017.

Also Apple has always made crazy amounts of profit, and any profit compared to Apple's is small profit. By the way A9 was fabricated on TSMC 16nm/Samsung 14nm while Tegra X1 was fabbed on TSMC 20nm, only reduced to 16nm in 2019.

[-]

theQuandary@reddit

You can look up the cost for yourself, but that was the only claim I could find and it was everywhere at the time.

Nvidia didn't lose money per chip, but the margins weren't super high because they had no leverage, because console margins are expected to be thin, and because the alternative was zero margins and writing off an entire chip design.

Savings per node in 2015 isn't what it is today. Apple made lots of money, but so did companies like Samsung which is the kind of company Nvidia was hoping to sell their chip to.

[-]

Johnny_Oro@reddit

The source is a guesstimation from Nikkei, and the article is paywalled. Yes Nvidia certainly didn't lose money but the Switch wouldn't cost that much to manufacture. Lenovo P2 was $200, and while it only had one gyroscope and no controllers, the screen was 1080p AMOLED with Gorilla Glass protection. The Switch cut so many corners that I remember the screen was filled with scratches due to the dock.

Perhaps it's $257 if the R&D is accounted for, but it's not like Nvidia Shield didn't have any R&D cost.

nachoz12341@reddit

This is an objectively better system than the download codes that exist this generation. It's a compromise. Between cost saving and letting a user own the license of the game they bought. Don't really see this as a negative. You can now resell and share a game that otherwise would have been locked to your console.

[-]

Marble_Wraith@reddit

Unless Nintendo or the game studio decides to stop supporting the game... in which case your key card amounts to a plastic paperweight.

[-]

nachoz12341@reddit

I mean is that any different from a regular cartridge? You'll always be somewhat dependent on nintendo having a functioning system. Seems like a weird nitpick. Even cartridges or disc's with the full game don't have things like updates included which are dependent on the store. And even more games only have part of the game on the physical media.

[-]

dieplanes789@reddit

Yes it is different. On a regular cartridge as long as your system still works so will the game even if it is a brand new one to you. This is the may not be getting updates anymore and the game might not either but you will still be able to play a game you have never owned or played before even if it is the version 1.0.

With the new system as soon as the download server goes down so does your ability to play a game you don't already have downloaded.

[-]

nachoz12341@reddit

Setting up a console still requires a working authentication server. There is always a dependency on Nintendo. I don't disagree that there needs to be better steps for game preservation, but let's not pretend this cartridge as a license system isn't a win for the consumer in an era of digital codes. Sony and Microsoft are both actively moving away from physical media as nintendo tries to incentivize it for publishers.

[-]

Marble_Wraith@reddit

I have an N64 from 1998 that still works. I can play any game i wish even though Nintendo has long discontinued production and support of all the game titles.

Now let's suppose I buy a switch 2, and a game key card.

Fast-forward to 2050...

Lets assume both switch 2 and key card are 100% functioning from a hardware perspective.

Will i be able to play the game?...

Probably not, because there is some server somewhere on the internet that does validation on the key card that authorizes you to access your own game. If that server doesn't exist anymore... you have no access to something you paid for.

It's not a weird nitpick, it's a literal scam. Because as pointed out to someone else, there is only 1 reason for moving to this model, and it's not because of some bullshit about game cartridge capacity.

Are there exceptions? Of course. For games that are exclusively multiplayer, having a server is fair enough and if they were going to take the players into consideration with an EOL plan (eg. open sourcing the server / ability to self host) there'd be a whole bunch of extra formalities and legalities to sort through.

But i'm talking single player games (which for Nintendo are a majority). There should be no single player game that requires an internet connection just to play it.

Even cartridges or disc's with the full game don't have things like updates included which are dependent on the store.

Which is just as shit, but out of scope for the discussion hence why it was not mentioned.

I don't know when marketing and sales go this idea of perverting DLC's but that too has to stop.

[-]

ManOnAHalifaxPier@reddit

I really don’t see why this is so commonly cited as some huge problem. The license is attached to the card, not the account or the console, so you can loan/sell a game no problem. Some games are just so big now that they don’t fit on the cartridges, so it’s a compromise. But a small compromise. Even on disc-based consoles, the game needs to be installed before it’s played.

The only downsides I can really see are that you a) need an internet connection - which is annoying but everyone buying a $450 Switch has an internet connection - and b) 10+ years down the road they shut the servers off and you can’t download the game anymore. By that time normal folks won’t care and hardcore gamers and techies will be able to mod the system and load backups.

[-]

Marble_Wraith@reddit

The license is attached to the card, not the account or the console, so you can loan/sell a game no problem.

We have been here before. CD keys in the late 90's / early aughts. It's just as shit now as it was then.

Some games are just so big now that they don’t fit on the cartridges

2TB in a 22mm x 30mm package...

https://www.pcworld.com/article/564466/microns-new-runty-nvme-4-0-ssd-can-pack-2tb-of-storage.html

Shieet even CFexpress 4.0 B cards for camera's are available in 4TB variants now.

Even on disc-based consoles, the game needs to be installed before it’s played.

Yeah and that's garbage as well

The only downsides I can really see are that you a) need an internet connection - which is annoying but everyone buying a $450 Switch has an internet connection - and b) 10+ years down the road they shut the servers off and you can’t download the game anymore.

point b is exactly the problem

By that time normal folks won’t care and hardcore gamers and techies will be able to mod the system and load backups.

These are huge assumptions. Not to mention given the litigious nature of Nintendo they have consistently done everything in their power to stop people from doing this with older titles / consoles.

[-]

ManOnAHalifaxPier@reddit

CD keys were shitty not because of the concept but because of the disjointed nature of the execution. With a tightly integrated platform where all keys are the same, the UX will be just like traditional physical games. To me that’s a total non-issue

Everyone is complaining about game prices, yet you are drawing comparisons to very expensive storage solutions. Hell Nintendo had to drop support for microSD in favour of microSD express just for usable load times. Game cards that are performant and capacious enough for AAA games would be too expensive for game devs to eat.

They are big assumptions. But the Wii and Wii U are some of the most easily modded devices out there these days. ROMs are absolutely everywhere. I feel stupid trying to pitch piracy as a solution to a clear design flaw. But I think it’s a fine compromise. Especially since the games won’t just disappear from the Switch’s memory when the servers are off

[-]

Marble_Wraith@reddit

CD keys were shitty not because of the concept but because of the disjointed nature of the execution. With a tightly integrated platform where all keys are the same, the UX will be just like traditional physical games. To me that’s a total non-issue

No they were shitty. If i buy a physical medium i expect that physical medium to contain the digital content i purchased. End of Story.

Everyone is complaining about game prices, yet you are drawing comparisons to very expensive storage solutions. Hell Nintendo had to drop support for microSD in favour of microSD express just for usable load times. Game cards that are performant and capacious enough for AAA games would be too expensive for game devs to eat.

Even CoD which held the record for quite some time as being the largest game was only 250GB.

https://www.bhphotovideo.com/c/product/1551702-REG/sandisk_sdcfe_256g_ancnn_256gb_extreme_pro_cfexpress.html

A sandisk CFexpress 256GB card is $120 USD, and that's consumer pricing without a bulk SLA or any kind of B2B arrangement in place.

In addition CFexpress is way better read/write then microSD.

I remember paying $99 for physical copies of games back in 2006. Admittedly they got cheaper, probably because of economies of scale (DVD / Blu Ray production picked up) but that can also happen for CFexpress cards.

Considering all that + inflation, you still think it's "too expensive"? 🤣 I think your sense of $value has been distorted or you're ignorant.

They are big assumptions. But the Wii and Wii U are some of the most easily modded devices out there these days. ROMs are absolutely everywhere. I feel stupid trying to pitch piracy as a solution to a clear design flaw.

Gee ya think 😑

What makes you think it's a design flaw?... If it's been designed, then by definition it's been intentioned which means Nintendo has a reason for doing it. And it's got nothing to do with memory card capacities (as evidenced above).

But I think it’s a fine compromise. Especially since the games won’t just disappear from the Switch’s memory when the servers are off

Irrelevant.

"One note worth mentioning is that an internet connection is only required when booting up a Game-Key Card for the first time, after which the game can be accessed offline. However, the Game-Key Card will still need to be inserted into the Switch 2 console in order to access the digital game."

https://www.gamespot.com/articles/certain-nintendo-switch-2-physical-games-are-game-key-cards-for-downloads/1100-6530576/

So you can play without the internet, which i could do with games in 1999... hooray 😑

That doesn't address the problem. What happens when support for the game goes away and they decide to retroactively remove your game / invalidate your license?

It only takes a few seconds connected to the internet...

Both your switch and your keycard could be 100% functional and you would still lose access.

Sony is the more egregious offender, but Nintendo is just as guilty / has precedent for this behavior (i trust you can google).

[-]

no_salty_no_jealousy@reddit

Somehow nintendo fans still blindly support that nintendo greedy BS move. They are just worse as apple fans.

[-]

Marble_Wraith@reddit

I rate Apple higher then Nintendo.

If you're inside Apple's "walled garden" at least they try to make it as nice as possible and offer maximum freedom within said garden. Good luck if you want to interact with anything outside the garden, but at least you know what you're buying.

Nintendo... It's like you buy the switch (your own walled garden), but once you get in there you notice it's a little sparse, and you wanna give it a glow up with some boujee drip (games)... as the kids say. Game key cards is effectively getting permitting Nintendo / game studios to cease support of a game and retroactively invalidating your purchase without refunding anything.

[-]

el_f3n1x187@reddit

Peiple are really sleeping on this.

The mode already exists on the switch 1.but it looks that the pushnis bigger this time around.

[-]

ea_man@reddit

Nintendo just sold me on the Retroid Pocket 5 for 210$.

[-]

nmkd@reddit

...okay? Tell me when you can play Zelda BotW or Cyberpunk 2077 on it.

[-]

ea_man@reddit

Would you like a list of the 100 000 games that you can play on the RP5 instead?

That should include BotW by the way.

[-]

nmkd@reddit

That should include BotW by the way.

Quick google and people say stuff like this lol

Stable high 20's to 30 fps. Some glitches in Shrines but manageable. Sometimes I crash out [...]

"Stable high 20s" is hilarious

[-]

ea_man@reddit

"Stable high 20s" is hilarious

That is how it runs on the Switch.

[-]

ea_man@reddit

For Botw you can simple google, there are many guides.

Listen that maschine can play hundread of thousands of games, 100x more than a Switch, it ain't sound to fixate on a single game that you can't play where there are thousands that run flawlessly.

Zelda

Donkey Kong

Kirby

Fire Emblem

Mario Odyssey

Metroid

Pokemon Legends

I get you're mad about the downvotes for your shitty replies, but you don't need to be an asshole.

[-]

AzovstalBBQPorkPit@reddit

HOW CAN U NOT PLAY MUH NINTENDO, BING BING WAHOOO!

grow up.

[-]

nachoz12341@reddit

How is that realistically any different from any other physical media at this point? Pretty much any media will require some form of download before you start playing. Wouldn't you rather at least own the copy and be able to resell/lend it out?

[-]

kuddlesworth9419@reddit

Hopefully the hardware and software isn't too weird so we can get an emulator for it in the future. Will be nice to play these games on desktop or other handhelds in the future.

[-]

teutorix_aleria@reddit

It's an ARM cpu with nvidia GPU, its basically just an upgraded switch architecturally. The main barrier to emulation is nintendo's litigious nature.

[-]

nmkd@reddit

Bad take.

Yeah this is something people are forgetting, even if the switch 2 is cheaper initially, as soon as you buy a couple of games you’re spending more than you would for other consoles.

[-]

The point is you can't compare a home console with a device that also doubles as a handheld and expect them to be the same cost.

The PS5 is ultimately a box that exists to sit in one place. For the same cost I get a console that I can also carry with me wherever I go. This means it needs a screen, in-built speakers, a microphone, and mechanisms for detachable controllers. Those components come with a cost.

Yes first party games are expensive but Nintendo is very good about not doing microtransactions and there is a huge, ever-expanding third party catalog on the Switch that will carry over to the Switch 2 as well.

The only thing that comes close to the value proposition of the Switch 2 is the Steam Deck and even that's not close because of the performance difference and the fact that there are still many games that flat out don't work on the Steam Deck

[-]

Sarin10@reddit

To be fair, a PS5 controller (which is included in the price) comes with a mic.

The only thing that comes close to the value proposition of the Switch 2 is the Steam Deck and even that's not close because of the performance difference and the fact that there are still many games that flat out don't work on the Steam Deck

And it's on old silicon.

[-]

inti_winti@reddit

It’s the value part that people are missing. What you value in that steam deck means nothing to someone who’s only interested in first party Nintendo titles, having gazillion more games in steam doesn’t change that.

They’re not the only handheld yes, but the only one that will play first party Nintendo titles (without the hassle of emulation). And it does so portably+docked. People said the same thing about the og switch given it was set to use the older Tegra SoC, yet it’s one of the best selling consoles of all time.

As for the greed part that’s a big LOL. I completely agree with you but you can make that argument for every company and you’d be right. If none of Sony/Microsoft/nintendo were greedy, exclusive games wouldn’t be a thing.

Valve is a little different, but I’ve played enough cs to know Valve is greedy too, just not in ways you’d notice unless you played their games long enough.

[-]

Vb_33@reddit

Is it? Sony raised prices awhile back and Sony barely makes any digital editions, most PS5s produced are the disc version.

[-]

DeusScientiae@reddit

Then buy that instead and live your life dude.

[-]

If you don't have it, then don't spend it. What's the problem here?

[-]

aj_thenoob2@reddit

Nintendo has always been the budget console though. This is a device for kids primarily. When has the Nintendo option been more than Sony or Xbox??

[-]

DeusScientiae@reddit

No they most certainly have not lol.

[-]

Sarin10@reddit

And Nintendo games themselves have always cost the most and almost never drop in price

This is the most egregious one.

You can probably convince dad to pony up an extra $100 on a $400 console vs a $300 console. It's much harder to convince them (or have enough money) to spend $70 on all the big games, vs. what, $20-40 after a few years?

[-]

DeusScientiae@reddit

And tbh? If I could experience BOTW for the first time all over again, I'd 100% buy it again at full price (as weird as that scenario sounds).

Sufficient-Diver-327@reddit

The PS2 was released at an MSRP of $299 USD. Adjusted for inflation, that's $550 USD.

[-]

aj_thenoob2@reddit

And the GameCube was $200.

The PS5 was $500 while the Switch was $300. The scale still stands.

[-]

Sufficient-Diver-327@reddit

That's also missing the point that a Switch would be more directly comparable to a Gamecube + Gameboy Advance/SP.

[-]

anival024@reddit

People forget the Nintendo 64 was 450 bucks

In what world? It was $200 at launch in the US. It was planned to be $250, but they dropped the price at the last minute.

[-]

The real competition for the Switch 2 is the $399 Steamdeck.

That extra $50 for the Switch 2 gets you:

almost an inch of screen
1080p@120Hz instead of 800p@60Hz
100 grams lighter (we'll see how much of this is battery later though) and smaller in every dimension.
If the rumors are correct, Switch 2 will have nearly 2x the GPU power. A78 is something like 15% faster than Zen2 per clock in SpecInt2006, so the CPU is probably fast at normal clockspeeds too. Even better, A78 was targeted at 5nm while the non-oled Steamdeck uses 7nm.

I can see that all being worth $50 to most people.

[-]

aj_thenoob2@reddit

I don't think so. The steam deck isn't something most people know about. Nintendo is. The competition for the switch 2 is the switch 1 and the 2 better have some good games to justify it being $150 more than its predecessor.

[-]

theQuandary@reddit

If you look for a handheld gaming system, it's Steamdeck, Steamdeck knockoffs, and the Switch.

You may not consider them competitors, but it's the closest you can come and the hardware/price comparison clearly shows that the Switch 2 isn't overpriced compared to the competition.

[-]

LucAltaiR@reddit

If Steam Deck is their real competition then Nintendo is in trouble. I'm expecting Switch 2 sales to blow steam deck out of the water in less than a month. How much units they're targeting with this? 100M? It's a different ballgame compared to the Deck. Different audience too of course.

Also, if we really wanna compare it to the Deck, it can't be down to hardware only. Take a look how much a game costs on the Nintendo store and how much they costs on Steam.

But I stick with my former point, I still wouldn't make that comparison, people will be buying Switch 2 for the 1st party games. 3rd party you can already find them much cheaper and better running elsewhere.

[-]

theQuandary@reddit

A large console that requires a TV is never going to compete with a handheld. That leaves the Steamdeck (which does a pretty good job emulating the Switch) and it's knock-offs. You could say that the other competitor is a phone, but they are massively more expensive and don't compete in game quality.

[-]

aj_thenoob2@reddit

Ps5 pro is dead on arrival. It has no games.

[-]

Zerasad@reddit

$299 in 2017 is $389 now. Still overpriced. And consoles tend to remain the same nominal price. Xbox One X was $499, just as the Xbox Series X. Ps3 was $499, PS4 was $399 PS5 was $499 or $399 digital. Rising the peice by 50% in just one generationis almost unheard of. The only console that did it is the PS3 which had a completly awful launch.

[-]

Sarin10@reddit

Percentage is a bit of a misleading statistic.

Going from 100 to $150 is a 50% increase.

It's obviously not equivalent to a $300 to $450 increase, which is also 50%

[-]

DeusScientiae@reddit

Still overpriced.

Uh huh. That's why it's going to be wildly successful, just like the last one, right?

[-]

aminorityofone@reddit

It will be successful because it is nintendo. They almost can do no wrong (Wii U is an exception). Any non gamer with kids/grand kids will buy nintendo because of its reputation of kid friendly. The switch had joycon drift which if it were sony or xbox would have been a recall. The nintendo online store/online gaming has been notorious for being shite. Nintendos constant anti fan behavior and the constant shutting down of older games. Never putting its games on sale despite being years old. These are just the ones i can think of. Yet, the switch was extremely successful. Also, not being able to afford something doesnt make it over priced.

[-]

gahlo@reddit

At least the PS5 had drift issues too

[-]

Perhaps you don't speak English as your first language, but I literally asked you what makes them shitty.

For that matter calling a product over priced or correct priced is an opinion. Not a fact.

Johnny_Oro@reddit

To be fair, Sony and Microsoft will use AMD and Intel CPUs that are most definitely cheaper than Nvidia SOC.

[-]

osirus35@reddit

It every game is 80. But yes 80 is expensive. I wonder if this is paying for the custom bigger size game card. Also yes 450 is also a lot for the console. BUT… let’s say it was the anticipated 400. Still expensive but a little better. That and a game is what 470ish. So that’s close to the 500 bundle anyways

[-]

littleemp@reddit

The $450 price is going to age like milk too, with the orange man now announcing insane tariffs on taiwan and china.

They are not the same type of oled, they also special display drivers built into the soc. Those display drivers are what are missing on the laptops and handhelds.

On top of that, most phones don’t have traditional VRR. They have dynamic refresh rates, that change based on a number of variables. For example, when in safari 120hz, when in calculator 30hz. I’m simplifying it but it’s not really traditional VRR.

[-]

CarVac@reddit

My pixel 8 has a visible color shift at low brightnesses when switching refresh rates. Generally it's 120hz when I touch the screen and a few seconds afterward.

[-]

rubiconlexicon@reddit

I'm unaware of any phone that has true VRR. What they have is switching between several static refresh rates depending on type of content.

[-]

andrewia@reddit

It took a lot of digging but you're right. Apple has the widest range of options but they're still discrete steps. https://developer.apple.com/documentation/quartzcore/optimizing-promotion-refresh-rates-for-iphone-13-pro-and-ipad-pro

[-]

[-]

Fromarine@reddit

We don't even know it's not mini led

[-]

Yes.

$80-$90 games is reasonable?

No.

[-]

uzzi38@reddit

You can't compare the performance of a handheld with a regularly powered console.

I think it's not very sensible to expect it to beat the existing PC handhelds when in handheld mode, nor to expect it to beat current gen consoles in docked mode.

Based on the battery rating of ~19.3WHrs and stated minimum battery life of 2 hours, the SoC will be limited to a very low 7-8W using Ampere tech on some 8nm derivative. And in docked mode, well it's still just 12 Ampere SMs, likely running closer to about 15-20W this time. That's not going to match up to a Series S in horsepower (even if the additional RAM will let it get some significant featureset wins).

[-]

theQuandary@reddit

I can't carry a Series S around with me. The only real comparison is the Steamdeck and the $400 steamdeck gives up a few things vs the $450 Switch 2.

The price is about what you'd expect for the features/performance.

[-]

uzzi38@reddit

I can't carry a Series S around with me. The only real comparison is the Steamdeck and the $400 steamdeck gives up a few things vs the $450 Switch 2.

The Switch 2 also gives up on several things too. Access to Steam (games will be and have already shown to be far more expensive on Nintendo's platform) and performance, in particular. Switch 2 with such a low power budget will almost certainly fall significantly behind the Steam Deck in terms of performance, and when you include the cost of pretty much 3 games you're already in Deck OLED territory in terms of pricing.

It wouldn't even take many games before you start to enter Windows PC handheld territory: and on that note Switch 2 will likely fall behind on docked performance too when compared against those devices. When you factor in the value proposition of Game Pass on those devices, Nintendo have a really poor value proposition on their hands.

The price is about what you'd expect for the features/performance.

No, from my point of view when you factor in the cost of games, the value proposition on Nintendo's part is pretty abysmal. It's by far and away the worst part of the Switch 2 release.

[-]

conquer69@reddit

I'm already hoping for a revision with power efficiency improvements and a bigger battery lol.

[-]

uzzi38@reddit

Well, one can definitely hope for something along those lines, potentially also with an OLED panel to boot. But I would be surprised to see a refresh any time in the next year or two, tbh. Maybe after that.

[-]

x3nics@reddit

Nintendo is held to a different standard compared to literally everybody else, it's maddening.

[-]

TheElectroPrince@reddit

That's what happens when you try and send enthusiasts to prison.

Nintendo understands the true meaning of why 60/120fps is important for console.

Unlike those 2 dumb console makers.

[-]

LasersAndRobots@reddit

I mean, let's not get ahead of ourselves. It'll run Super Mario and Mario Kart at 120 fps, pretty much every other first party title at 60 (for the first year or two) and every port at 30 or below.

The thing is expected to have about 70-odd percent of the PS4 Pro's horsepower in docked mode, and perform pretty similarly to a base PS4 in handheld. DLSS will definitely help it punch above its weight, but as we've seen in the PC space it's just as likely to become a crutch and will probably be leveraged to drive higher resolutions than higher framerates.

Funnily enough

I've been talking to some tech dudes all agreed that no game shown in the direct had dlss on from what they could tell

Still disappointed this isn't an OLED, but I hope it is at least a good quality LCD and not like the LCD Steam Deck.

VRR and 120Hz were surprising inclusions. This also makes 40 FPS games in 120Hz containers viable now too, which is sooooo much better than 30 FPS which I honestly cannot stand anymore.

This will be an awesome indie machine with lighter 2D indie games likely being able to push 120Hz modes, and even if they can't quite reach that target, VRR will come in to save the day.

[-]

ConsistencyWelder@reddit

As someone who has never had an OLED screen that didn't end up with burn in: I think people are better off without it. It will shorten the reasonable life span of the device considerably.

[-]

mrheosuper@reddit

Is it ?, my samsung phone with oled screen which is 5YO at this point still has perfect screen. And if nothing change, i bet i can get 1 or 2 years out of it

[-]

teutorix_aleria@reddit

Ive had multiple OLED phones and the burn in was only noticeable on static bright backgrounds. And ive spent way more time on tiktok than i would playing any singular game on a switch. Unless you are playing the same game with bright static elements 24/7 its not too much an issue IMO.

That said im happy with the LCD for the VRR support and resolution without making the thing 600$

[-]

Thoromega@reddit

Is the Wi-Fi chip still dogshit?

[-]

Johnny_Oro@reddit

Yeah no wifi 6E in a $450 device.

[-]

no_salty_no_jealousy@reddit

That's greed move by Nintendo. My $400 MSI Claw even has Intel Killer Wi-Fi 7.

[-]

$450 for overpriced overhyped Switch 2 trash?? No thanks Nintendont!! Not to mention switch game price going up, also you have to pay for their garbage online service. Not to mention switch lack so many game variety which makes this console even more lame.

Meanwhile there are so many Windows handheld gaming right now. I bought brand new MSI Claw Ultra 7 with much better specs than this overpriced switch for only $400, and you know what? My handheld has been gaming beast, it runs every game, even switch game runs better on my Claw than on switch itself with better graphics too, not to mention i can play every AAA game which isn't available on switch and i don't even need to spends stupid online paid service. Also Xbox gamepass is insanely cheap, i spends less money for hundreds game.

[-]