• Hey everyone, staff have documented a list of banned content and subject matter that we feel are not consistent with site values, and don't make sense to host discussion of on Famiboards. This list (and the relevant reasoning per item) is viewable here.
  • Furukawa Speaks! We discuss the announcement of the Nintendo Switch Successor and our June Direct Predictions on the new episode of the Famiboards Discussion Club! Check it out here!

StarTopic Future Nintendo Hardware & Technology Speculation & Discussion |ST| (New Staff Post, Please read)

Reader mode
Steve said:
Can you show the forum where the post was posted.

Also wouldn’t be surprised if a floodgate of leaks with either happen next week or the week after that.
Click to expand...
It's on ResetEra
www.resetera.com

Switch 2 | Release Speculation and Rumors [check threadmark for potential specs]

At GDC and while I doubt I can hunt down much info, Nintendo of America is here with a meeting only booth and it’s all behind closed doors stuff with them being careful of people passing by, so it’s got me going 👀 a bit for sure.
www.resetera.com www.resetera.com
 
Neoxon said:
It's on ResetEra
www.resetera.com

Switch 2 | Release Speculation and Rumors [check threadmark for potential specs]

At GDC and while I doubt I can hunt down much info, Nintendo of America is here with a meeting only booth and it’s all behind closed doors stuff with them being careful of people passing by, so it’s got me going 👀 a bit for sure.
www.resetera.com www.resetera.com
Click to expand...
kinda didn't want to flood this guy's inbox or put too much heat on him (because I think he might be a developer) which is why I didn't link and put it in hidden but whatever lmao, info will probably leak regardless
 
Last edited:
goombaicebro said:
* Hidden text: cannot be quoted. *
Click to expand...

bane-batman.gif

You merely adopted the cope; I was born in it, molded by it.
 
Hidden content is only available for registered users. Sharing it outside of Famiboards is subject to moderation.
 
MP! said:
Other than the fact it still exists I just don’t think we get much out of knowing it’s at gdc anyway…
Click to expand...
I'm leaning towards that camp too. Unless the physical device is there we likely won't get anything good imo
 
Mr Swine said:
Nintendo would start producing Switch 2 around October if they are targeting a March-April release?

Could they go with a late January early February release if possible?
Click to expand...
Waiting that long for a reveal would be crazy but I don’t even know what to expect anymore 😂
 
NZZN said:
Just found this low power Ampere mobile SKU with 2048 CUDA, 4.7 TFLOPS, 96 GB/s, and a 15 Watt TDP
Is this a closer PC match for GA20B than the 2050?
https://www.techpowerup.com/gpu-specs/geforce-mx570.c3919

fpsta7.png
kjai8z.png

Edit: RT/DLSS are disabled and it's even more memory starved than the 2050 in most configurations (4GB>2GB). Maybe not.
15 watts from a GA107 (which is 115w in the 3050) is interesting for scalability of SEC8N though.
Click to expand...

So.....8nm is back on the table? LOL Who knows, but this does make it seem a bit more possible. It has 25% more GPU cores than T239, can clock as high as 1155Mhz and still operate within the 15w TDP budget. It does make me wonder if the GDDR6 memory is a bigger power hog than we realize. The MX570 only has 2GB and perhaps this is a big reason why it can operate on a 15w TDP. T239 saves die space by having 25% fewer GPU cores, but it also has to fit the CPU cores, so it is probably pretty close to a wash. Maybe some further tweaks and optimizations could get it down under the 200mm2 dies size, but at 8nm, that's probably pretty close to what it would be.

The AYA Neo Air is about the same size as the Nintendo Switch and they use a 7350 or 10,000 mAh batteries (depending on configuration) compared to the 4310 mAh battery in the Switch. Leads me to believe that Nintendo would quite easily implement a battery with twice the capacity of the Switch without issue. The TDP for the Tegra X1 in portable mode is around 5w max I believe. With a battery that has twice the capacity, perhaps T239 could draw 8-10w and still maintain battery life north of 3 hours. The AYA Neo Air has a peak TDP for its APU of 18watts, there is no reason to believe that t239 couldn't draw 15-20watts in docked mode. I think a lot of our assumptions is that SNG wouldn't go with power draw above the original Switch, but if that isn't the case, then perhaps a more power hungry SOC is acceptable.

So why go with 12sms when they could have gone with 8sm, resulting in a smaller die size and clock it higher? Tensor and raytracing cores would be a possible answer. These cores are directly linked to the number of sm's. 8sm's might not have had enough Tensor cores to use DLSS and not enough ray tracing cores to make good use of that technology.

Not saying its going to be 8nm, I still hope it isn't, but after looking at the MX570 alongside the possibility that Nintendo settled on higher power draws than with the Switch thanks to a bigger battery, I would say its possible.
 
It's an easy rule-of-thumb: You announce stuff shortly before mass production starts.

That's when more people and hands get on board, so it's the moment where you risk leaks because the situation gets difficult to control.
 
Goodtwin said:
Some games/game engines just hit the CPU harder than others, even if there doesn't appear to be anything that should be pegging it.
Click to expand...
KarlRemarx said:
Or...or... we could take the developer at their word when they explicitly say that the game is intensive on older CPUs due to the nature of what they're simulating for the player and large numbers of NPCs in a dynamic world
Click to expand...
Goodtwin said:
Ill wait for the performance test to roll out, but unless we are seeing multiple CPU cores maxed out, then it is just another example of the game not being properly multi threaded.
Click to expand...
Two things can be true. A game's load can be caused by the high level of simulation, and the engine can be straining to take advantage of modern CPU designs. Video games are basically a worst case scenario for multi-core/multi-threaded hardware. It'll be interesting to see if the 10th gen goes with 12 cores, or stays with 8. I imagine they stay with 8, simply because engines aren't taking sufficient advantage of multi-core to make the cost of the extra hardware worth it.

I don't know of any engines that are "good" at multi-threading come to think of it. I'd be curious what folks think the best examples of multi-threading are on the video game side. To my knowledge, gaming engines are still using separation-of-concerns style threading models, with some work-sharing in the subsystems, rather than a global work-sharing (or even work-stealing) model.
 
Hermii said:
In some capacity maybe l, but probably not mass production.

For the simple reason that it won't take long from mass production starts until there's a leak. Nintendo probably want to get something official out there ahead of that.
Click to expand...
It's probably a batch or test production. Essentially a smaller scale production period in a controlled space to see how it would go with mass production. To test for production faults and potential defect issues with consoles. Mass production in other, contracted manufacturing plants would likely happen when they're confident in mass production.
 
oldpuck said:
Two things can be true. A game's load can be caused by the high level of simulation, and the engine can be straining to take advantage of modern CPU designs. Video games are basically a worst case scenario for multi-core/multi-threaded hardware. It'll be interesting to see if the 10th gen goes with 12 cores, or stays with 8. I imagine they stay with 8, simply because engines aren't taking sufficient advantage of multi-core to make the cost of the extra hardware worth it.

I don't know of any engines that are "good" at multi-threading come to think of it. I'd be curious what folks think the best examples of multi-threading are on the video game side. To my knowledge, gaming engines are still using separation-of-concerns style threading models, with some work-sharing in the subsystems, rather than a global work-sharing (or even work-stealing) model.
Click to expand...
I've been racking my brain to think of any examples and genuinely can't think of anything. The Last of Us remake on PC is "good" at multi-threading, in that it uses all of your CPU cores to render an empty warehouse with two fuckin bottles, but I wouldn't say it's a legitimately good example of multi-threading. Truthfully, I don't think we'll see any engines or video games that are good at multi-threading for a looooooong time just due to the nature of how video games are made, how many engines are tied to older logic and design principals (hello Creation Engine), and partially due to the variability in spec on the PC market.

Engine development isn't my wheelhouse at all, but I'd imagine changing to a global work-sharing or work-stealing multi-threading model would require a topdown rewrite of the engine, and at that point is it even worth the investment if you're not Epic Games or some other middleware company?
 
Goodtwin said:
So.....8nm is back on the table? LOL Who knows, but this does make it seem a bit more possible. It has 25% more GPU cores than T239, can clock as high as 1155Mhz and still operate within the 15w TDP budget. It does make me wonder if the GDDR6 memory is a bigger power hog than we realize. The MX570 only has 2GB and perhaps this is a big reason why it can operate on a 15w TDP. T239 saves die space by having 25% fewer GPU cores, but it also has to fit the CPU cores, so it is probably pretty close to a wash. Maybe some further tweaks and optimizations could get it down under the 200mm2 dies size, but at 8nm, that's probably pretty close to what it would be.

The AYA Neo Air is about the same size as the Nintendo Switch and they use a 7350 or 10,000 mAh batteries (depending on configuration) compared to the 4310 mAh battery in the Switch. Leads me to believe that Nintendo would quite easily implement a battery with twice the capacity of the Switch without issue. The TDP for the Tegra X1 in portable mode is around 5w max I believe. With a battery that has twice the capacity, perhaps T239 could draw 8-10w and still maintain battery life north of 3 hours. The AYA Neo Air has a peak TDP for its APU of 18watts, there is no reason to believe that t239 couldn't draw 15-20watts in docked mode. I think a lot of our assumptions is that SNG wouldn't go with power draw above the original Switch, but if that isn't the case, then perhaps a more power hungry SOC is acceptable.

So why go with 12sms when they could have gone with 8sm, resulting in a smaller die size and clock it higher? Tensor and raytracing cores would be a possible answer. These cores are directly linked to the number of sm's. 8sm's might not have had enough Tensor cores to use DLSS and not enough ray tracing cores to make good use of that technology.

Not saying its going to be 8nm, I still hope it isn't, but after looking at the MX570 alongside the possibility that Nintendo settled on higher power draws than with the Switch thanks to a bigger battery, I would say its possible.
Click to expand...
I seen this argument thrown around, but doesn't the AI/ RT performance scale with the clock speeds of the rest of the gpu? If so, why would it help having more if they're slower clocked?
 
The explanation for 12SMs is simply the chip's performance target, 8 or 10 SMs would have to use very high clocks to offer the performance that Nintendo is looking for.
 
lattjeful said:
Engine development isn't my wheelhouse at all, but I'd imagine changing to a global work-sharing or work-stealing multi-threading model would require a topdown rewrite of the engine, and at that point is it even worth the investment if you're not Epic Games or some other middleware company?
Click to expand...
Yeah, I assume the same. And I know I'm in the minority in the video game space, but I'm okay with developers using the power of new hardware to make games cheaper/faster (by letting the hardware brute-force through legacy engine tech) rather than requiring that every game that has ambition to also have the most up-to-date engine.

Work-stealing and work-sharing are the "fastest" designs, but they also have latency costs. One of the real struggles of game engine development is that you need to optimize for both latency and throughput. The fastest game engine on the planet is not good if you have a second of input lag.
 
oldpuck said:
Two things can be true. A game's load can be caused by the high level of simulation, and the engine can be straining to take advantage of modern CPU designs. Video games are basically a worst case scenario for multi-core/multi-threaded hardware. It'll be interesting to see if the 10th gen goes with 12 cores, or stays with 8. I imagine they stay with 8, simply because engines aren't taking sufficient advantage of multi-core to make the cost of the extra hardware worth it.

I don't know of any engines that are "good" at multi-threading come to think of it. I'd be curious what folks think the best examples of multi-threading are on the video game side. To my knowledge, gaming engines are still using separation-of-concerns style threading models, with some work-sharing in the subsystems, rather than a global work-sharing (or even work-stealing) model.
Click to expand...
The only recent game I could find that has really good multithreading is Mount & Blade II: Bannerlord, and that’s the exception that proves the rule since it has a proprietary engine that had to be redesigned 3 separate times because of multithreading.
 
oldpuck said:
And I know I'm in the minority in the video game space, but I'm okay with developers using the power of new hardware to make games cheaper/faster (by letting the hardware brute-force through legacy engine tech) rather than requiring that every game that has ambition to also have the most up-to-date engine.
Click to expand...
I will die on that hill with you. There's something to be said about already knowing how your engine is designed and functioned. That familiarity may be able to take you far and squeeze a bit more out of the engine, despite the on-paper performance cost being higher. Not to mention the time savings that come with knowing the tech like the back of your hand, and being able to onboard new hires quicker because of it. Time is money, after all.

What good is constantly chasing the latest and greatest if you have to relearn and redo your entire workflow, and slow down development as a result? Kinda goes against the idea of having your own tools to streamline development lol
 
Hermii said:
I seen this argument thrown around, but doesn't the AI/ RT performance scale with the clock speeds of the rest of the gpu? If so, why would it help having more if they're slower clocked?
Click to expand...

Perhaps, but the clock speeds would have to be about 35% higher to offset the reduced core count. Again, my theory would assume that portable clock speeds would be right at 8nm Ampere peak efficiency, roughly 550-650Mhz I believe? Because an 8sm chip would have to be clocked 35% higher just to reach the same performance, it will operate in a less efficient parts of the power draw curve. What is the power draw for a 12sm chip at 650 Mhz compared to 8sm's at 1Ghz? Even worse on the bell curve would be docked power draw where the 12sm chip could be clocked at 1.1Ghz vs a 8sm chip at 1.6Ghz. The 8sm chip might actually draw more power at those clock speeds compared to the 12sm chip at the lower speed.

This is obviously just spit balling ideas and it of course is 8sm 8nm vs 12sm 8nm. There is no argument to me made against 4N in terms of performance or efficiency. Just looking 8nm and how a 8sm design might compare to a 12sm design. Once we remove the power budget limitations of the Switch and instead increase the power budget to match the efficiency sweet spot for 8nm, then things can start to make sense. As long as we are under the impression that the TDP for the SOC cannot be higher than 5w, it makes it impossible to square that with 8nm, but if that power budget is now 8-10 watts for portable mode and 15-20 watts for docked, suddenly 8nm is plausible again.

Not advocating for 8nm, still hoping its 4N, but if it does end up being 8nm, perhaps a higher power draw was an accepted part of the design from the get go.
 
Maxrunner said:
whats so terrible about the ram besides the quantity?
Click to expand...
It's slow, as @ItWasMeantToBe19 points out. It also has a two-pool design, which isn't inherently bad, but in the Series S causes additional problems.

Series S has two pools of memory, an 8GB pool which runs at 224 GB/s and a second 2GB pool which runs at 56 GB/s. Ideally, developers never need to tap into that second pool, and you just let the OS live there, and you go about your life. But that 8GB pool is pretty constrained - it's less than the One X had, after all, and much less than the Series X/PS5 - so developers often need to tap that second pool.

Getting the best performance out such a setup requires really carefully managing your resource allocation, something no other platform needs. The failure mode is that memory performance falls off a cliff somewhat unpredictably - texture streaming lags, or the GPU and the CPU stall out waiting for data from RAM.
 
eye0h said:
I'm leaning towards that camp too. Unless the physical device is there we likely won't get anything good imo
Click to expand...
yeah exactly I mean, even then we wouldn't get ram amounts or clock speeds, storage amount, probably wouldn't have any clue about new joycons or other features

I would even argue when we found out about "the zelda demo" and "the matrix demo" we still only learned what we already knew... that a new device could play old games at higher frames and resolutions... and that it could handle some type of modern current gen feature set.

Though we did get fast load times confirmed I suppose, still doesn't tell us much...

Anyway i'm repeating myself for no reason but there's just not a lot of new info we could get from another demonstration behind closed doors.
 
Steve said:
And the only thing I can hope is that they’ll improve with their smaller budget games and try giving them either more time or more budget.
Click to expand...
HankyPanky said:
just played the Peach Demo and got damn the frame rate is rough to the extent where is detracts from the experience. if first party games can't maintain a stable framerate they must understand a new system is more than overdue.
Click to expand...

Many games that are clearly more demanding than PPS run without a hitch on Switch 1. Similarly, Good Feel's previous game, Otogi Katsugeki Mameda no Bakeru: Oracle Saitarou no Sainan!!, doesn't seem to be the most technically impressive game in the Switch 1's library, and yet also has fluidity problems. I couldn't agree more with Steve: it's all about time and budget, in other words, the absolutely crucial issue here is optimization.This is by no means a criticism of Good Feel, but rather an acknowledgement that development conditions are important, and that we can't constantly just talk about the hardware.

It's a question that extends far beyond Nintendo: transistors won't be able to shrink indefinitely, rapid power gains and the "pro" versions of such and such consoles won't be able to suffice indefinitely. Reducing everything to a question of chips or power at a time when Tears of The Kingdom has just received an award for its technology seems questionable, to say the least.

I'm not at all saying that power doesn't matter, of course. That's not what I think at all. However, by necessity, you have to deal with energy, economic, technical and release schedule constraints. Even when you're a Nintendo EPD and you're recognized for your expertise in optimization, you're impacted by all these constraints. We're already starting to think about the possible power of a Nintendo Switch 3 to rival the PS5, but the limits of the race for power even apply to home consoles at the moment, when unlike the Switch they don't have to deal with the limitations induced by the potentially portable nature of an hybrid console.

Of course, the exciting thing about technology is that things are always changing, and innovations can be unpredictable. However, after years and years of discourse centered on power gains, I get the impression that we might have entered the era of time gains. I'm the first to jump up and express my disagreement when I read that the developers of Mario Odyssey have supposedly "done nothing since". The fact remains that Nintendo is not immune to the problems raised by the significant increase in development time. They make huge efforts to optimize most of the time, but this has its drawbacks, or they also produce less demanding games, but this has other drawbacks, as we can see with sports games for instance. I think the solutions to this challenge lie in more than just power. And perhaps through automation processes that I don't necessarily find reassuring for developers.
 
Mr Swine said:
Nintendo would start producing Switch 2 around October if they are targeting a March-April release?

Could they go with a late January early February release if possible?
Click to expand...
I imagine there's only so much they can do to slow/delay the production side of things, especially if the hardware is complete. The last Funcle related insight said something about July, it seems plausible mass production will kick off around then regardless of whether the system was coming Q4 or Q1 2025. I think it's easier to run with the software holdup theory given everything we know, which might mean a reveal sooner than otherwise out of neccesity.
 
0
oldpuck said:
in
Click to expand...

NZZN said:
Just found this low power Ampere mobile SKU with 2048 CUDA, 4.7 TFLOPS, 96 GB/s, and a 15 Watt TDP
Is this a closer PC match for GA20B than the 2050?
https://www.techpowerup.com/gpu-specs/geforce-mx570.c3919

fpsta7.png
kjai8z.png

Edit: RT/DLSS are disabled and it's even more memory starved than the 2050 in most configurations (4GB>2GB). Maybe not.
15 watts from a GA107 (which is 115w in the 3050) is interesting for scalability of SEC8N though.
Click to expand...
So... SEC8N is not off the table after all...kkkkk
 
0
Kise Ryota said:
I confess you got me curious on this one. Could you bring an example where using MB/ms makes it easier to grasp what's going on?
Click to expand...

If you say "500 GBs per second!" it caused me to subconsciously wonder what the big deal was with RAM speed. Who cares about 100 GB/s vs. 500 GB/s? The entire game is probably less than 150 GBs! We're getting in that less than a second!

But you only have 16.66 or 33.33 ms of frametime for a game and then 100 MB/ms vs. 500 MB/ms comes across as much more relevant.
 
Goodtwin said:
Perhaps, but the clock speeds would have to be about 35% higher to offset the reduced core count. Again, my theory would assume that portable clock speeds would be right at 8nm Ampere peak efficiency, roughly 550-650Mhz I believe? Because an 8sm chip would have to be clocked 35% higher just to reach the same performance, it will operate in a less efficient parts of the power draw curve. What is the power draw for a 12sm chip at 650 Mhz compared to 8sm's at 1Ghz? Even worse on the bell curve would be docked power draw where the 12sm chip could be clocked at 1.1Ghz vs a 8sm chip at 1.6Ghz. The 8sm chip might actually draw more power at those clock speeds compared to the 12sm chip at the lower speed.

This is obviously just spit balling ideas and it of course is 8sm 8nm vs 12sm 8nm. There is no argument to me made against 4N in terms of performance or efficiency. Just looking 8nm and how a 8sm design might compare to a 12sm design. Once we remove the power budget limitations of the Switch and instead increase the power budget to match the efficiency sweet spot for 8nm, then things can start to make sense. As long as we are under the impression that the TDP for the SOC cannot be higher than 5w, it makes it impossible to square that with 8nm, but if that power budget is now 8-10 watts for portable mode and 15-20 watts for docked, suddenly 8nm is plausible again.

Not advocating for 8nm, still hoping its 4N, but if it does end up being 8nm, perhaps a higher power draw was an accepted part of the design from the get go.
Click to expand...
If by some wizardry they are able to hit peak efficiency in handheld mode 8nm makes sense. If they are not, it wont be 8nm. That's where Im at.
 
NZZN said:
Just found this low power Ampere mobile SKU with 2048 CUDA, 4.7 TFLOPS, 96 GB/s, and a 15 Watt TDP
Is this a closer PC match for GA20B than the 2050?
https://www.techpowerup.com/gpu-specs/geforce-mx570.c3919

fpsta7.png
kjai8z.png

Edit: RT/DLSS are disabled and it's even more memory starved than the 2050 in most configurations (4GB>2GB). Maybe not.
15 watts from a GA107 (which is 115w in the 3050) is interesting for scalability of SEC8N though.
Click to expand...
I'm not going to lie, but removing the 2GB DDR6, and counting for 12SMs @ 660MHz, we get just above 4.2W consumption, if we count improvements such as clock gating seems like a possible target.
 
Reader mode
You must log in or register to reply here.
Please read this new, consolidated staff post before posting.

Furthermore, according to this follow-up post, all off-topic chat will be moderated.
Last edited by a moderator:


Back
Top Bottom