• Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.

Nintendo Switch Dev Kit Stats Leaked? Cortex A57, 4GB RAM, 32GB Storage, Multi-Touch.

Status
Not open for further replies.
I think it may have been me. I was saying- accounting for more modern architecture, CPU outperforming PS4/XB1, and 512GFLOPS GPU at the lowest (when docked)- it would be a safe bet to call this 5x stronger than a Wii U.

It seems I was wrong (or DF is using the wrong specs to base their assumptions on, but I'm starting to lose hope on that).

Lets just hope you arent also wrong about 16nm Maxwell..In my mind if Pascal isnt even being discussed anywhere then this thing is going 20nm Maxwell especially with those clocks...
 
Lets just hope you arent also wrong about 16nm Maxwell..In my mind if Pascal isnt even being discussed anywhere then this thing is going 20nm Maxwell especially with those clocks...

The DF article explicitly talks about Pascal features being part of the Switch SoC's customizations. They clearly don't know what the customizations are, but they speculate some of them may be a 16nm process (they say this in the video).
 

NateDrake

Member
The DF article explicitly talks about Pascal features being part of the Switch SoC's customizations. They clearly don't know what the customizations are, but they speculate some of them may be a 16nm process (they say this in the video).

And this may have been the root of the confusion among my sources. Like I've said, I know Nintendo looked into Pascal for Switch. If the custom Maxwell is making use of several Pascal features then I can see how confusion set in and the information relayed to me was a mix of truths.
 
Why do these developers think it will have such a huge install base?...Actually, let me rephrase that:

Why do they think it will have the right install base for them? Especially if the Wii and "casual market" have left such a bad taste in their mouth?

The Wii U had 12-13 Million in sales the 3ds is somewhere in the 60 Million range. Nintendo combined their hardware and software development teams from their portable and home console divisions to make the Switch and combine both user bases.

This makes it very very easy for Nintendo to use exclusive IPs from both their portable division and home console divisions to create a steady stream of first party content to drive the user base to levels the Wii U could never have imagined.

If you look at the teaser trailer for the Switch you dont see children in it at all , which developers noticed and commented on. They know the target audience the Switch is going for is different and if the casuals decide to hop on then it is cake.

Times are changing people are on the move more often and want to take their entertainment with them (this is why mobile is huge). The Switch does exactly what people wanted the Wii U to do which was go outside the house. The Switch just does this with cleaner hardware and without the confusion of the name and concept.. this is why they believe the base will be there.
 

Vena

Member
The DF article explicitly talks about Pascal features being part of the Switch SoC's customizations. They clearly don't know what the customizations are, but they speculate some of them may be a 16nm process (they say this in the video).

Is this speculation or based on their spreadsheet of documentation? If the latter, might actually be rather important in sorting out this confusing mess.
 

Hermii

Member
Then how do you explain the fan being in the tablet? Even at 768Hz docked why is a fan necessary? You don't have any concerns about battery life when docked, and the Pixel C is passively cooled at that clock rate, right?

This would be such a bizarre decision. Unless they decided to do this to get rid of the need for a fan?
Does the pixel c get hot and have a shitty battery life when playing a demanding game? Tablets aren't primarily gaming devices.the switch is designed to run at 100% capacity for hours.
 

NateDrake

Member
Is this speculation or based on their spreadsheet of documentation? If the latter, might actually be rather important in sorting out this confusing mess.

Now, to be fair to Nvidia, Tegra X1's Maxwell was the final iteration of the architecture and does have technological aspects that are found in Pascal: specifically, double-rate FP16 support. We're also told that Switch has bespoke customisations that may involve pulling in other Pascal optimisations.

From the EG article.
 

sfried

Member
If you look at the teaser trailer for the Switch you dont see children in it at all , which developers noticed and commented on. They know the target audience the Switch is going for is different and if the casuals decide to hope on it is cake.

Times are changing people are on the move more often and want to take their entertainment with them (this is why mobile is huge). The Switch does exactly what people wanted the Wii U to do which was go outside the house. The Switch just does this with cleaner hardware and without the confusion of the name and concept.. this is why they believe the base will be there.

So you're telling me "better advertising" is the Switch's secret weapon?

I kind of find that very hard to believe, and not to mention they would need to out-ad Sony and MS.
 

Vena

Member
From the EG article.

Hum... then the direct 1:1 correlation that Eurogamer is applying to the X1 seems to be shot. Can't imagine you'd customize the chip like that, and then underclock everything else without reason if the specs are just the default X1 specs.

On the flip, there's a good shot this actually is a 16nm Maxwell chip.
 

Lonely1

Unconfirmed Member
The way I'm reading the DF article is that they have solid info on the clocks (from a dev manual) but that they don't know much about the chip other than is Maxwell based. Their performance targets are based on applying the clocks to a stock TX1.

Is this correct?
 
And this may have been the root of the confusion among my sources. Like I've said, I know Nintendo looked into Pascal for Switch. If the custom Maxwell is making use of several Pascal features then I can see how confusion set in and the information relayed to me was a mix of truths.

Yeah, I've been saying this a lot over the past couple days. Even without any other customizations, simply a die shrink to 16nm while retaining Maxwell architecture might appear to be Pascal based to some people. Which is certainly how your sources may have heard Pascal.

Is this speculation or based on their spreadsheet of documentation? If the latter, might actually be rather important in sorting out this confusing mess.

Their spreadsheet only says anything about clock speeds. They base the rest of their speculation on the "leak" from the OP of this thread, and say that developers have told them that it is close to what Nintendo is telling them to expect. Close in what way, no one knows.

Does the pixel c get hot and have a shitty battery life when playing a demanding game? Tablets aren't primarily gaming devices.the switch is designed to run at 100% capacity for hours.

Based on benchmarks linked here it runs at max clocks for a decent period of time without throttling and with passive cooling only, though I have no idea how hot it gets. But the clocks for the Switch are even lower for portable- much lower- which means there should be absolutely no way the Switch requires active cooling in portable mode. I think a poster said the Switch, with these clocks, will have the same TDP as the Vita but a far bigger surface area.

When docked it's looking at about the same clock rate as the Pixel C, yet doesn't have to run on battery. So even docked it doesn't make much sense to require a fan. Let alone two fans, potentially.
 

Vena

Member
The way I'm reading the DF article is that they have solid info on the clocks (from a dev manual) but that they don't know much about the chip other than is Maxwell based. Their performance targets are based on applying the clocks to a stock TX1.

Is this correct?

Yes, and they also note that they believe there are features of Pascal that have been pulled into the chip (but it remains Maxwell).
 
So you're telling me "better advertising" is the Switch's secret weapon?

I kind of find that very hard to believe, and not to mention they would need to out-ad Sony and MS.

Better targeted marketing is part of the reason they believe it will succeed.. It is clear that the device is going to outperform the Wii U and the 3ds.. People who have a 3ds will want to upgrade at some point and the loyalists from the Wii U will move to the Switch because they want more first party games. Those 2 user bases will drive other third parties to consider making games for the platform or at least take a chance on it a few times.

I use to work in a retail store and I know exactly how hard the Wii U would have been to sell to people coming in. It was too confusing to sell to people outside the initial gold rush.

The Switch isnt going to have the Wii name working against it and some confusing concept. People are going to walk in and ask someone about the Switch and the sales associate is going to say it is the new Nintendo that you can play on your TV or on the Go and the name is going to click ..

Clever marketing does wonders.. How do you think Apple is able to recycle the same product with lesser specs than the competition every year with minor improvements and features and still sell millions?
 

Zil33184

Member
True. PS4 is 10x the Wii U give or take a flop.

GCN vs R700 comparison though, and the PS4 has almost 14x the memory bandwidth.

I wonder how Switch in portable mode will fare against the Wii U/last gen. Kind of puts into perspective how much you have to sacrifice performance in order to accommodate portability with consistent performance.

My expectations from the initial leak were full TX1 performance when docked and 40-50% downclock in portable mode. However, from the Eurogamer article we're seeing Vita levels of aggressive downclocks. The docked mode downclock is just bizarre unless the form factor has severe thermal issues and Nintendo want this thing to be as silent as possible.
 

bomblord1

Banned
And yet again, after Wii speculation, after WUST, the "lowest reasonable estimate" touted around still turns out to be more than twice the real thing.

Lowest reasonable expectation is still well within the realm of possibility given we don't know the core count
 

ZOONAMI

Junior Member
Yeah, I've been saying this a lot over the past couple days. Even without any other customizations, simply a die shrink to 16nm while retaining Maxwell architecture might appear to be Pascal based to some people. Which is certainly how your sources may have heard Pascal.



Their spreadsheet only says anything about clock speeds. They base the rest of their speculation on the "leak" from the OP of this thread, and say that developers have told them that it is close to what Nintendo is telling them to expect. Close in what way, no one knows.



Based on benchmarks linked here it runs at max clocks for a decent period of time without throttling and with passive cooling only, though I have no idea how hot it gets. But the clocks for the Switch are even lower for portable- much lower- which means there should be absolutely no way the Switch requires active cooling in portable mode. I think a poster said the Switch, with these clocks, will have the same TDP as the Vita but a far bigger surface area.

When docked it's looking at about the same clock rate as the Pixel C, yet doesn't have to run on battery. So even docked it doesn't make much sense to require a fan. Let alone two fans, potentially.

Hopefully the clocks are wrong or we have 3-4 sm
 

Mr Swine

Banned
From what I understand, this is what they're essentially doing (downclocking to conserve battery/reduce heat).

Yeah but I mean that they are doing it on both portable and console mode. That they are limiting both so that they can get good battery life and later down the road unlock it so that it can get 20-25% better performance or something like that.

I know that Devs could use more out of PSP after Sony released the firmware that unlocked the CPU/GPU speed
 

inner-G

Banned
Clever marketing does wonders.. How do you think Apple is able to recycle the same product with lesser specs than the competition every year with minor improvements and features and still sell millions?

Apple puts out expensive hardware that is regularly the very fastest available, and has a huge library of third party software. People don't just buy $1000 iPhones because of clever marketing.

They're the de facto smartphone because they provide the best user experience.
 

Ryoku

Member
And yet again, after Wii speculation, after WUST, the "lowest reasonable estimate" touted around still turns out to be more than twice the real thing.

Well, for one, you don't get an accurate measure of performance based on the clockspeeds alone.

The Switch is weak as hell if we assume it is using standard X1 chip (500 man-years for what?). Now how likely is that?
 

breakfuss

Member
Apple puts out expensive hardware that is regularly the very fastest available, and has a huge library of third party software. People don't just buy $1000 iPhones because of clever marketing.

They're the de facto smartphone because they provide the best user experience.

And don't explode
 

Lonely1

Unconfirmed Member
IMHO, a third SM would be unlikely, but I do expect much better memory setup than a standard TX1 (or any other mobile chip for that matter). In other words, a substantial amount of embedded, fast VRAM or other kind of memory cache. A stock TX1 downlocked to 300Mhz would have troubles with running Wii U/Ps3/360 software.
 
Apple puts out expensive hardware that is regularly the very fastest available, and has a huge library of third party software. People don't just buy $1000 iPhones because of clever marketing.

They're the de facto smartphone because they provide the best user experience.

You must not have met any iSheep.
 

Mr Swine

Banned
IMHO, a third SM would be unlikely, but I do expect much better memory setup than a standard TX1 (or any other mobile chip for that matter). In other words, a substantial amount of embedded, fast VRAM or other kind of memory cache.

Why would it likely not to have 3SM if it's clocked that low?
 

Mr Swine

Banned
Faster memory doesn't come for free.

Nothing does and I think that paying more now for 3SM will be beneficial for them in the long run. But then again they will probably go with 1 SM instead of 2

Why not simply go with a Tegra K1 and tinker it from there instead of a TX1?
 

ZOONAMI

Junior Member
Nothing does and I think that paying more now for 3SM will be beneficial for them in the long run. But then again they will probably go with 1 SM instead of 2

Why not simply go with a Tegra K1 and tinker it from there instead of a TX1?

Yeah honestly it makes no sense to even use the x1 and not k1 if you're going to downclock the x1 this much.
 

Lonely1

Unconfirmed Member
The way I see it, clocking 2SMs higher could require a bigger battery than paying for an extra SM and clocking them all lower. Meaning, the tradeoff might be battery cost vs die/core cost.

Embedded memory also takes no trivial ammount of die space.

Nothing does and I think that paying more now for 3SM will be beneficial for them in the long run. But then again they will probably go with 1 SM instead of 2

Why not simply go with a Tegra K1 and tinker it from there instead of a TX1?

Huh?
 

EVH

Member
Can anybody explain what this 2-3-4SM talk is about? What is a 3SM and the difference between 4SM? Why is it relevant?

I feel that you guys just found another thing to get your hopes up only to end dissapointed again.
 
Even if Nintendo somehow can go 16nm. There isn't enough place for such experiemts aka every CUDA unit more is expensive.

It doesn't make sense.
 

Raet

Member
About the battery (5-8 hrs) rumor, considering this is Nintendo we're talking about and their experience with handheld devices, isn't it possible they could have downgraded the chip to achieve this? It could be some variant of Maxwell 20 nm that is less powerful to achieve that kind of battery life.

Not saying this is is what I want. I'm hoping for 400-500 GFLOPS portable, 800 docked, but would be fine with 250-400.
Looks like I was onto something (unfortunately). Although it looks to be even more downclocked in portable mode to achieve even less than 250 GFLOPS.

And what's up with not running at full speed even in docked mode? Is this solely to keep the docked from being way too powerful, and keep the ratio at 2.5 so all that changes is the resolution to 1080p? Is it too farfetched to hope that retail units may be clocked at full speed in docked mode, assuming these numbers are from the devkit?

This will probably be a great system anyway, but I will now have to lower my expectations. Used to be a guaranteed day one purchase, now I will need to wait and see how the January conference turns out. Will probably still get it since I skipped out on the Wii U, but this news means third party games will not be as easy to port, so I will need to get it for Nintendo games. Which is fine by me, I prefer their games to almost any third party anyway.
 
A question:
Let's assume all of Nintendo's own games will be native res on the Switch portable (720p per rumors), would the GPU increase when docked result in a 1080p image most of the time?
The GFlops numbers pin the docked mode at being 2.5x more powerful than the portable mode which should be enough to cover doubling the pixel count to 1080p, no?
Nintendo's simpler art styles in 1080p or 720p on the go would look pretty great if that's the case.
 

Enduin

No bald cap? Lies!
A question:
Let's assume all of Nintendo's own games will be native res on the Switch portable (720p per rumors), would the GPU increase when docked result in a 1080p image most of the time?
Nintendo's simpler art styles in 1080p or 720p on the go would look pretty great if that's the case.

I keep seeing people say that you need 2.25 times greater GPU performance to go from 720p to 1080p, which these specs line up with, actually a little bit over that when docked. So yeah if devs all design their games around handheld mode at 720p they will then have enough headroom to increase the resolution to 1080p when docked.
 

Lonely1

Unconfirmed Member
Why even use an x1 if you're going to downclock it this much?

Downclocking the TX1 would yield better results than an equivalent TK1. Also, my opinion is that the switch doesn't have an stock TX1. I believe that a 300-750Mhz TX1 with access to faster memory would outperform a 3 SM TX1.
 
Can anybody explain what this 2-3-4SM talk is about? What is a 3SM and the difference between 4SM? Why is it relevant?

I feel that you guys just found another thing to get your hopes up only to end dissapointed again.

The way I understand it: Nvidia GPUs are comprised of shader modules (SMs) which contain 128 CUDA cores each, and these cores perform the graphical processing. The base Tegra X1 has 2SMs, meaining 256 CUDA cores and can perform 512 (EDIT: million) Floating point operations per second when clocked at 1GHz (core count x 2 x 1GHz).

We know the chip used by the Switch is custom, though we have no idea in what way it has been customized. The fact that it has at least one fan and vents though makes it seem like it is designed to draw enough power such that active cooling is required. The clock rates discussed in the DF article seem very low, likely too low to require active cooling if they are just assuming it's a base Tegra X1 at these clock rates. I think the portable mode system draws about as much power as a Vita, which doesn't have a fan or vents.

So speculation says, since we know there is a fan and vents, it could be possible that there are more SMs (likely not 4) running at this low clock rate which could draw enough power (and therefore generate enough heat) to require a fan, in both portable and docked modes.
 

LordRaptor

Member
A question:
Let's assume all of Nintendo's own games will be native res on the Switch portable (720p per rumors), would the GPU increase when docked result in a 1080p image most of the time?
Nintendo's simpler art styles in 1080p or 720p on the go would look pretty great if that's the case.

Its enough to comfortably handle higher resolution geometry and minor stuff like UI font size changes or whatever for basically 'free' (as in you wouldn't need to change anything or re-QA everything if it was just an internal resoltuion change from 720p to 1080p).

What would need extra work would be changing anything that was originally set for a 720p target resolution that you then wanted to increase to a 1080p target resolution - so things like draw distance, shadow resolution, particle count, VFX resolution, grass density, those sorts of bells and whistles.

Basically, it should be 'free' to just up geometry rendering but keep other visual elements at whatever they were set for at 720p
 

LordOfChaos

Member
A question:
Let's assume all of Nintendo's own games will be native res on the Switch portable (720p per rumors), would the GPU increase when docked result in a 1080p image most of the time?
The GFlops numbers pin the docked mode at being 2.5x more powerful than the portable mode which should be enough to cover doubling the pixel count to 1080p, no?
Nintendo's simpler art styles in 1080p or 720p on the go would look pretty great if that's the case.

The bandwidth however doesn't increase by nearly as much.


And unlike on say, the PS4 Pro, it's not going from an older GPU generation undocked to a newer one with newer compression while docked.


Not saying it will or won't , but there are considerations beyond GPU clock for that.
 

MisterR

Member
The way I understand it: Nvidia GPUs are comprised of shader modules (SMs) which contain 128 CUDA cores each, and these cores perform the graphical processing. The base Tegra X1 has 2SMs, meaining 256 CUDA cores and can perform 512 (EDIT: million) Floating point operations per second when clocked at 1GHz (core count x 2 x 1GHz).

We know the chip used by the Switch is custom, though we have no idea in what way it has been customized. The fact that it has at least one fan and vents though makes it seem like it is designed to draw enough power such that active cooling is required. The clock rates discussed in the DF article seem very low, likely too low to require active cooling if they are just assuming it's a base Tegra X1 at these clock rates. I think the portable mode system draws about as much power as a Vita, which doesn't have a fan or vents.

So speculation says, since we know there is a fan and vents, it could be possible that there are more SMs (likely not 4) running at this low clock rate which could draw enough power (and therefore generate enough heat) to require a fan, in both portable and docked modes.

Nintendo fans should have learned by now. When it's Nintendo assume the worst possible outcome when it comes to hardware. Always assume it's less powerful than you even imagined it could be. You'll usually be right and you won't be disappointed and who knows, maybe once in a blue moon you'll get a happy surprise.
 

Hermii

Member
I swear I read this exact same post in the WUST except swapping U instead of Switch.
Ok so I got a little overhyped.

Whatever, Nintendo games will still be great. I just hope it does decently financially.

I do think they should have made it two separate devices, like everyone thought based on Iwatas comments.
 

atbigelow

Member
Wonder if someone could chime in with some CPU benches. It seems troubling going from a faster tri-core PPC to a quad A57.

The new 3DS has a quad-core ARM CPU clocked at 804Mhz, but a very old one. Sure it won't be comparable at all, but definitely worth noting.
 
Nintendo fans should have learned by now. When it's Nintendo assume the worst possible outcome when it comes to hardware. Always assume it's less powerful than you even imagined it could be. You'll usually be right and you won't be disappointed and who knows, maybe once in a blue moon you'll get a happy surprise.

Yeah right now my expectations are in the gutter but I'm still hoping for the best. The other thing though is that we've still heard nothing but positive news regarding ports and third parties, so even if the specs are worse than we could have imagined I'm perfectly if we're getting a solid amount of multiplats. Solid meaning more than Wii U got.
 

Lonely1

Unconfirmed Member
Nintendo fans should have learned by now. When it's Nintendo assume the worst possible outcome when it comes to hardware. Always assume it's less powerful than you even imagined it could be. You'll usually be right and you won't be disappointed and who knows, maybe once in a blue moon you'll get a happy surprise.

The switch is more powerfull than what I expected form a Nintendo handheld 1 year ago, TBH.
 
I keep seeing people say that you need 2.25 times greater GPU performance to go from 720p to 1080p, which these specs line up with, actually a little bit over that when docked. So yeah if devs all design their games around handheld mode at 720p they will then have enough headroom to increase the resolution to 1080p when docked.
The GPU increase is 2.5x when docked and the pixel count increase from 720p to 1080p is 2.25.
I'd probably be pretty happy if they can manage. that.
720p on the handheld would look pretty great while 1080p on the TV is really all I need for now.
 

AzaK

Member
...

Urban legend SoCs aside, the article raises more questions than it answers.

Yeah like "WTF were Nintendo thinking"

It's roughly 157 GFLOPS for the portable mode and 393 GFLOPS for the docked mode assuming 2 SM.

Wow, that is laughable if true but what I fully expected from Nintendo. I don't see this boding well for third party support.

The GPU increase is 2.5x when docked and the pixel count increase from 720p to 1080p is 2.25.
I'd probably be pretty happy if they can manage. that.
720p on the handheld would look pretty great while 1080p on the TV is really all I need for now.


You'd really be happy with Wii U level graphics at 1080........4 years after Wii U and 10 years after equivalently powered competitor consoles?
 
Status
Not open for further replies.
Top Bottom