• 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.

A Nintendo Switch has been taken apart

z0m3le

Banned
Sorry, but it is not what it looks like.
OK, there is people that assuming the clocks and other configurations that DF leak as the worst case, when switch is a ~400 GFlops machine with 3 under clocked A57 chips and 25Gb/s bandwidth.

But there is there other side too.
There are people that think the switch use a 16mm pascal custom Tegra. That gives numbers close to 600 GFlops. But they talking more how Nvidia GFlops is roughly 30% better than AMD (what let switch with a performance close to a 780 GFlops AMD in theory), plus the advantage of it utilize FP16, what can increase that number roughly by 50%, what let it at 900 GFlops with mixed precision, or equivalent to a 1170 Gflop GPU from AMD, that is, basically, a Xbox one.
They sum it with the possibility of it use A72 on CPU, what, even with 4 cores, can put switch in the same ballpark as Ps4 in that area.

So yes, there is people that are hoping to a "almost" Xbox one machine.

You want to point out a single post in this thread from such a person? I'm pretty optimistic and even I think fp32 472gflops is the best we can assume and that the nvidia advantage over AMD is going to be in the TEENS at best, fp16 being the best case scenario would allow it to be noticeably weaker than XB1, but run the same games at reduced quality, if all the stars align that is.
 
https://www.reddit.com/r/NintendoSw...the_teardown_and_pc_analysis_the_cpu/ddz3hp4/

Reading through this, seems that blue test sheet floating around might be from the tear down unit, meaning it is certainly a devkit as retail units don't have root access. Note, the 512 cuda cores is likely just the system reading the 256 cores twice with FP16 being used.

The Denver cores there are a big red flag to me. Going from A57s in one devkit to Denver in another would be a problem for developers, right? That honestly doesn't look like the most reliable leak.

Sorry, but it is not what it looks like.
OK, there is people that assuming the clocks and other configurations that DF leak as the worst case, when switch is a ~400 GFlops machine with 3 under clocked A57 chips and 25Gb/s bandwidth.

But there is there other side too.
There are people that think the switch use a 16mm pascal custom Tegra. That gives numbers close to 600 GFlops. But they talking more how Nvidia GFlops is roughly 30% better than AMD (what let switch with a performance close to a 780 GFlops AMD in theory), plus the advantage of it utilize FP16, what can increase that number roughly by 50%, what let it at 900 GFlops with mixed precision, or equivalent to a 1170 Gflop GPU from AMD, that is, basically, a Xbox one.
They sum it with the possibility of it use A72 on CPU, what, even with 4 cores, can put switch in the same ballpark as Ps4 in that area.

So yes, there is people that are hoping to a "almost" Xbox one machine.

First of all, can you back up your claim that Snake Pass on XB1 is 1080p/60fps? I've never seen any XB1 footage and the PS4 version is 1080p 60fps but not a locked 60fps, while the Switch footage is 1080p and locked 30fps. And was also ported just a month ago, so it might not be the final version.

And second, we're saying it's possible that the Switch can get close to those levels effectively for exclusives, and not likely for ports. Any console will perform a lot better for a properly optimized exclusive than it will for a multiplat, this isn't some magic or secret sauce. The Switch is no exception there. If the Eurogamer clocks are final, we'll essentially have a ceiling of "effective PS4/XB1 GFlops" around 800-900 I think. A bit higher for the Foxconn clocks.
 

Schnozberry

Member
The Denver cores there are a big red flag to me. Going from A57s in one devkit to Denver in another would be a problem for developers, right? That honestly doesn't look like the most reliable leak.

Anybody seriously considering the idea that Nintendo would move from A57 to Denver just for giggles in between dev kit revisions is huffing paint.
 

z0m3le

Banned
The Denver cores there are a big red flag to me. Going from A57s in one devkit to Denver in another would be a problem for developers, right? That honestly doesn't look like the most reliable leak.
Anybody seriously considering the idea that Nintendo would move from A57 to Denver just for giggles in between dev kit revisions is huffing paint.
He is assuming a custom CPU, not knowing that it is one. Though Nvidia's Tegra Volta is suppose to have a new custom CPU iirc, so who knows, since we have no confirmation of final hardware. This is all google translated, so not the best way to read this stuff.

The big reason to believe this IMO? it's a defective unit, that he seems to have access to, the 4th CPU doesn't read right according to the picture as far as I can tell. Also this picture happened within a day of the tear down and from china which is where the tear down came from... Seems like a pretty reasonable assumption to look this leak over more.

http://tieba.baidu.com/photo/p?kw=n...d81463e01213fb80ebd847d433fd12f2eb8389478/pn1 here is the original, maybe someone can translate this?

Edit: was there any back story to the unit in the OP? China doesn't have a retail unit for sale AFAIK, wouldn't this unit being in China mean it is a defective unit, such as a devkit? I don't think it would be possible to get your hands on a retail unit very easily, since they ship from Foxconn directly outside the company and defective units are destroyed, right?
 

KingSnake

The Birthday Skeleton
https://www.reddit.com/r/NintendoSw...the_teardown_and_pc_analysis_the_cpu/ddz3hp4/

Reading through this, seems that blue test sheet floating around might be from the tear down unit, meaning it is certainly a devkit as retail units don't have root access. Note, the 512 cuda cores is likely just the system reading the 256 cores twice with FP16 being used.

I just can't read through that, I get a headache. Is it translated with Google?

Can you quote the part that says that the benchmark is running on the teared-down unit in the OP? And why is it believable?

Edit: I managed to push myself to read the whole thing, I still don't see how it's linked with the unit in the OP. Bonus: somebody in the main thread warning that the poster of the images is a scammer.
 

MCN

Banned
If Switch can't make a game like that run in the same quality of the Xbox one, then, it is weaker to that machine, don't matter how much FP16 and Nvidia GFlops help it.

*looks at size of Xbox One*

*looks at size of Switch*

Well no fucking shit, Sherlock.
 
I can confirm to mods that the Switch is likely X1 based, and the CPU is likely woefully underpowered. I am a developer working on titles on other platforms. I do not work on Switch titles or have access to Switch SDK. I came across it through an alternative source (which is legit). If a mod wants to verify this information I will be happy to do so under the condition that the raw information presented not be published.
Why edit the post?
 

blu

Wants the largest console games publisher to avoid Nintendo's platforms.
This is absolutely not true. We have done FP16 before on the RSX, and other platforms, and that tale is full of woe discovering where the precision breaks down in lighting and you need to do local promotion to avoid artifacts. And on architectures that require additional instructions to pack/unpack into registers, whether the ALU savings amortize out over the shuffling overhead.
A multitude of mobile GPUs that are still in use today use only fp16 in their fragment shaders.
BTW, what's local promotion?
 
I can confirm to mods that the Switch is likely X1 based, and the CPU is likely woefully underpowered. I am a developer working on titles on other platforms. I do not work on Switch titles or have access to Switch SDK. I came across it through an alternative source (which is legit). If a mod wants to verify this information I will be happy to do so under the condition that the raw information presented not be published.

"Woefully underpowered"

As compared to what? You are a dev, looking at the most impressive/powerful portable system ever and you say the CPU is woefully underpowered? Ok.

EDIT: Oh, I see you deleted your original post. Changed your mind about being mod-vetted hey?
 

Polygonal_Sprite

Gold Member
I still find it mind blowing that a technology company will not list its tech specs and we have to guess or take it apart to find out what is in it.

Can you imagine someone like Apple releasing a new iPhone and not telling anyone what is inside and just expecting people to buy it? What about TVs or virtually any other electronics sold? Specs don't matter to everyone but they should be out there for the people that care.

Nintendo want you to focus on the Switch unique selling point - a modern console that you can take on the go with you. Not raw specs which would be directly compared to PS4 Pro and Scorpio specs and laughed at.

I like your avatar :)
 
I still find it mind blowing that a technology company will not list its tech specs and we have to guess or take it apart to find out what is in it.

Can you imagine someone like Apple releasing a new iPhone and not telling anyone what is inside and just expecting people to buy it? What about TVs or virtually any other electronics sold? Specs don't matter to everyone but they should be out there for the people that care.

Nintendo want you to focus on the Switch unique selling point - a modern console that you can take on the go with you. Not raw specs which would be directly compared to PS4 Pro and Scorpio specs and laughed at.

I like your avatar :)

I'm pretty sure Nintendo stopped releasing specs for their consoles ever since Sony managed to twist their Gamecube specs such that people still to this day think it was way weaker than the PS2. They now see no benefit in officially announcing specs if anything they say can be turned against them in some way.

Better to let everyone expect very little then (potentially) exceed expectations.

If Eurogamer is right the CPU is woefully underpowered.

In your original post why did you keep saying "likely" for everything if you have a source with the official specs?
 
Guys, he edited for a reason I'm sure, leave him alone.

Maybe he decided not to risk his job/contact's job, which is fair enough.

It's just a little odd to me to claim to have a legit source showing specs, then claim that it's likely X or Y, rather than saying it definitely. Oh well, if he/she doesn't want it out there that's fine.

Has anyone figured out yet what the unit from the OP is? A final retail unit really doesn't make sense due to showing up torn down on a Chinese website days before the press are getting theirs.
 

blu

Wants the largest console games publisher to avoid Nintendo's platforms.
If Eurogamer is right the CPU is woefully underpowered.
Switch's SoC has been 'indirectly' confirmed to be TX1 based to hell and back. I'm not sure what you're trying to prove here.

BTW, for the record: the one canonical way to confirm what uarch Switch uses would be to read MIDR_EL1 register (main ID register, Exception Level 1), which holds that info in plain sight. Alas, that register is normally above user-space privilege (Exception Level), and it's upon the good will of the kernel to pass down this info. Or pass it down truthfully, and not as 0xdeadbeef.

I'm pretty sure local promotion would mean using FP32 in intermediate steps (that is, input and output are FP16 but the calculations inbetween might have to be higher precision to avoid inaccuracy).
Ok, makes sense. For some reason when I read it I was thinking in graphics algorithms terms, and not just straight type promotion.

Though I guess the main reason you'd do that instead of straight FP32 would be for memory usage or bandwidth reasons?
Not necessarily. For instance, you may have an accumulator that gathers quantities from a loop - even though all such quantities might be fp16, and the accumulator be eventually reduced back to fp16 before further use, the very act of doing the accumulations as fp32 could produce more precise results.
 

sits

Member
It's just a little odd to me to claim to have a legit source showing specs, then claim that it's likely X or Y, rather than saying it definitely.

I'm guessing his info leads him to believe DF clocks are final, but he doesn't know definitively.

Next Friday can't come soon enough, I say.....
 

pulsemyne

Member
https://www.reddit.com/r/NintendoSw...the_teardown_and_pc_analysis_the_cpu/ddz3hp4/

Reading through this, seems that blue test sheet floating around might be from the tear down unit, meaning it is certainly a devkit as retail units don't have root access. Note, the 512 cuda cores is likely just the system reading the 256 cores twice with FP16 being used.

Which would make sense as the Gflops mentioned would be doable for a Tegra in half precision. There's nothing surprising about that number.
 

matthewuk

Member
I'm not entirely sure the statement that the switch CPU is woefully underpowered is entirely fair. Just by looking at dmips. Is equivilent to a 2ghz espresso for general purpose code. And in terms of simd and other features it's a huge leap. I think for for Devs it will be a case of "it's nothing special but we can work with it and at least it's not killing us like the Wii u ppc did"
 

Vena

Member
I'm not entirely sure the statement that the switch CPU is woefully underpowered is entirely fair. Just by looking at dmips. Is equivilent to a 2ghz espresso for general purpose code. And in terms of simd and other features it's a huge leap. I think for for Devs it will be a case of "it's nothing special but we can work with it and at least it's not killing us like the Wii u ppc did"

I think its more that its without context. If the CPU is well balanced relative to the rest of the hardware than the context makes the statement meaningless, or if not then the statement makes sense. But to just compare it against different hardware in completely different regimes and form factors, and call that woefully underpowered is a case of missing the forest for the trees.

Isn't that the same for ps4 Jaguar? They're also underpowered...

Its really more a matter of bottlenecks and imbalances in hardware, and of course what is tenable to achieve at a given time for a given price.

Running a Pentium III with a 1080Ti is woefully underpowered. :p
 

Padinn

Member
*looks at size of Xbox One*

*looks at size of Switch*

Well no fucking shit, Sherlock.

I know the point you are making but comparing the size of an electronic doesn't really mean much. Otherwise those 60s era pcs would rock. Maybe switch is a quantum computer ;)

More seriously while the bandwidth of switch appears low the Maxwell chips have some very nice tech to make it less impactful overall. Tile based rendeting and delta color compression so itll punch anove weight. I am hoping for 60% of an xb1 in docked mode.
 

BDGAME

Member
You want to point out a single post in this thread from such a person? I'm pretty optimistic and even I think fp32 472gflops is the best we can assume and that the nvidia advantage over AMD is going to be in the TEENS at best, fp16 being the best case scenario would allow it to be noticeably weaker than XB1, but run the same games at reduced quality, if all the stars align that is.

Good to know your true expectations.
In defense of myself, tell me, Do you never read people thinking it can use pascal and have 600 or more GFlops? Am I the first to tell about the possibility of it use A72 CPU? How many people do you seem hopping for a 30% better permanece in practice than in paper because Nvidia?

The only thing I do is put all these expectations together.


First of all, can you back up your claim that Snake Pass on XB1 is 1080p/60fps? I've never seen any XB1 footage and the PS4 version is 1080p 60fps but not a locked 60fps, while the Switch footage is 1080p and locked 30fps. And was also ported just a month ago, so it might not be the final version.

And second, we're saying it's possible that the Switch can get close to those levels effectively for exclusives, and not likely for ports. Any console will perform a lot better for a properly optimized exclusive than it will for a multiplat, this isn't some magic or secret sauce. The Switch is no exception there. If the Eurogamer clocks are final, we'll essentially have a ceiling of "effective PS4/XB1 GFlops" around 800-900 I think. A bit higher for the Foxconn clocks.

http://m.neogaf.com/showthread.php?t=1342575&page=100000

People assume it is 1080p 60fps on that thread and no one try to fix it. So I assume it's true.


*looks at size of Xbox One*

*looks at size of Switch*

Well no fucking shit, Sherlock.

Size? Really?

I mean, is there a weak argument than the size?
 

Schnozberry

Member
I'm not entirely sure the statement that the switch CPU is woefully underpowered is entirely fair. Just by looking at dmips. Is equivilent to a 2ghz espresso for general purpose code. And in terms of simd and other features it's a huge leap. I think for for Devs it will be a case of "it's nothing special but we can work with it and at least it's not killing us like the Wii u ppc did"

Yeah, as with everything, you have to ask "compared to what?"

It's a huge upgrade from Wii U, and it fits in a tiny little tablet form factor. You can say "they should have crammed in 4SM's, 4 A73 + 4 A53 cores, 6GB of RAM on a 128-bit bus, an all day battery and a 1080p display" all day, but then again who is going to buy it for $600-700?
 

psyfi

Banned
I got to hold one today when I ran into a Nintendo rep at my local GameStop. Damn thing was ridiculously light. Lighter than a Vita.
That's great to hear. I was kinda worried it'd be too heavy to play in bed for long play sessions.
 
http://m.neogaf.com/showthread.php?t=1342575&page=100000

People assume it is 1080p 60fps on that thread and no one try to fix it. So I assume it's true.

Have you actually read that thread? As you can tell I posted in it quite a bit. There's only PS4 footage shown, not XB1. And most people seem to agree that the PS4 footage is not locked at 60fps.


EDIT:
Snake Pass to me is an example of a game that shows off how well the Switch will punch above its weight. You have the PS4 version struggling to maintain 60fps, and the Switch version locked at 30fps with essentially the same visuals (I've heard different lighting but I haven't seen any appreciable changes) which, theoretically would put the Switch version at >50% of the PS4 version.

Whereas people going by pure flops keep reminding us that the Switch has less than 25% the flops of the PS4 going by the July devkit specs and Eurogamer clocks. So performing >50% should indicate that flops tell a very small part of the overall story.
 

Donnie

Member
Lots of talk about how switch can be stronger than we think, but for me, that game give us a good idea of what is possible:

Snake-Pass-1.jpg


Snake Pass is 1080p 30fps on switch and 1080p 60fps on Xone.

That game beauty, but it's slow, with not much thing on screen and a repetitive scenery.

If Switch can't make a game like that run in the same quality of the Xbox one, then, it is weaker to that machine, don't matter how much FP16 and Nvidia GFlops help it.

OK... does the XBox One version run at 1080p 60fps? Because I haven't seen that confirmed. Even if it does you're assuming that means it could look better on XBox One but not so on Switch. Well what if it's unstable 60fps on Xbox One and stable 30fps on Switch? What would that tell us? Also what on earth does any of that have to do with fp16?, this is a port and not a UE4 game afaik.
 
OK... does the XBox One version run at 1080p 60fps? Because I haven't seen that confirmed. Even if it does you're assuming that means it could look better on XBox One but not so on Switch. Well what if it's unstable 60fps on Xbox One and stable 30fps on Switch? What would that tell us? Also what on earth does any of that have to do with fp16?, this is a port and not a UE4 game afaik.

It is a UE4 game actually, and I don't know where he got that performance on XB1 from because it's 1080p and an unstable 60fps on PS4. There's no XB1 footage yet as far as I can tell.
 

HeelPower

Member
If Nintendo would treat this as a handheld first that you can play on a TV rather than a console that you can take on the go, their situation would be much better. Its an amazing handheld.

But they're going with "homeconsole first that you can take on the go" narrative,and that's why comparisons to competitors are warranted.
 

Donnie

Member
Sorry, but it is not what it looks like.
OK, there is people that assuming the clocks and other configurations that DF leak as the worst case, when switch is a ~400 GFlops machine with 3 under clocked A57 chips and 25Gb/s bandwidth.

But there is there other side too.
There are people that think the switch use a 16mm pascal custom Tegra. That gives numbers close to 600 GFlops. But they talking more how Nvidia GFlops is roughly 30% better than AMD (what let switch with a performance close to a 780 GFlops AMD in theory), plus the advantage of it utilize FP16, what can increase that number roughly by 50%, what let it at 900 GFlops with mixed precision, or equivalent to a 1170 Gflop GPU from AMD, that is, basically, a Xbox one.
They sum it with the possibility of it use A72 on CPU, what, even with 4 cores, can put switch in the same ballpark as Ps4 in that area.

So yes, there is people that are hoping to a "almost" Xbox one machine.

You're talking complete nonsense. Nobody thinks Switch is 600Gflops fp32, I really don't think you understand what you're reading here (no offence, just you clearly don't have a good understanding of the English language).

BTW if those A57 CPU's weren't "underclocked" they'd be twice the performance of PS4's CPU cores, imagine if those dogs were underclocked..

It is a UE4 game actually, and I don't know where he got that performance on XB1 from because it's 1080p and an unstable 60fps on PS4. There's no XB1 footage yet as far as I can tell.

Fair enough, but like you I've been unable to find any info on the XBox One version.
 

LordOfChaos

Member
Size? Really?

I mean, is there a weak argument than the size?

Well, yeah, that's actually a good argument, since we only have 3.5 years of fabrication plant advancement to account for and the Switch most likely isn't even using the most advanced ones. You can only cool so much chip in such a small enclosure feasibly, and can only have so much die size to improve performance without raising cooling costs within a budget.

Think of how much GPU you can manage in a 4 pound laptop vs a 12 pound desktop. Even if the desktop is from 2013 and the laptop is from now, it's not enough to overcome the thermal gap.

Especially as we now see a 120mm2 die, similar to a TX1, and most things are pointing to 20nm so far.
 

phanphare

Banned
If Nintendo would treat this as a handheld first that you can play on a TV rather than a console that you can take on the go, their situation would be much better. Its an amazing handheld.

But they're going with "homeconsole first that you can take on the go" narrative,and that's why comparisons to competitors are warranted.

it's both but it's clearly a handheld first

who cares about PR spin, just look at the actual hardware
 

PrimeBeef

Member
If Nintendo would treat this as a handheld first that you can play on a TV rather than a console that you can take on the go, their situation would be much better. Its an amazing handheld.

But they're going with "homeconsole first that you can take on the go" narrative,and that's why comparisons to competitors are warranted.

I thought Nintendo was actually pitching it as a hybrid, not cementing if it was either and NoA was saying a home console on the go.
 

HeelPower

Member
it's both but it's clearly a handheld first

who cares about PR spin, just look at the actual hardware

Because it matters if Nintendo treats it as the 3DS successor.That would mean its gonna get an amazing stream of games.

As a home console,it has less chances of getting third parties due to power gap.
 

Donnie

Member
http://m.neogaf.com/showthread.php?t=1342575&page=100000[/url]

People assume it is 1080p 60fps on that thread and no one try to fix it. So I assume it's true.

So your entire argument is based on nobody correcting an assumption?, amazing. Are you even serious here?
 

phanphare

Banned
Because it matters if Nintendo treats it as the 3DS successor.That would mean its gonna get an amazing stream of games.

As a home console,it has less chances of getting third parties due to power gap.

well like I said it's both. it's also already getting a bunch of games that would be right at home on 3DS. focus less on the PR and more on what's actually happening. let the pudding be the proof, as they say.
 

mario_O

Member
If Nintendo would treat this as a handheld first that you can play on a TV rather than a console that you can take on the go, their situation would be much better. Its an amazing handheld.

But they're going with "homeconsole first that you can take on the go" narrative,and that's why comparisons to competitors are warranted.

All Nintendo's ads show both modes, docked and handheld, even the tabletop mode. And people in the West prefer home consoles, they're not going to downplay it.
 
Top Bottom