The batteries seems to be once again very easy to replace which is a good thing.
I guess a teardown of the retail switch will follow soon.
-
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
- Thread starter Dystify
- Start date
The batteries seems to be once again very easy to replace which is a good thing.
I guess a teardown of the retail switch will follow soon.
Yeah results seem really solid for a device of that size, but still not good enough for a proper console. Anyway, i doubt the iPhone outperforms the Switch GPU wise, the CPU is obviously much better.
By the way, foxcoon leak said that the joycon batteries were 525mAh, right?
LordOfChaos
Member
Yup, crazy considering that the iPhone is also one of the smaller phones. Apple silicon is no joke.
I really wish what we could see what that puppy could do with active cooling like the Switch... Apple TV is a missed opportunity for a decent microconsole.
Sapiens
Member
UnemployedVillain
Member
The Tegra X1 outperformed the A8 in a number of tests and I don't think the A9 is "much better" than it either. People forget the X1 is still among the most powerful mobile SOCs out there, especially considering its price. I could be mistaken though
Schnozberry
Member
If Apple thought they could be wildly profitable making game consoles, they certainly could, but I think the margins are too low for them.
In terms of performance, I think you're seriously underselling how much of a difference there is between making games for iOS/Android on general purpose devices vs. a custom kit dedicated to gaming with lots of Nvidia, Nintendo, and middleware support.
LordOfChaos
Member
Sure, but I was only directly answering the posed question
A dedicated gaming platform has plenty of advantages, though I think API wise with low level APIs like Metal or NVN we'd be hitting a point of diminishing returns between them.
The single biggest thing is...You don't put as much of a budget into making a game you sell for 5 dollars, or dollars in microtransactions, as you do a 60 dollar AAA title. Here's the Switches biggest selling point. The A10 Fusion SoC could certainly run a hell of a lot, but it only gets whatever Nintendo decides to release for it, and no console class AAA blockbusters.
Eolz
Member
Don't know if this has been mentioned yet but Nintendo's Japanese support site says the Switch supports UHS-I Class 1 cards (just like those licensed HORI cards).
"Q: What microSD cards could be used with the Switch?
A: Nintendo Switch can support microSD/microSDHC/microSDXC memory cards.
It also supports the UHS-I standard as well as UHS Speed Class 1."
--------------
On other news, the SD Association has just introduced the UHS-III standard which still utilizes the same layout as UHS-II cards.
So UHS-1 V10 would be the max supported, and not UHS-1 V30?
blu
Wants the largest console games publisher to avoid Nintendo's platforms.
You need to clarify what you mean by v10 and v30. UHS-I is SD v3.01, UHS-II is SD v4.0 and UHS-III is SD v6.0.
https://en.wikipedia.org/wiki/Secure_Digital#Ultra_High_Speed_.28UHS.29_bus
Pokemaniac
Member
Eolz
Member
Ah yeah, sorry, I was referring to this picture:
So U1/U3, which apparently is the same speed as V10/V30.
It's weird since some cards have different symbols yet the same speed (like the circled 10 being apparently the same as U1).
blu
Wants the largest console games publisher to avoid Nintendo's platforms.
The SD consortium is an example of how _not_ to make speed rating systems for the consumer. That's why most vendors adhere to providing the UHS class (I, II, III) which tells the bus capability of the card, and the top read speed (e.g. 45MB/s).Ah yeah, sorry, I was referring to this picture:
So U1/U3, which apparently is the same speed as V10/V30.
It's weird since some cards have different symbols yet the same speed (like the circled 10 being apparently the same as U1).
Regarding Apple A10 and while I think they have done an outstanding job with their custom cores, remember that it's a 2+2 cluster switching setup, so while their IPC is top tier it would struggle even more than X1 with games that use multi thread and that's without thinking to fit the OS.
Of course if Apple really wanted they could design another core setup, but I don't think A10 would be a better SoC than X1 for what the Switch needs.
Also even pursuing another A10 with more cores would have some extra hurdles seeing as their core area is huge by ARM standards (4.18mm Hurricane vs 1.7mm~ A57) and A10 is already pushing 125mm2 die size.
And that's without talking about the gpu side of things where X1 should probably still be ahead.
Of course if Apple really wanted they could design another core setup, but I don't think A10 would be a better SoC than X1 for what the Switch needs.
Also even pursuing another A10 with more cores would have some extra hurdles seeing as their core area is huge by ARM standards (4.18mm Hurricane vs 1.7mm~ A57) and A10 is already pushing 125mm2 die size.
And that's without talking about the gpu side of things where X1 should probably still be ahead.
TanookiTom
Member
Okay I am not a techie but I've been trying to follow this post.
Can someone here explain to me why the gaming media (such as Gamespot etc.) seem to report on roughly 1 TF for the Switch when much lower numbers are being used in here?
I am trying to calm some people down who expect Xbone-like power levels but all the sources I can find online are pointing at those 1 TF...
Can someone here explain to me why the gaming media (such as Gamespot etc.) seem to report on roughly 1 TF for the Switch when much lower numbers are being used in here?
I am trying to calm some people down who expect Xbone-like power levels but all the sources I can find online are pointing at those 1 TF...
Maybe link to the sources?
Is that reddish brown stuff the thermal paste? What the fuck.
There was a report from some website that said it was 1 TFLOPS, but that was in half-precision numbers, whereas it is normal to give the FLOPS in full precision numbers, which is 512 GFLOPS. Beyond that, the report didn't take into account that this system has a lower clock speed than the stock Tegra X1, which decreases the FLOPS count to 393 GFLOPS when docked and 197 GFLOPS when undocked. The report that all those news sites are running was plain wrong, and showing an annoying lack of basic understanding of what they were reporting.
Thanks for posting this, it's nice to have a higher resolution image. Comparing it to the images in the OP, it seems the prior photos were of a unit with a non-final battery, but everything else seems to be the same (from what I can see).
There's a metal panel between the Switch's innards that you see there and the rear of the case. The thermal paste is there to provide a thermal connection to this panel to allow the system to passively dissipate heat more effectively when the fan is off.
Skittzo0413
Member
The bottom right portion of the unit in this image
looks substantially different than in the above one. Although this other image from the OP:
looks the same. It seems like a component was removed from that area in the first image, any ideas what it was?
Graphics Horse
Member
Game card slot, no?
Ah, that explains it. Took me by surprise.
Skittzo0413
Member
Ah, yeah. That appears to be right. So I guess they removed the game card board for whatever reason from that first pic I quoted? Weird.
Or the silver portion in that picture is a cover?
Thanks.
Finally.
That white block definitely looks like the HD Rumble.
It's just too bad that I still can't prove yet where it's from, especially Alps Electric.
A reminder that this is what the Haptic Reactor from Alps Electric looks like:
These are its Dimensions (W×D×H) 9×10×22.6mm
http://www.alps.com/prod/info/E/HTML/Actuator/index.html
ShadowFox08
Member
Most likely its just fp16 numbers for X1(while fp32 is 500GFLOPs) nvidia specs and at full clock speeds. According to eurogamer leaked specs from dec(and they seem pretty sure about it even now) the final dev clockspeeds were downclocked. GPU by like 25% and CPU clockspeeds are cut in half, due to conserving battery life and throttling from heat.
Devs can't 100% utilize fp16 in games anyway, as fp32 is required in some areas. Ubisoft devs claim they can use as much as 70% fp16 though.
Edit:whelp someone already said half precision(meaning fp16).
As Graphics Horse says, there's a daughter board for the game card slot (and headphone jack) which was removed in the first photo (presumably to get photos of it, as they posted a separate photo of the daughter board on its own).
Here's something interesting I noticed on the new photo, though. There's a small low-profile connector on the motherboard which isn't connected to anything:
This is also unconnected in all the other photos we've seen, and none of the components we've seen would interface with it. It's also in a position where there really wouldn't be much space for anything to sit if it were connected.
One theory that did occur to me is that it's an alternative for the eMMC connector. That is, it's designed for a small board which sits above the wifi/bluetooth module, just as the eMMC board does, but which needs a different interface. This could be embedded UFS (or even in theory BGA NVMe), and it would help explain why the eMMC module is on its own board in the first place.
Edit:
So here's something interesting: the left and right joy-cons have different antenna configurations:
Following recent reports of the left joy-con not having quite as good wireless connectivity as the right one, it's interesting to see that they do have physically different antenna configurations. On the image of the right joy-con we can see the antenna connection on the PCB, and a wire running to a metal plate, which it's soldered to. This metal plate is likely acting as the antenna (and may wrap around out of view of this photo).
On the left joy-con, though, we don't see the antenna connector, the wire or the metal plate. The joy-con definitely has an antenna (it physically has to to transmit wireless signals), but it's not visible in this photo, and it's different to the antenna Nintendo is using in the right joy-con. That's not necessarily to say that it's worse, but it's not the same.
It's interesting to consider why Nintendo didn't just use the same antenna configuration between both units. It's possible that the extra data transferred by the right joy-con while using the infra-red camera on the bottom required a higher quality connection with Switch, so they upgraded it to a beefier antenna while leaving the left joy-con one as-is.
Finally, something that explains why the left Joy-Con was different to the right Joy-Con with regards to signal strength.
So this practically confirms that the images in the OP are from a similar unit to the retail ones. right?
Well, the Left and Right Joy-Cons are different. You can read amiibo on the right Cons but not the Left Cons. But I'm not sure how that'd change anything. The right also has that IR pointer on the bottom on it.
Digital Foundry
How does these power draw numbers line up with expectations/speculation about the Switch's hardware?
I've never seen anyone in these threads suggest wattage that high, even for docked mode on Maxwell architecture. With Eurogamer clocks, and their assumption that very little changes were made to the Tegra chip, shouldn't we be seeing much lower numbers? Does this hint at potentially higher clocks on retail devices, some other changes/additions to the SoC, or heck, maybe even some poor design choices here and there in the system? What say you guys?
http://www.eurogamer.net/articles/digitalfoundry-2017-nintendo-switch-review
How does these power draw numbers line up with expectations/speculation about the Switch's hardware?
I've never seen anyone in these threads suggest wattage that high, even for docked mode on Maxwell architecture. With Eurogamer clocks, and their assumption that very little changes were made to the Tegra chip, shouldn't we be seeing much lower numbers? Does this hint at potentially higher clocks on retail devices, some other changes/additions to the SoC, or heck, maybe even some poor design choices here and there in the system? What say you guys?
Skittzo0413
Member
I'm wondering this too, and I've asked about it in a few threads. The Shield TV for reference maxes out at 20W, likely at max load, so 2GHz CPU and 1GHz GPU. The Switch uses a bit more RAM but does not have a hard drive to power.
Basically, 16W would suggest getting a lot closer to the Shield TV's peak clocks at 20W, but DF is telling us that the clocks max out at 1GHz for the CPU (50% of Shield TV) and 768MHz (75% of Shield TV). Could those differences really only amount to a 4W difference?
Again, a thing to note is RAM- the Switch has more and it might use the RAM at a higher frequency or bandwidth. I don't know if we have any numbers for RAM in the Shield TV.
TLZ
Banned
Confirmation the joycon issue is hardware related and can't be fixed with a fw update then.
Skittzo0413
Member
The thing on the bottom of the right joy-con is an IR camera, not a pointer. Depending on the resolution it could take a reasonably amount of bandwidth to stream to the Switch (certainly a lot more than normal button input), so it would make some degree of sense that it has different requirements than the left joy-con when it comes to the wireless interface.
Not necessarily. Range reportedly improves with both joy-cons when connected to the charging grip, which suggests they can operate with increased signal power if required. They could potentially adjust the left joy-con's signal strength when disconnected in order to improve the connection at the cost of a bit of battery life. If they feel they need to, that is.
-shadow-
Member
Such a strange choice to have one with and one, seemingly, without an extra antenna. Possibly that these two are still two different batches for some reason (as in a newer left joy-con potentially also having an antenna)? Anyone else happened to have opened the joy-con?
I don't think the power draw when docked is going to tell you too much, you have a charging battery that takes up power draw, I thought we had a 12w figure for when the Switch was docked and not charging the battery?
I'm not sure how the foxconn clocks exist, so I'll just wait for more information to come in. Seems Eurogamer is using works like "likely" to detail the switch and their chart they keep posting with the clock speeds is a eurogamer chart and not from the updated devkit so if a developer said "yeah the undocked clock is 384mhz now" that doesn't mean they would know the other clocks, a 921mhz gpu clock seems needed if they raised the undocked clock but that is just my speculation and its always been on a leak proven true.
The tear down here certainly didn't come from a final production unit as the battery is different from the one on the last page, so like some of us assumed, it is a prototype/earlier production/devkit unit, meaning it can't confirm final hardware, but that is just over a day away.
We are close to real answers, at least we aren't getting shell shocked from our speculation as Eurogamer's clocks were definitely real and gave us the minimum spec that has already been improved on with the performance boost Eurogamer detailed. I'm glad we will be able to stop speculating on hardware soon, this is by far my favorite hardware design from Nintendo and they did a great job with performance. Battery life does seem to be a possible problem, so it will be confusing if Switch is indeed 20nm because they could have avoided this problem with a shift to 16nm. I get my switch in about 40 hours (when I'll actually pick it up) I won't be opening it up, but I'll help dig into whatever information comes in here, as long as I'm not busy playing Zelda that is.
I'm not sure how the foxconn clocks exist, so I'll just wait for more information to come in. Seems Eurogamer is using works like "likely" to detail the switch and their chart they keep posting with the clock speeds is a eurogamer chart and not from the updated devkit so if a developer said "yeah the undocked clock is 384mhz now" that doesn't mean they would know the other clocks, a 921mhz gpu clock seems needed if they raised the undocked clock but that is just my speculation and its always been on a leak proven true.
The tear down here certainly didn't come from a final production unit as the battery is different from the one on the last page, so like some of us assumed, it is a prototype/earlier production/devkit unit, meaning it can't confirm final hardware, but that is just over a day away.
We are close to real answers, at least we aren't getting shell shocked from our speculation as Eurogamer's clocks were definitely real and gave us the minimum spec that has already been improved on with the performance boost Eurogamer detailed. I'm glad we will be able to stop speculating on hardware soon, this is by far my favorite hardware design from Nintendo and they did a great job with performance. Battery life does seem to be a possible problem, so it will be confusing if Switch is indeed 20nm because they could have avoided this problem with a shift to 16nm. I get my switch in about 40 hours (when I'll actually pick it up) I won't be opening it up, but I'll help dig into whatever information comes in here, as long as I'm not busy playing Zelda that is.
Skittzo0413
Member
Based on what people are saying I think we'll all be too busy playing Zelda...
Indeed... my theory is that the left joycon actually communicates with the right through a different protocol, not Bluetooth. I guess we could test that by removing the right joycon from range and see if the left still connects... the left needs much less bandwidth since it's just standard inputs. Also the configurations of joycon supported would exceed the max 7 (plus host) the Bluetooth has.
TanookiTom
Member
Thanks a lot guys! I'll admit that I started a commenting war with a user on Amazon.de that just loves to spread false information, such as the power of the Switch being almost on par with Xbox One. Sadly a lot of the German gaming media seems to have referenced that VentureBeat/Gamespot report from December, reporting it basically as facts.
They each have an antenna (as I mentioned any device capable of wireless communication has to), the antenna on the left on is just different than the right one. It's hard to read too much more into it without seeing the left joycon's antenna, though, which we can't from this angle.
That's a really interesting theory, I hadn't thought of that. If you wanted to be absolutely sure you could detach the battery in the right joycon before testing (it looks easily detachable from the photo).
Marcan has already one and it seems that the OS is based on FreeBSD.
https://twitter.com/marcan42/status/837481011508195329
https://twitter.com/marcan42/status/837481011508195329
Lol I actually decided not to answer that idiot after I saw someone already started discussing with him!
Great, this should end the clocks talk pretty quickly. Chipworks will do the rest with the node.
If there is eurogamer clocks we know its not 16nm. So I guess what will be left is trying to find out exactly why this is a custom soc.
I don't know how quickly, he's playing Zelda for now.
There is no guarantee chipworks will be kind enough this time.
Dont really have to do that, as I hid my right away and left behind my back 5 meters was operational.
Now here's the funny part, my right controller has connection issues at same distance and position.
For me the left has better connection.