-
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
TannerDemoz
Member
Yeah, I agree but I'm still seeing a lot of mixed reception from people. I think 2.5 hours as the lowest standard just comes across as a shite number, but when you're comparing that to the power of the hardware, surely it's really impressive?
I generally struggle to accept criticism on its battery life as it's so easy to be near a power socket anyway?
I thought the new NVIDIA Shield TV is exactly the same as the old one but smaller.
There is a revision of the Tegra X1 in the new one.
2.5 is the minimum and Zelda is 3.5 hours long.
It depends on the game.
I'm hoping something like Shovel Knight would run for 6 hours.
TwilightPrincess
Member
Not that I don't believe you but do you have any links about this.
As when I read articles on the new Shield TV majority of them stated it's the same as the old one.
Someone mentioned it earlier but apparently it benchmarks higher for some reason.
Its the same as in people had expected maybe a parker chip. I don't think anyone has actually gone as far as scanning the chips etc. for changes. The New shield does seem to perform better in benchmarks but there could be OS changes or other things that have led to that.
We need chipworks before we can definitively say anything about what changes are there including 16nm versus 20nm etc
The revisions could be so tiny that Nvidia didn't make a big deal out of it. It's linked somewhere earlier in this thread that the new one does very slightly better in benchmarks.
Just like ms said Xbox one s was the same, while in actuality it's a little better
Spyder_Monkey
Banned
The motherboard is so tiny it's adorable.
Unknown Soldier
Member
You should see the motherboards on phones.
Better benchmarks, and different chips:
Original Shield TV:
New Shield TV:
Edit: rehosted the first pic.
Kadin
Member
I'm trying to recall, what's Nintendo's track record with estimated battery usage versus real world?
othersteve
Member
(Regarding active cooling):
Wow, I seriously didn't know this. And I'm happy about it too, because it greatly reduces the odds of thermal throttling, which I was quite concerned about previously!
When Nvidia were talking about how many man hours they put into the custom chip that powers the Switch they worded it perfectly because what they really meant was they put all those man hours into the chip that powers both their own refresh of the Shield and also the Switch.
I think it's worth looking at what today's info tells us about the game cards and their interface. First we have the photo of the Switch game card from earlier today:
Here we can see 16 physical contacts, which is a lot more than the 5 I had assumed from earlier (blurrier) photos. DS/3DS cards, by comparison, have 17 pins, although they use a parallel interface and have to accommodate writes as well as reads (whereas Switch cards are read-only).
Of those pins, we can reasonably assume the following:
Data - ?
Ground - 1 or more pins
Vcc - 1
There's also the possibility of a clock pin (although this can be embedded in the data signal using an 8b10b encoding), or command or control pins, or removal detection pins. In theory the data bus may be as wide as 12-bits, but my guess would be we're looking an an 8-bit data bus, as these things are almost always power-of-two.
Then we have the front and back of the game card reader:
On the back we can see a wide connection to the mainboard, with a very large number of traces going to a small chip, which seems to have a small number of traces exiting it (to the game card, presumably). This chip is presumably a SERDES, taking the narrow, high speed serial data coming from the game card and de-serialising it, to produce a wide, lower-clocked parallel signal which then goes out to the mainboard and on to the SoC.
One thing that interests me is just how wide the interface is between the SERDES and the mainboard. That's indicative of a very wide parallel interface on both ends, which means we could be looking at very high read speeds from the game cards. Quite a while back I speculated on the potential speeds that Nintendo could get from a serial game card interface:
Given how dense the traces are between the SERDES and the motherboard, it looks like the game cards may well be designed for the 32-bit reads I speculated about above, putting performance well ahead of the HDDs used in PS4/XBO. It's possible that this varies on a game-by-game basis (bigger games would use faster cards), but what we see would indicate a pretty speedy upper limit.
Here we can see 16 physical contacts, which is a lot more than the 5 I had assumed from earlier (blurrier) photos. DS/3DS cards, by comparison, have 17 pins, although they use a parallel interface and have to accommodate writes as well as reads (whereas Switch cards are read-only).
Of those pins, we can reasonably assume the following:
Data - ?
Ground - 1 or more pins
Vcc - 1
There's also the possibility of a clock pin (although this can be embedded in the data signal using an 8b10b encoding), or command or control pins, or removal detection pins. In theory the data bus may be as wide as 12-bits, but my guess would be we're looking an an 8-bit data bus, as these things are almost always power-of-two.
Then we have the front and back of the game card reader:
On the back we can see a wide connection to the mainboard, with a very large number of traces going to a small chip, which seems to have a small number of traces exiting it (to the game card, presumably). This chip is presumably a SERDES, taking the narrow, high speed serial data coming from the game card and de-serialising it, to produce a wide, lower-clocked parallel signal which then goes out to the mainboard and on to the SoC.
One thing that interests me is just how wide the interface is between the SERDES and the mainboard. That's indicative of a very wide parallel interface on both ends, which means we could be looking at very high read speeds from the game cards. Quite a while back I speculated on the potential speeds that Nintendo could get from a serial game card interface:
Thraktor said:
Given how dense the traces are between the SERDES and the motherboard, it looks like the game cards may well be designed for the 32-bit reads I speculated about above, putting performance well ahead of the HDDs used in PS4/XBO. It's possible that this varies on a game-by-game basis (bigger games would use faster cards), but what we see would indicate a pretty speedy upper limit.
That's actually really impressive. Almost no space wasted while keeping parts like the card slots and batteries easy to replace. It looks like it doesn't use any glue either. Overall this is a well engineered piece of kit. I'm looking forward to what the ifixit guys think about it.
It might not be 100% exactly the same (I mean, the print is different
), but from everything indicated here the differences certainly don't seem all that large.Unknown Soldier
Member
I think it would be a good portable, as a home console it's a disaster and I wish Nintendo would just give up on this hybrid concept
weeman_com
Neo Member
What would you suggest to be the best option for a sdxc card to buy for switch. I seen a 64gb with 90mbs read/write speed on offer that I was thinking of buying to start me off. I was thinking the higher a read speed the better but would 90mbs be counter intuitive?
https://m.mymemory.co.uk/product?p=46889
Thanks for posting this. From the codes on the dies we have:
Shield TV v1: TM670D-A1
Shield TV v2: TM670M-A2
Switch: ODNX02-A2
Buggy Loop
Member
What? No it's far from a disaster. Graphics have plateaued pretty much thanks to silicon limits. Never in gaming history mobile have come this close to comparison with console hardware.
With :
Nvidia flops vs AMD flops
Vulkan or NVN API
FP16
This thing will be really really neat.
I would not be surprised if next gen hybrids become the norm, mobile chips are becoming incredibly good.
I also don't like a world where this
is a disaster. Gameboy player was freaking nice all of nintendo's portable/console on 1 platform, thank god for hybrid.
JonnyDBrit
Member
Man, nice to have some perspective on that. Am kinda curious: How would those possibilities compare to read speeds on an SD card? Since Switch games evidentty won't need to install to launch, would digital and physical copies potentially have a difference in loading times?
Astral Dog
Member
I like that they are the only ones taking handheld gaming seriously by releasing this little Beast,much better than another Wii or Wii U
A bit surprising that Gamespot equates an X1 to X360/PS3. Digital Foundry says it's somewhere between last gen and this gen.
https://youtu.be/C6cTt_nxTWo?t=125
LordOfChaos
Member
Where do you see better benchmarks for the new Shield TV vs old?
I've ordered one of these, which seems to be the best value in terms of capacity vs price. Anything with a ~90MB/s read speed from a reputable manufacturer should be fine, write speeds aren't going to be as important as it'll be doing a lot more reading than writing (and writes will usually be bottlenecked by your broadband connection, anyway).
Faster than regular MicroSDs (~250MB/s vs ~90MB/s), although in theory UHS-II MicroSD cards would match it. I very much doubt Switch supports UHS-II, though, as it's only made its way to high-end cameras at the moment.
JonnyDBrit
Member
So physical versions could potentially have games read from storage around 2.5x faster? I presume it's not exactly a consistent or necessarily maximised process.
Elfotografoalocado
Banned
But, the original Shield TV didn't throttle in benchmarks, as in, the GPU ran at 1GHZ and the CPU at 2GHZ, while in games there was power throttle to the speeds those july devkits were running at.
Are the load speeds of this new Shield TV known? What if it runs at higher speeds because it's built in 16nm? Wouldn't make sense to use that botched node when you have plenty of 16nm wafers already in production.
Die shot suggests strongly that its a K1, not an X1 or newer, with modifications for Nintendo
Some K1's carry the A2 code too.
Bwahahaahaha
No.
I think it's worth looking at what today's info tells us about the game cards and their interface. First we have the photo of the Switch game card from earlier today:
Here we can see 16 physical contacts, which is a lot more than the 5 I had assumed from earlier (blurrier) photos. DS/3DS cards, by comparison, have 17 pins, although they use a parallel interface and have to accommodate writes as well as reads (whereas Switch cards are read-only).
Of those pins, we can reasonably assume the following:
Data - ?
Ground - 1 or more pins
Vcc - 1
There's also the possibility of a clock pin (although this can be embedded in the data signal using an 8b10b encoding), or command or control pins, or removal detection pins. In theory the data bus may be as wide as 12-bits, but my guess would be we're looking an an 8-bit data bus, as these things are almost always power-of-two.
Then we have the front and back of the game card reader:
On the back we can see a wide connection to the mainboard, with a very large number of traces going to a small chip, which seems to have a small number of traces exiting it (to the game card, presumably). This chip is presumably a SERDES, taking the narrow, high speed serial data coming from the game card and de-serialising it, to produce a wide, lower-clocked parallel signal which then goes out to the mainboard and on to the SoC.
One thing that interests me is just how wide the interface is between the SERDES and the mainboard. That's indicative of a very wide parallel interface on both ends, which means we could be looking at very high read speeds from the game cards. Quite a while back I speculated on the potential speeds that Nintendo could get from a serial game card interface:
Given how dense the traces are between the SERDES and the motherboard, it looks like the game cards may well be designed for the 32-bit reads I speculated about above, putting performance well ahead of the HDDs used in PS4/XBO. It's possible that this varies on a game-by-game basis (bigger games would use faster cards), but what we see would indicate a pretty speedy upper limit.
If it adds anything, Breath of the Wild on Wii U will require a 3GB install for faster loading times, etc., while the Switch version will not require that at all. They're likely using a very fast card for Breath of the Wild to cut down the load times and pop-in.
I really like how these cards are moving past CDs and making things quiet.
Rushersauce
Banned
Die shot suggests strongly that its a K1, not an X1 or newer, with modifications for Nintendo
Some K1's carry the A2 code too.
eh... no
LordOfChaos
Member
Die shot suggests strongly that its a K1, not an X1 or newer, with modifications for Nintendo
Some K1's carry the A2 code too.
Again, "Ax" markings are mask revisions. The first produced version of a chip is A0. If there are defects, A1, and so forth. The actual logic (and so capabilities) are not changed.
All A2 indicates then is A0 was taped out, found a flaw, A1 was taped out, etc.
I think most of that time went into creating the nvn API, the tools and the development environment which by all accounts is exceptional.
UnemployedVillain
Member
It might not be 100% exactly the same (I mean, the print is different
What's indicated here? We literally just have a picture of the silicon and nothing else, and a bunch of people who are simply trying to match pictures of the back to indicate it's 1:1. It's ridiculous
LordOfChaos
Member
250 people working 2 years isn't much for chipmaking. That also includes the NVN API, the OS integration, some sort of physics library they mentioned, etc.
Skittzo0413
Member
It might not be 100% exactly the same (I mean, the print is different
I mean, don't we know already that the 4 A53 cores have been removed? Isn't that a fairly substantial difference? Is there anything which would make sense to replace those with that could line up with the July specs?
weeman_com
Neo Member
Yeah that's what I was thinking. I was looking at the card you linked but it was not as cheap as that when I was... So I've just ordered it hahaha. I just wanted to have. Effective read speed for the card as to not limit the system from loading the data.
The price of the cards are extortionate... Over 128gb with moderate read speeds and your paying around £100 if not more.
LordOfChaos
Member
It's a miniscule difference. A Cortex A53 is 0.7mm2 on 20nm. It may have made more sense to leave them in and disable them, who knows until die scans.
When you see a Chinese SoC vendor advertising 17 cores, this is why, making a big high IPC core is expensive, sprinkling in A53s is almost free.
Wii U optical disc read speeds are only around 25MB/s (afaik), so even on the lowest possible end of the spectrum Switch game cards would be massively faster (even ignoring the improvement in random access).
The talk of card speeds has reminded me of this tweet, though, where an analyst claimed that Switch has an "internal bus speed at 5Gbps, faster than 3DS's 128Mbps". He's only an analyst, and it's likely that he was actually talking about something else (e.g. the USB-C data connection), but 3DS's card interface is 128Mb/s, and it's occurred to me that if Switch's game cards do operate at 62.5MHz, as I assumed, then a 64-bit interface from the ROM would actually require exactly a 5Gb/s serial interface if 8b/10b encoding was used (i.e. 4Gb/s data + 1Gb/s protocol overhead).
As I say, he's only an analyst, and I don't see Nintendo needing 500MB/s read speeds from the game cards (if nothing else it would make the MicroSD and eMMC seem pretty sluggish in comparison). Nonetheless, it's an interesting thing to consider.
frankie_baby
Member
Do we actually know the A53 cores were removed? They weren't in the dev docs but maybe that's because they aren't accessible to devs?
Skittzo0413
Member
Didn't the dev docs explicitly say what was on the hardware and then separately what was accessible to devs? Which would mean no A53s on the die. That at least confirms there would be some physical customization to the die itself, though apparently very minor.
frankie_baby
Member
200gb cards are about 70 quid
Lol, A2 is just a revision number, it actually starts at 0.Die shot suggests strongly that its a K1, not an X1 or newer, with modifications for Nintendo
Some K1's carry the A2 code too.
<edit>
Actually after seeing that 2017 shield TV PCB, its definitely an X1, which still destroys any suggestion/rumor of it being a newer chip
LordOfChaos
Member
He says Displayport over USB C, he's undoubtedly talking about the USB-C dock data connection it would seem?
It was based on this comment earlier:
But searching now I see different results over the time for the original Shield, so it might be also due to newer Android versions.