Castef
Banned
Yet, the 4GB gap in RAM is really, really wide.
Nintendo Switch Dev Kit Stats Leaked? Cortex A57, 4GB RAM, 32GB Storage, Multi-Touch.
- Thread starter Robert at Zeboyd Games
- Start date
- Status
Yet, the 4GB gap in RAM is really, really wide.
calc84maniac
Member
If rumors are correct, as far as games are concerned there will be a 2GB difference. Nintendo won't be going with a heavyweight OS.
3.2GB (720p-1080p) vs 4.5/5/5.5GB (900p-4k).
3.2 vs 5-5.5GB isn't a 4GB gap. The amount of RAM pissed me off (i wanted/expected 6GB total), but let's not rewrite history.
th3sicknness
Member
If they are using Maxwell instead of Pascal with little to no modification then wtf does this quote mean from Nvidia?
"The development encompassed 500 man-years of effort across every facet of creating a new gaming platform: algorithms, computer architecture, system design, system software, APIs, game engines and peripherals. "
"The development encompassed 500 man-years of effort across every facet of creating a new gaming platform: algorithms, computer architecture, system design, system software, APIs, game engines and peripherals. "
Schnozberry
Member
It's very possible the chip design was finalized in 2015. Nvidia had no way to predict how their 16nm products would scale at that time in full production, and they did in fact have supply issues with GeForce Products after launch due to fab constraints.
You can't rule out 16nm, but there's no evidence to support it other than speculation, either. I think people should prepare for this to be pretty much a Tegra X1 in terms of performance and design, with perhaps some light customization to facilitate the dock/TV output and perhaps some cache modifications to overcome the anemic nature of the primary memory interface. That's not necessarily bad. Most of the issues with the X1 were related to Android and it's horrible GL performance. Maybe we'll get lucky and the Switch will have more GPU cores than a standard X1.
The final dev kits being Maxwell is a dead giveaway.
Saying that its a heavily modified Maxwell doesn't tell us the node its on, it just IS a modified Maxwell because that's the chip is started as. They aren't going to customise it exactly into a Pascal chip, that wouldn't be custom and they didn't start it as Pascal so its not custom Pascal. Whatever the chip has ended up as (more or less power efficient, powerful ect than Pascal) its a custom Maxwell, that doesn't tell us anything though unfortunately.
Schnozberry
Member
All true, but Nate did say not much has changed for the final kits, so..... ¯\_(ツ
_/¯We don't have any evidence other than every other Maxwell Tegra product being on 20nm. Given that Nintendo was supposedly given a killer deal on chips, perhaps they bought up the remaining 20nm fab commitments from Nvidia? That would explain the 20nm Tegra X1 Shield Tablet disappearing from the roadmap, and the Shield TV channels drying up in anticipation of the 16nm refresh.
All true, but Nate did say not much has changed for the final kits, so..... ¯\_(ツ
We don't have any evidence other than every other Maxwell Tegra product being on 20nm. Given that Nintendo was supposedly given a killer deal on chips, perhaps they bought up the remaining 20nm fab commitments from Nvidia? That would explain the 20nm Tegra X1 Shield Tablet disappearing from the roadmap, and the Shield TV channels drying up in anticipation of the 16nm refresh.
Nate also said its Pascal based though didn't he? Unfortunately none of this is revealing. What I'd like to hear from these leakers is plain and simple, what node is it on. Throwing around chip names is irreverent.
No offence meant to Nate BTW, just saying that Maxwell/Pascal really isn't relevant unless you specify that you mean the node is the same.
I mean this isn't like it was with WiiU when we were discussing wether it was R700 based or something current, because in that case the architecture was so different, what kind of shaders and features it would have would be so different depending on what chip it was more similar to. But in this case Maxwell and Pascal both use the same shaders, same number of them, same features. Only really significant difference is the node and what it could mean for performance per watt.
Schnozberry
Member
Nate was the one who had a source that tipped him off to the final devkits being out there. They are Maxwell and not a lot was changed from the previous kits.
I think you're grasping a bit to hold onto the 16nm dream. I think it's best to let it go, and be pleasantly surprised if we hear good news.
Skittzo0413
Member
The other thing suggesting 16nm is the rumored 5-8 hours for the battery life target. Even heavily downclocked that would be pretty difficult for a 20nm device.
Then again that comes from Nate who also said Pascal, so I guess if we are discounting that rumor then we might not be able to trust the 5-8 hour rumor either.
There is only one source talking about 800mb OS unless I'm mistaken, it'd be dangerous to accept it as a fact
How does my post have any suggestion of grasping to a dream of 16nm?.. What I'm trying to explain is that these leaks don't tell us anything and people are just wasting their time with the current fuss and obsession with "Its Maxwell" or "Its Pascal". Because by the very nature of Maxwell/Pascal (similarity of the chips and the fact Switch GPU is custom and started development long before Pascal ever came along) trying to decide what the chip will end up as based on the name Maxwell or Pascal is pointless.
If any of these leakers would just make a claim on the node being used, or number of cores, or clock speed it would be worth talking about.
So, in the Switch patents there are several things confirmed and relevant for this thread:
There is a fan in the tablet and it runs (at reduced RPM) also in mobile mode:
Portable mode downclock confirmed.
And different resolutions.
There is a fan in the tablet and it runs (at reduced RPM) also in mobile mode:
[0419] The main unit 2 includes a cooling fan 96 for radiating heat from inside the main unit 2. With the cooling fan 96 in operation, the air outside the housing 11 is introduced into the housing 11 through the air inlet hole 11d and the air inside the housing 11 is discharged through the air outlet hole 11c, thereby discharging heat from inside the housing 11. The cooling fan 96 is connected to the CPU 81, and the operation of the cooling fan 96 is controlled by the CPU 81. The main unit 2 includes a temperature sensor 95 for detecting the temperature inside the main unit 2. The temperature sensor 95 is connected to the CPU 81, and the detection results of the temperature sensor 95 are output to the CPU 81. The CPU 81 controls the operation of the cooling fan 96 based on the detection results of the temperature sensor 95.
Portable mode downclock confirmed.
And different resolutions.
Skittzo0413
Member
I don't see it explicitly mentioned, no.
Schnozberry
Member
Every Tegra product based on Maxwell is 20nm. That should be the assumption unless we get decent information that proves otherwise. I agree the obsession with Maxwell or Pascal is off base in some ways because of how similar they are, but it does give us a rough idea of the ballpark we're in when it comes to power consumption and performance expectations. Thraktor's post earlier in this thread outlined how similar performance could be achieved between the two.
The leakers will have no idea what node it's on, because they are getting information second hand from developers or studio insiders who aren't going to be able to pull lithography information from the tools at their disposal.
Not as far as I've seen. Just one fan mentioned, in the main unit.
And this is the main issue right now. Unless the die size is documented, we aren't going to be able to get specifics outside of the architecture the Switch is based upon. Like I said, final dev kits are Maxwell based. Began distribution in November.
The Pascal information I received months ago is from different contacts than the 5-8hr target battery range - remember, this is a range that was being targeted and never a sure thing.
So new news then?
Nothing there is new and it's stuff I've said numerous times before.
You said you were going to find out more. i was referring to that.
Yeah but every Maxwell product is based on 20nm because once they dropped Maxwell to 16nm they called it Pascal to make it sound like it was some great new technology. 20nm can be seen as the worst case scenario, as it always has been.
Polygonal_Sprite
Member
Right but knowing the final RAM target is a lot different to knowing the specific chip architecture final hardware will use especially when they are so similar from a performance standpoint anyway whether they go with Maxwell or Pascal.
So you think Emily has access to finished retail hardware ?
All the info info she has is based on devkits.
I personally couldn't give a toss which architecture it uses, I'm buying it to play Nintendo games regardless.
Just to be clear, how do you check your sources are legit ? I'm sure you do a better job than just listening to random twitter users who send you private messages
.I think she has access to sources that have access to final retail hardware yes.
Interfectum
Member
Bottomline, eventhough this has been asked/answered numerous times I'm sure but... based on your contacts and their impressions of the dev kits could the Switch theoretically run the big AAA third party current gen games without notable sacrifice?
Beyond everything else... Maxwell, amount of RAM, etc this is the most important question.
Skittzo0413
Member
Good to know, thanks. Something is weird about the Switch having an internal fan in the tablet (and apparently not in the dock, though that can be added even if it's not in the patent) yet still reaching 5 hours of battery. I'm not sure if they'll realistically reach that target, but it would only make sense if it is on a 16nm process.
That article sure did cause a Gaf meltdown didn't it...
Schnozberry
Member
That's not true. There were some major changes made between the Tegra X1 and Pascal based GP104 for desktop cards. Especially related to FP16 performance, which I believe will be important to Nintendo's strategy to get the most out of the Switch.
http://www.anandtech.com/show/10325/the-nvidia-geforce-gtx-1080-and-1070-founders-edition-review/5
The Tegra X1 SM design would actually be a better choice than Pascal for this reason. Especially if they were able to die shrink it down to 16nm.
sir_bumble_bee
Member
Did the document specifically mention the fan being active in portable mode?
Skittzo0413
Member
brochiller
Member
Would it be normal for the final dev kit and shipped product to have different chip architectures in them? Seems weird but I don't know much about this
sir_bumble_bee
Member
Interesting. I hope this thing doesn't get too noisy while in my hand (or face depending on if Nintendo utilize the VR-like capabilities present in the documents)
Just to be clear, how do you check your sources are legit ? I'm sure you do a better job than just listening to random twitter users who send you private messages
Pascal information was provided by individuals I've worked with for years. The talk of Pascal wasn't limited to only me. Other outlets heard the information as well and most were 99% certain it was happening.
A sub paragraph after the one posted they specifically talk about the fan noise and lowering its RPMs in order to reduce that noise along with power consumption, whether that means inaudible or just not that loud is unclear.
What I want to know is why people are taking this crazy article as fact? It's like if an article is published, it must be true kinda thing. Is there something about this that warrants believing it?
Emily seems to disagree with you on this one though.
He said the final devkits are definitely Maxwell
The final dev kits are Maxwell.
Pascal is smaller, generates less heat, and is more power efficient than Maxwell so Pascal is the better option.
The difference isn't anything mind blowing though in performance the main thing that would suffer from retail units using Maxwell would be battery and heat.
brochiller
Member
The guy who wrote it authored some video game related books and seems to have been pretty well respected. The way this article reads makes it seem like he forgot how to be a journalist, however.
So does that mean that the final hardware has to be Maxwell also? I'm going to assume that it does, but the differences might not matter since we're talking dev kits vs. actual production hardware.
The guy has no technical knowledge. He probably hasn't ever heard of precision in his life. As for 16nmFF being cheaper, we're talking about the same company that used 65nm in a 2011 handheld. They don't care, especially if the handheld team is calling the shots now.
- Status