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

Status
Not open for further replies.
It is going to be better than the Wii U / 3ds and it's going to be portable.. which I think is win win regardless of the weak specs.

The specs you listed are for a chip called "Parker" which is made by Nvidia and used in self driving cars. It's not a leak- those specs were released by Nvidia months ago.

Regarding the Switch, it is not using "Parker" because Nvidia themselves have said it uses a custom SoC, which "Parker" is not. So I don't know why those specs were presented as a leak for the Switch specs.


Now, the Switch MIGHT wind up using a custom SoC somewhat similar to Parker, since Parker itself is fairly similar to Tegra X1, which is what is reportedly in the Switch devkits. However, certain parts of Parker, such as the Denver CPU cores, would apparently not make any sense to be used in a game console. For a car computer they do make sense.

Either way, we've landed on a fairly commonly shared consensus that the maximum docked performance (if that's even a thing) of the Switch GPU will be between 512 and 768 GFlops, which is ~3-4x more powerful than the Wii U, not accounting for architectural/toolchain differences.

Thank you!

I saw this News right here : http://www.zeldainformer.com/news/rumor-nintendo-switch-nvidia-architecture-unveiled-by-japanese-journalist
 
Really curious to see the detailed specs of this custom Tegra Nvidia is doing, not cause i really care about raw power on a Nintendo console, which i will mainly use as handheld also, though it could be crucial in order to get multiplatform games with XB1 and PS4...So now beint TFlops so common right now to describe performance, what's the most realistic TFlops number that we can expect from Switch based on what we have? Xbox One is 1.25 TeraFlops, i guess if Switch hits at least 1TFlops ports should not be a problem right?
 
I am hoping that Nintendo decides to go with 1 pool of faster memory. From what ive read it seems that two pools of memory just require more work from a developer to take advantage of. This also seemed to be a concern for the x1 which seems to have gone the same route using 2 pools of memory rather than 1.

Even if the specs are rather low.. it would seem if the architecture was rather simple and easy to develop for it would go a long way to at least appease some concerns for third parties.
 
Really curious to see the detailed specs of this custom Tegra Nvidia is doing, not cause i really care about raw power on a Nintendo console, which i will mainly use as handheld also, though it could be crucial in order to get multiplatform games with XB1 and PS4...So now beint TFlops so common right now to describe performance, what's the most realistic TFlops number that we can expect from Switch based on what we have? Xbox One is 1.25 TeraFlops, i guess if Switch hits at least 1TFlops ports should not be a problem right?

It's completely speculative at this point. When in portable mode, I think we're likely to see somewhere around .5 TFlops. While in docked mode, somewhere near .8 TFlops. This is all assuming Nintendo is on 16nm and using the newer Pascal GPU from Nvidia. If they stuck with the older Maxwell part, both of those numbers go down by 20-30%
 
Didn't that bet include FP16 performance too?

I'd be shocked if it's not over 1TFlop in FP16 precision. Sorry AzaK :(
In portable mode? I wouldn't be shocked. (I also wouldn't be shocked if it reaches just about that. It's borderline)

Anyway, I think using "TFLOPS" for FP16 without explicitly qualifying that every single time you use it is problematic.
 
In portable mode? I wouldn't be shocked. (I also wouldn't be shocked if it reaches just about that. It's borderline)

Anyway, I think using "TFLOPS" for FP16 without explicitly qualifying that every single time you use it is problematic.

No, I was talking about the maximum power- as in when docked. The bet was simply that the Switch would never get over 1 TFlop, FP32 or FP16, in any scenario (if I remember correctly).

And yeah I can see the whole FP32/FP16 thing getting confusing. I think realistically we should only be talking about FP32 performance because it's very difficult to actually determine what percentage of code could reliably use 16 bit precision, and because that number would likely change drastically from game to game.
 
Zeldainformer said:
Someone in Japan may have just leaked some final specs for the NVidia architecture inside Nintendo Switch. It wasn't too long ago that we got a supposed Dev kit leak about the current specs in those units at that time that has had some outside confirmations, but now we have something that appears on the surface to be a much more tantalizing prospect towards the final spec sheet. According to Nishikawa Zenji, who apparently is a really famous Japanese journalist who knew about the PlayStation 4 Pro long before it was unveiled and simply just in the planning stages, the Nintendo Switch will likely be based upon the Pascal "Parker" architecture.

For those unfamiliar with those terms, Pascal is the actual architecture used in the 1050+ range of GPU cards (their latest GPUs, and would line up with NVidia's own press release) and the fact "Parker" is mentioned is a reference to the yet to be released Tegra X2. So, what we can infer (while having no exact specs) is that this is a custom Tegra X2 chip feature Pascal architecture. While this doesn't really give us a full hint at what that means, here is what is currently known about the Tegra X2:

Parker delivers 50 to 100 percent higher multi-core CPU performance than other mobile processors(2). This is thanks to its CPU architecture consisting of two next-generation 64-bit Denver CPU cores (Denver 2.0) paired with four 64-bit ARM Cortex A57 CPUs. These all work together in a fully coherent heterogeneous multi-processor configuration.

The Denver 2.0 CPU is a seven-way superscalar processor supporting the ARM v8 instruction set and implements an improved dynamic code optimization algorithm and additional low-power retention states for better energy efficiency. The two Denver cores and the Cortex A57 CPU complex are interconnected through a proprietary coherent interconnect fabric.

A new 256-core Pascal GPU in Parker delivers the performance needed to run advanced deep learning inference algorithms for self-driving capabilities. And it offers the raw graphics performance and features to power multiple high-resolution displays, such as cockpit instrument displays and in-vehicle infotainment panels.
parker_specifications_two__large.jpg


It's notable this comes out to about 1.5 TFLOPS of performance, which is more than the Xbox One but less than the PlayStation 4. However, the journalist notes that it is likely the performance will be knocked down to 1 TFLOP in order to have a longer battery life when taken on the go. This would also be the first real suggestion we have had that performance is gimped purposefully to make the device mobile, though it may still be possible to increase performance when docked to support the 1080p resolution natively, ramping up an internal fan for cooling. Even still, at 1 TFLOP that makes the difference between the Nintendo Switch and the Xbox One about the same as the the difference between the Xbox One and the PlayStation 4.

That means it would still be not quite at Xbox One levels, but it would be close enough to where porting current generation games shouldn't be a huge hangup, especially when the portable side is likely sporting a 720p resolution (which will save on resource management). It is also notable that Tegra X2 has been purposefully designed for use in cars, which explains why it would be a fully custom chip based on that current technology rather than a straight X2. Thus, it's possible the overall performance is slightly higher or slightly lower in the final build. Either way, if true, this paints a much rosier path for the Switch.

The journalist also notes this could pave way for future Switch hardware that, like the Pro or Xbox One S, could see updated releases as Tegra technology advances forward every three years.

Personally, I hope this is all true. Even with the Pro and Scorpio on the horizon, the original Xbox One and PlayStation 4 will still be getting a vast majority of the big AAA titles making the possibility of those titles coming to Switch far more likely. The fact there is also a clear upgrade path down the line bodes well for Nintendo not falling behind too much, as even with Scorpio boasting anywhere from 6 to 8 TFLOPS, with the Switch being a success it's not inconceivable that advancements in Tegra tech could make the Next Switch come really close as well (if there is next), which may be behind what the PlayStation 5 or Xbox Two does, but close enough to still be there.

Essentially, if true, this is making the Switch viable for all the big AAA third party games, even if they won't be the very best resolution or fps on the market. The bottom line is that this speaks well for anyone looking into buying a Switch next year.
http://www.zeldainformer.com/news/r...-architecture-unveiled-by-japanese-journalist
 
Why does that article talk about FP16 Tflops and compare them to FP32 Tflops of the Xbone and PS4? Bunch of garbage.
 
It's perfectly fine to wish for certain specs but I wish people like the ZI writer above didn't frame it in terms of what PSXB games the Switch is likely to get. That's misunderstanding the point and also setting yourself up for disappointment.
 
ZeldaInformer said:
For those unfamiliar with those terms, Pascal is the actual architecture used in the 1050+ range of GPU cards (their latest GPUs, and would line up with NVidia's own press release) and the fact "Parker" is mentioned is a reference to the yet to be released Tegra X2. So, what we can infer (while having no exact specs) is that this is a custom Tegra X2 chip feature Pascal architecture. While this doesn't really give us a full hint at what that means, here is what is currently known about the Tegra X2..
Oh FFS.
 
Seriously I don't care to much about the flop count as flops are just a number and do not scale linearly to real world performance. If the CPU is on par or better than the competition, and the memory, is decent which it usually is for Nintendo hardware they should be able to get some decent downports even with a much weaker gpu and first party stuff will be amazing.
 
Are you interested in buying a bridge?

No, but since Pascal is 4 times more efficient than Maxwell, which is already way more efficient than the old GCN architecture in PS4 and Xbone, I think 1 Tflops in a portable console this small is absolutely in the realm of possibility.
 
No, but since Pascal is 4 times more efficient than Maxwell, which is already way more efficient than the old GCN architecture in PS4 and Xbone, I think 1 Tflops in a portable console this small is absolutely in the realm of possibility.

How amenable are you to wearing a 130lb backpack made of lithium ion batteries and water cooling equipment?
 
Wait, what did I miss about 1TF rumors/tease? Likely just docked and FP16, but still.
Nothing. The 'net is still running off JTX1, and now Parker.
 
No, but since Pascal is 4 times more efficient than Maxwell, which is already way more efficient than the old GCN architecture in PS4 and Xbone, I think 1 Tflops in a portable console this small is absolutely in the realm of possibility.
You are talking about FP32 here, right?

I won't just sell you a bridge, I'll sell you an entire country.
 
I think Switch will be 250gflops in portable mode (32fp) and 500gflops (32fp) when docked, it sounds most reasonable

I have come to terms that Nintendo never does what it should with hardware. But nintndo always tries to eclipse the previous platform. I can't remember the exact numbers but if the Wii U is 176 - 352 GFLOPs then the switch is going to be higher docked and in portable form imo.
 
The X1 is 512Gflop with the same 25GB/s bandwidth. So obviously Nvidia feel the Maxwell GPU with 256 cores is a ok match for that bandwidth. Plus if they did drop processing power they'd go for a lower clock rather than cutting cores.
I think Nintendo might want better memory bandwdith for their purposes with the GPU than Nvidia did with a Android TV box.

Also, smaller chip makes for cheaper chip. Nintendo wants to make profit per device. This will help them with that while underclocking does almost nothing unless the yields on the chip are terrible. 256 cores for 25GBps is terrible match and waste of die space and throwing away money IMO.

Basically either we'll get a 128 cores with 64bit bus, or 256 cores with 128bit bus.
 
I don't agree with this.

People are very commonly discussing performance in terms of which ports might and might not be possible. When doing that, you need to take minimum (that is, portable) performance as your metric, since every single game will have to run at those frequencies and the resultant performance.

Of course, this is particularly relevant on the CPU side.

I agree that you do need to take the minimum configuration into account when considering ports, but that doesn't mean that all quotes of potential performance should be taken as referring to portable performance. In any case, one would expect that, with a smaller screen, the handheld will display at lower resolutions and/or framerates in line with the reduction in performance (excepting, as you say, CPU performance, which I would expect to remain relatively constant between the two modes).

In this thread most of the time clock speeds are implicitly assumed whenever people compare the maximum possible performance of the chips in question.

Most discussions arguing that the CPU is comparable to the Jaguars or that GPU performance relative to resolution is not terribly far away from other consoles only look at the maximum performance at maximum clock speeds and thus disregard the fact that the target for multi-platform games will be what the CPU and GPU can deliver in mobile mode without killing the batteries in one or two hours.

My issue here is you're not actually responding to a poster and arguing that their expectation of CPU clock speeds are unrealistic, or providing a counter-argument of what you think is a realistic expectation, but instead claiming that "Everyone is assuming that the CPU cores will run at maximum clock speed" without actually quoting or citing a single poster assuming as much, let alone providing evidence that every single person in this thread is making such an assumption.

This is in the context of several pages of discussion where the broad, untargetted "You guys have far too high expectations" posts vastly outnumbered the posters who actually had unrealistic expectations, which detracts from those of us who want to have a reasonable discussion about the technical possibilities and are happy to respond to excessive exuberance in a positive and non-confrontational manner. I'll readily admit that you're far from the worst offender in this regard, I was simply annoyed at the proliferation of such posts in a thread that, from my experience on GAF, probably has the most realistic consensus expectations of any technical thread of the sort.

You are talking about FP32 here, right?

I won't just sell you a bridge, I'll sell you an entire country.

Obviously the claim of 4x improvement over Maxwell is bull, but it would theoretically be possible to hit 1TF FP32 if you go with a wide enough architecture at a low enough clock speed and have pretty low standards when it comes to battery life. Say 6 SMs at 650MHz (which would keep Azak happy at 998 Gflops). Of course that kind of sized chip would push the cost of the machine well beyond what Nintendo would be comfortable charging, so there's no chance that it would actually happen.
 
I think Nintendo will probably be satisfied with the difference between the Switch and Xbox one being the same as the difference between the xbox one and ps4. So expect Switch to do around 900 gflops FP32 at the very most. Which is honestly kinda amazing given the form factor.
 
I think not using a Parker as the starting point for the Switch SoC would be quite a misstep. Forget the Denver cores, those can be eliminated, but using the 16nm node and the better memory controller seems like the best approach. Even in terms of costs, unless Nvidia somehow asked for a so much higher price for this design compared to a custom Maxwell Tegra. I don't see any reason why Nvidia would do that. It would provide a better promotion for their Tegra line and Parker is their current chip to sell.
 
This is in the context of several pages of discussion where the broad, untargetted "You guys have far too high expectations" posts vastly outnumbered the posters who actually had unrealistic expectations, which detracts from those of us who want to have a reasonable discussion about the technical possibilities and are happy to respond to excessive exuberance in a positive and non-confrontational manner. I'll readily admit that you're far from the worst offender in this regard, I was simply annoyed at the proliferation of such posts in a thread that, from my experience on GAF, probably has the most realistic consensus expectations of any technical thread of the sort.

Thank you for saying it. As far as tech speculation threads this one seems pretty reserved, and people coming in and saying we're all gonna be let down don't really appear to be reading the majority of posts. Also, this is a tech thread, people asking why we are interested in specs for a Nintendo console when all we get is Nintendo games should be aware of where they're posting.


Also that ZeldaInformer post is very misinformed. Which is ironic, considering the name. This thread (and the other one) have some very good in depth posts which detail what we should be expecting and why.
 
Nothing. The 'net is still running off JTX1, and now Parker.

It's like we were months ahead of everyone else on Switch speculation because we were sceptical of the stupidest shit of claims of NX being powered by something that isn't Nvidia because some loud people and "leakers" on Twitter thought:

  • Nintendo wouldn't be creating a hybrid because of an out of context quote from Iwata about merging hardware divisions
  • That apparently the TX1 was a placeholder for a DMP GPU because Nintendo are making a handheld (not hybrid) using it
  • That apparently the NX is a home console powered by AMD because of a Semi-Custom Design win for something using ARM that does more than gaming
  • That outlets like Eurogamer, Kotaku, IGN and WSJ were lying and conspiring to get clicks
  • That apparently some tablet similar to the design of the Switch meant that Eurogamer's sources were lying because... ??? Reaching much?
  • That apparently because Nvidia are using Tegra which is a low-powered SoC for automotive and IoT that supposedly Tegra no longer does gaming. LOL

Ahh... And the list goes on.
 
Why does that article talk about FP16 Tflops and compare them to FP32 Tflops of the Xbone and PS4? Bunch of garbage.

For a lot of operations a flop is a flop, many of them don't need the precision offered by fp32. Of course they can't be compared 1 to 1 because some things do require fp32, but at the same time Fp16 shouldn't be considered as half as powerful as fp32 either.
 
Unless I'm missing something this dude at gamefaqs seemed to leak accurately loads of Switch stuff in August and claims that the SCD is alive and well.

Bullshit or possible true?
 
Unless I'm missing something this dude at gamefaqs seemed to leak accurately loads of Switch stuff in August and claims that the SCD is alive and well.

Bullshit or possible true?

Uhh could you quote them for me? Because all I'm seeing is bullshit.

He said Switch would have a 540p screen and that the dock would have an upgrade slot for an SSD and that apparently two years from now there will be a partnership between Google, Samsung and Nintendo to make a Switch Phone.

"Accurate"...
 
I think Nintendo might want better memory bandwdith for their purposes with the GPU than Nvidia did with a Android TV box.

Also, smaller chip makes for cheaper chip. Nintendo wants to make profit per device. This will help them with that while underclocking does almost nothing unless the yields on the chip are terrible. 256 cores for 25GBps is terrible match and waste of die space and throwing away money IMO.

Basically either we'll get a 128 cores with 64bit bus, or 256 cores with 128bit bus.

I hope they do want as much bandwidth as possible. But that doesn't mean the biggest graphics hardware developer around got it vastly wrong when they matched their 512Gflop chip with 25GB/s main memory.

As far as number of cores/clock goes. These are small SoC's to begin with and will have very low yield problems, considering its a mobile device heat/power consumption are of a higher priority than saving a tiny amount more per chip . Also to be honest its amazing you can make such a statement regarding cores and bandwidth.

Its akin to looking at Xbox One main memory bandwidth vs PS4 and saying that anything other 750Gflops for that console is a total waste of money.
 
Uhh could you quote them for me? Because all I'm seeing is bullshit.

He said Switch would have a 540p screen and that the dock would have an upgrade slot for an SSD and that apparently two years from now there will be a partnership between Google, Samsung and Nintendo to make a Switch Phone.

Accurate...

It's a mix but he seemed to have quite a few things correct months ago.



The NX (name subject to change) won't physically "dock" into any docking station but a charging cradle much like the Wii U controller does. But it is true that it will be a portable/console hybrid. Think of the NX as a reverse Wii U whereas the Wii U controller was a dummy that only received video from the console, with the NX the controller has the actual computing power that streams (doesn't dock) video into a console-type or wrongly named "docking station".

Clear and simple, so far so good. But as always, the devil is in the details... The NX with its integrated screen will only output and process 540p mostly to keep things cool and extend battery life. Meanwhile, the "docking station" that comes standard will be a bare bones receiver that outputs this 540p image on your TV which is decidedly a low resolution for today standards. Where it gets interesting is that this barebones docking station will have 2 upgrade slots (much like the N64 expansion pack slot before it) one for a Solid State Hard Drive and the other and most interesting, a custom chip capable of converting that lowly 540p image into a 1080p image or possibly even 4K.

Now, Nintendo is aiming to have this controller/screen/portable/dock sold in the $179-$200 price range. Custom chip image converter and the solid state drive sold separately. If I were a betting man, I'd say the'll have a more expensive deluxe package where everything is included.

Stop making too much out of the "detachable controllers". Understand, the main controller is the console itself and those detachable controllers will be used when you're using it as a portable on the go and IF the game supports a more simplistic multiplayer mode. Mainly, you'll be playing your NX without detaching it especially if played on a TV.

A few extra tidbits:

The NX will have 64GB available rewritable memory so you can upload and unload any games you've purchased via the nintendo network and take them on the go or display them on a TV. You can also just pop a 3DS-like cartridge and play that way. And if you've bought the SSD attachment for your "dock" you can pull games wirelessly on the fly and/or upload them from the SSD to take them on the go.

The NX dock will be compatible with Wii U Pro controllers.

The NX has integrated 3DS backward compatibility only. Again, if I were a betting man, I'd say nothing prevents nintendo from emulating Gamecube, Wii or even Wii U. Nothing even prevents the docking station from being connected to a device that reads and plays such discs. Will they? Only time will tell...

It will use Pascal architecture and a custom Nvidia chip. No finer details to report here but memory on the other hand...

It will only have 4GB of RAM but "word has it" that with the NX use of cartridges instead of disc media, the RAMifications won't dent performance much at all.

The real system seller being lined up is not Zelda BotW, but a new 3D Mario game and a souped up version of PokemonGO.

Nintendo is going all-in and for the long-haul. This is the last "console" you'll ever see from them. That's to say their plan is to keep updating it for millenia to come for anyone who is willing to spend more on technological prowress. Nintendo is fed up with constantly waging console war and they think they'll finally be able to concentrate, literally speaking, on making games.
 

...

Are you familiar with the term: Throw shit at the wall and see what sticks?

That is exactly what this guy is doing, he is no different to the other "leakers" on Twitter.

I just... Don't get how you can find this guy believable considering he got some lucky guesses from people's speculation of the NX hybrid from Eurogamer and Emily Rogers.

Are you not seeing the hundreds of other things he got wrong?

Seriously?
 
The laws of physics still apply in this universe

Nintendo is above the law /s.

Denver isnt necessarily out of the question. If utilized properly it can actually lend itself very well for games. I dont think Nintendo will use it mostly because you have to be careful with how you write ur code or your gonna stall the CPU. Most devs arent gonna take the time to deal with Denvers unique design.

Im still suprised people are grasping on to this fp16 thing. Im aware UE4 uses fp16 extensively but there were a lot of sacrifices made visually to make it work. You also have to be very cheeky with your shaders algos to make something that looks nice which warrents more work than many devs are willing to put in. Expecting NX to visually match XB1 is just unrealistic. You cant expect a 10-15 watt device to perform similarly to something that has a power envelope of nearly an order of magnitude higher.

I remember with WUST very similar things were happpening. "Oh that unknown logic block will do free lighting". "There is more shader cache" "VLIW5 vs VLIW4".

A lot of those didnt really do much and it turned out that unknown logic block just did DSP stuff for the gamepad.

Im not trying to be a downer or anything and insult anyone. I just dont want ppl to get overhyped then hit with dissapointment later on. With WiiU, after we learned more there was so much negativity haha. I just want to avoid that lol.
 
Status
Not open for further replies.
Top Bottom