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

Status
Not open for further replies.
The tweet in Japanese has more details. Here's a translation from Google:

また、最近問題の解像度に関し、特許情報だけでは1080P外部→720Pということは断定できず、と安田さん。むしろ内部解像度WQHD→1080Pという可能性も、と。小型液晶の解像度も1080Pの割合が上がってきており、あえて720Pにする動機は乏しい、と。

Also, regarding the resolution of the problem recently, it can not be concluded that patent information alone is 1080P external → 720P, Mr. Yasuda. Rather the possibility of internal resolution WQHD → 1080P, and so on. The resolution of small liquid crystal is also increasing at 1080 P, and motivation to dare to 720 P is poor.
 
No. NO. This device does not have a 1080p screen. Just NO.

It would be exceedingly strange, especially considering that UE4 preset. If that preset is correct as a general guideline, then the 1440p would drop to 960p resolution, rather than 1080p as this Ace Yasuda claims. Is Ace Yasuda a reliable source for this stuff at all (who is he anyway)?
 
The bus speed being 5 gbps is also super wrong. Tegra K1 is 17 Gbps, Tegra X1 is 25 Gbps. Eurogamer's spec sheet has 25.8 Gbps.

I'd just ignore those tweets without other evidence.
 
The screen is probably going to be 720p. It was just a tidbit from a Japanese analyst going through the recent postulated patents it seems.
God damn relax lol. Still all rumors here.

On the Jimmy Fallon show the port on the bottom of the Switch did look like USB Type C....
It is an USB Type-C connector.
 
God damn relax lol. Still all rumors here.

On the Jimmy Fallon show the port on the bottom of the Switch did look like USB Type C....

Hehe I am good.. just wanted to get ahead of this one early. But yeah, USB-C is pretty much a given, which is a great move considering the form factor.
 
Remember, Takashi is quoting an analyst. The same analyst said in November this:



__



Internally, yes. It's in the patents. The output from the dock is HDMI.

And that's that! Time for sleep.

Hehe I am good.. just wanted to get ahead of this one early. But yeah, USB-C is pretty much a given, which is a great move considering the form factor.

Ok cool haha. Displayport and USB Type C look kinda similar from afar but with all the info out now I agree it is a given.
 
You always make good points, and educated ones at that. But in this case I sort of have to point out that although you may be correct in a vacuum, you also admit to a range of variance in the final Switch specs that pretty much agrees with my assertion that we don't know what performance will be upon release. End of the day, we know neither the specifics of the hardware nor the customization Nintendo and Nvidia worked toward.

It's too soon to settle on givens.

Thanks, I don't mean to sound concrete, if you take GTX 1060 and test a bunch of games against RX 480, you will see it wins most and loses some. This is more about optimizations at this point and if they choose to use fp16 (mixed precision) which can have a small or large impact depending on what the GPU is asked to do, a lot of effects can be done in fp16 and if that is what is majorly being asked, it can improve performance drastically.

We should also look at Doom 2016, it is probably one of the best optimized games we have seen this generation, that is optimized at an API level surrounding vulkan, which Switch is somewhat built around, it's own NVN API is rumored to be based on, NVN is speculated to mean Nintendo Vulkan Nvidia.

In terms of other customizations my post is ignoring them unpurpose to give a base performance, no mixed precision was speculated, no checkerboard rendering, memory compression or other features that we know Switch can do. What I did was give context to SM counts we assume, Clocks we know and Architecture we know (again there is no performance difference between maxwell and pascal) that means we know the base performance of the device given any number of cuda cores, so when we say 2SM or 3SM, we know what that entails.

I'm trying to give posters who don't really understand tech, a reference of power, something other than the complete sky is falling knee jerk reactions we've gotten over the past 2 days, so giving baseline performance right now is more important than talking about if a developer utilizes everything available to them. We just shouldn't expect them
because it will be more rare, and the variance is huge, mixed precision alone could mean the difference between 45% xb1 and 80% xb1. Checkerboard rendering can mean the difference between 720p 60fps and 1080p 60fps, reality is that graphics have met a new horizon in the last 9 months or so, but the industry as a whole hasn't moved in that direction and a large part of that is only x1/pascal/vega offer these new breakout features.
 
I think it is fair to mention that there is no rumour that mentions 3 SM. Not even a fake one. As of now, 3 SM is only a speculation created here to try to cope with the conflicting information.

It's not total fantasy, but it might well be that Occam provides another reason for the fan in the tablet.
 
Thanks, I don't mean to sound concrete, if you take GTX 1060 and test a bunch of games against RX 480, you will see it wins most and loses some. This is more about optimizations at this point and if they choose to use fp16 (mixed precision) which can have a small or large impact depending on what the GPU is asked to do, a lot of effects can be done in fp16 and if that is what is majorly being asked, it can improve performance drastically.

We should also look at Doom 2016, it is probably one of the best optimized games we have seen this generation, that is optimized at an API level surrounding vulkan, which Switch is somewhat built around, it's own NVN API is rumored to be based on, NVN is speculated to mean Nintendo Vulkan Nvidia.

In terms of other customizations my post is ignoring them unpurpose to give a base performance, no mixed precision was speculated, no checkerboard rendering, memory compression or other features that we know Switch can do. What I did was give context to SM counts we assume, Clocks we know and Architecture we know (again there is no performance difference between maxwell and pascal) that means we know the base performance of the device given any number of cuda cores, so when we say 2SM or 3SM, we know what that entails.

I'm trying to give posters who don't really understand tech, a reference of power, something other than the complete sky is falling knee jerk reactions we've gotten over the past 2 days, so giving baseline performance right now is more important than talking about if a developer utilizes everything available to them. So we just shouldn't expect them
because it will be more rare, and the variance is huge, mixed precision alone could mean the difference between 45% xb1 and 80% xb1. Checkerboard rendering can mean the difference between 720p 60fps and 1080p 60fps, reality is that graphics have met a new horizon in the last 9 months or so, but the industry as a whole hasn't moved in that direction and a large part of that is only x1/pascal/vega offer these new breakout features.

Once again thanks for the explanation. I see where you are coming from, and your reasoning, much better now. It will be interesting to see if Nintendo focuses on some of the things that are available to them with this modern architecture. 3rd parties as well.
 
Except someone did his homework and it is competitive.

A customized ARM CPU not on windows which reserves power could easily pull that, regardless of console/pc would allow for that number.. However that won't be happening and it's downclocked. I've already stated I didn't expect downclocks. Yet be like others and forget I did nail what type of CPU it was going to end up with, though to be fair thraktor/blu were on the money once hints came out.
It's not just Thraktor/blu - sane gaffers have been saying ARMv8 made the most sense for NX way before there were any actual hints of what its CPU architecture was - we've been going over potential ARMv8 configurations in speculation threads for .. I dunno, since there were rumors of NX?

Re the Switch/ps4 juxtaposition, here's a very rudimentary one: https://browser.primatelabs.com/v4/cpu/compare/1398129?baseline=1396658

To see the DF picture, divide the left-side scores by 1.55. Let me help - first column is A57 @ 1.55GHz, second column - Jaguar @1.6GHz, third colum - first column / 1.55:

Code:
Single-Core Score       1304        967         841
AES                     945         916         610
LZMA                    1221        1191        788
JPEG                    1561        1177        1007
Canny                   1659        988         1070
Lua                     1173        799         757
Dijkstra                1416        1366        914
SQLite                  1206        751         778
HTML5 Parse             1381        902         891
HTML5 DOM               1259        819         812
Histogram Equalization  1697        1022        1095
PDF Rendering           1631        926         1052
LLVM                    1562        825         1008
Camera                  1700        1003        1097
SGEMM                   472         300         305
SFFT                    977         472         630
N-Body Physics          778         650         502
Ray Tracing             892         744         575
Rigid Body Physics      1424        977         919
HDR                     1743        1333        1125
Gaussian Blur           1284        813         828
Speech Recognition      964         1012        622
Face Detection          1355        861         874
Memory Copy             1399        1056        903
Memory Latency          1821        3459        1175
Memory Bandwidth        1312        647         846

And this, ladies and gentlemen, is how you bring a superior uarch to mediocre performance.
 
I think it is fair to mention that there is no rumour that mentions 3 SM. Not even a fake one. As of now, 3 SM is only a speculation created here to try to cope with the conflicting information.

It's not total fantasy, but it might well be that Occam provides another reason for the fan in the tablet.

Yeah, you're right. I'm one of the offenders here, both because of wishful thinking and because of the fan. I will totes own up to it and eat crow once those silly hopes are dashed properly.
 
It's not just Thraktor/blu - sane gaffers have been saying ARMv8 made the most sense for NX way before there were any actual hints of what its CPU architecture was - we've been going over potential ARMv8 configurations in speculation threads for .. I dunno, since there were rumors of NX?

Re the Switch/ps4 juxtaposition, here's a very rudimentary one: https://browser.primatelabs.com/v4/cpu/compare/1398129?baseline=1396658

To see the DF picture, divide the left-side scores by 1.55. Let me help - first column is A57 @ 1.55GHz, second column - Jaguar @1.6GHz, third colum - first column / 1.55:

Code:
Single-Core Score       1304        967         841
AES                     945         916         610
LZMA                    1221        1191        788
JPEG                    1561        1177        1007
Canny                   1659        988         1070
Lua                     1173        799         757
Dijkstra                1416        1366        914
SQLite                  1206        751         778
HTML5 Parse             1381        902         891
HTML5 DOM               1259        819         812
Histogram Equalization  1697        1022        1095
PDF Rendering           1631        926         1052
LLVM                    1562        825         1008
Camera                  1700        1003        1097
SGEMM                   472         300         305
SFFT                    977         472         630
N-Body Physics          778         650         502
Ray Tracing             892         744         575
Rigid Body Physics      1424        977         919
HDR                     1743        1333        1125
Gaussian Blur           1284        813         828
Speech Recognition      964         1012        622
Face Detection          1355        861         874
Memory Copy             1399        1056        903
Memory Latency          1821        3459        1175
Memory Bandwidth        1312        647         846

And this, ladies and gentlemen, is how you bring a superior uarch to mediocre performance.

Sorry vut I dont understand anything of this...could you explain it a little bit for an idiot like me? :3
 
Yeah, you're right. I'm one of the offenders here, both because of wishful thinking and because of the fan. I will totes own up to it and eat crow once those silly hopes are dashed properly.

I'm just going to mark the fan up to needing to cool the device in 100f+ degree weather.

Sorry vut I dont understand anything of this...could you explain it a little bit for an idiot like me? :3

middle column is PS4, right column is switch CPUs per core performance.
 
Sorry about double post!

3SM means a more expensive chip than a standard X1 chip for the same/less performance.

I mean why not?

Actually a bit more performance in docked mode than a standard X1. Thraktor made a post in which he explained (with graphs and stuff even) that having more SMs as opposed to higher clock speeds would save battery life as it draws fewer watts. That would be the reason why they could choose that.

Most likely, though, they just didn't, and went cheap. But it is at least something we can think and speculate about (speculation doesn't hurt anyone - except if you cling to that speculation as a hope for better days, which frankly would be one's own fault).
 
I'm just going to mark the fan up to needing to cool the device in 100f+ degree weather.



middle column is PS4, right column is switch CPUs per core performance.

Could you explain how this would translate to the overall CPU picture? The per A57 core situation is just about similar at 1 GHz, but what configuration of cores can we reasonably expect for this chip?

Edit: I found on NVIDIA's website that the Tegra X1 chip has 4x A57 cores on it. So that would mean that the CPU has 4x this similar core performance, while Jaguar (if I'm not mistaken) has currently 7 cores freed for development (at first it had 6 I believe, but they added one, at least in the PS4 - but maybe it is not fully accessible?), I think.
 
And this, ladies and gentlemen, is how you bring a superior uarch to mediocre performance.

It will be interesting if they bump the CPU clocks if they run into any problems. Also interesting if they open up the A53 cores that I'll assume is on there for the OS, leaving 1 or 2 cores for the OS still leaves 2 or 3 for developers, and they perform around 53% A57 performance iirc.

I can see why developers aren't having much issue here running their games on Switch if they targeted PS4 before, as long as they weren't using all cores at full tilt, it should be fairly straight forward, and slowing down some games to 30fps from 60fps would be all that they need as a worst case.
 
Could you explain how this would translate to the overall CPU picture? The per A57 core situation is just about similar at 1 GHz, but what configuration of cores can we reasonably expect for this chip?

I'm just got four cores from what we know. It might not even have the four A53 cores that other Tegra X1 devices have.
 
Could you explain how this would translate to the overall CPU picture? The per A57 core situation is just about similar at 1 GHz, but what configuration of cores can we reasonably expect for this chip?
Well, it says quad core in the OP.
So per thread, the CPU is roughly on par with PS4 / Xbone, but half* the cores to scale to.

*depending on OS core reservation

CPU looks to be Switch's strongest point, competitively.
 
I'm just got four cores from what we know. It might not even have the four A53 cores that other Tegra X1 devices have.

Yeah we should just speculate on this mostly, there is plenty of overhead if Nintendo can't pull of ports of games they want though, it wouldn't be unheard of to see a device wide clock increase, so don't worry too much about the CPU.

Basically 4 switch cores are going to perform around 4 PS4 cores. Developers have access to 6 cores, possibly 7?

Well, it says quad core in the OP.
So per thread, the CPU is roughly on par with PS4 / Xbone, but half* the cores to scale to.

*depending on OS core reservation

CPU looks to be Switch's strongest point, competitively.

Couldn't be more wrong, it is the weakest link, while it is closer to PS4's performance overall, CPU performance is paramount to a good port, you'll see fewer frames on Switch if 5, 6 or 7 cores are used on PS4, but if this becomes a serious enough issue, Switch does have the ability to increase clocks or possibly access hidden cores meant to run the OS, it just isn't something we should be too concerned with as things can still be scaled, logic is just a bit more of a pain.
 
Yeah we should just speculate on this mostly, there is plenty of overhead if Nintendo can't pull of ports of games they want though, it wouldn't be unheard of to see a device wide clock increase, so don't worry too much about the CPU.

Basically 4 switch cores are going to perform around 4 PS4 cores. Developers have access to 6 cores, possibly 7?



Couldn't be more wrong, it is the weakest link, while it is closer to PS4's performance overall, CPU performance is paramount to a good port, you'll see fewer frames on Switch if 5, 6 or 7 cores are used on PS4, but if this becomes a serious enough issue, Switch does have the ability to increase clocks or possibly access hidden cores meant to run the OS, it just isn't something we should be too concerned with as things can still be scaled, logic is just a bit more of a pain.
The more cores you have, the harder it gets to distribute workload between them all.
 
Could you explain how this would translate to the overall CPU picture? The per A57 core situation is just about similar at 1 GHz, but what configuration of cores can we reasonably expect for this chip?

Edit: I found on NVIDIA's website that the Tegra X1 chip has 4x A57 cores on it. So that would mean that the CPU has 4x this similar core performance, while Jaguar (if I'm not mistaken) has currently 7 cores freed for development (at first it had 6 I believe, but they added one, at least in the PS4 - but maybe it is not fully accessible?), I think.
7th core isn't fully available i think.
 

Not entirely sure that applies, we are only talking about 4 to 6 cores (I believe PS4's 7th core is half access because of OS) but more importantly, if you code to a core specifically, you just have that core doing that thing you told it to, for instance if they give sound to just 1 core, you get that same benefit at 3 cores or 100 cores, the other cores are free from having to deal with sound.

7th core isn't fully available i think.

yeah pretty sure it is half accessible and might not be available without turning off some features of the os, could mean that you have to ask sony for permission even.
 
I think that once the device is out and we see what's inside, we'll have a good laugh on how Nintendo managed to screw up another hardware design.

Some baffling decisions here and there. Not having a big.LITTLE setup sounds stupid in this day and age.
 
I think that once the device is out and we see what's inside, we'll have a good laugh on how Nintendo managed to screw up another hardware design.

Some baffling decisions here and there. Not having a big.LITTLE setup sounds stupid in this day and age.

That is just speculation, a dev sheet wouldn't need to know about locked processors.

I agree though that they screwed up on clocks, everything should be 25% higher clocked, it doesn't make it a different device but it would be much easier to deal with ports from ps4.
 
That is just speculation, a dev sheet wouldn't need to know about locked processors.

I agree though that they screwed up on clocks, everything should be 25% higher clocked, it doesn't make it a different device but it would be much easier to deal with ports from ps4.



It doesn't make it a different device indeed, but it does limit the potential much more. 25% of performance is important when it represent 6fps on a 30fps game and 15fps on a 60fps game.

I don't know who's responsbiel for Switch hardware design, but it's baffling that even when Nintendo is shopping for a nearly off the shelves design, they manage to drag it down in a way or another.

They're trying to design an hybrid console without giving themselves the means to do so. In the end, it'll end up a good handheld with barely enough juice to translate what's on its screen to a bigger screen.
 
It doesn't make it a different device indeed, but it does limit the potential much more. 25% of performance is important when it represent 6fps on a 30fps game and 15fps on a 60fps game.

I don't know who's responsbiel for Switch hardware design, but it's baffling that even when Nintendo is shopping for a nearly off the shelves design, they manage to drag it down in a way or another.

They're trying to design an hybrid console without giving themselves the means to do so. In the end, it'll end up a good handheld with barely enough juice to translate what's on its screen to a bigger screen.

yeah, I understand what you are saying, but it is still 45% of XB1 without any real funny business, that is still enough to do 720p version of 1080p xb1 games. It is also still 4 times more powerful than Wii U, without the need to display graphics on a second screen. Honestly Bayonetta 2's Starfox scene was pretty good, considering that can do 4x better with Bayonetta 3 or Starfox resurrection (because that is what it needs at this point) along with all the modern effects like god rays, and if the developers do use mixed precision, it's going to be a decent console.

big.little makes sense when you have phones with long idle times. For gaming? Not so much

You need something to run the OS in the background, using full A57 cores just kills battery life, meanwhile A53 cores sip at 1ghz, I wouldn't be surprised if it has them and they just aren't accessible by developers. We will know if we hear about any cpu core split for developers, but I think they need all 4 cores to really do any porting IMO.
 
yeah, I understand what you are saying, but it is still 45% of XB1 without any real funny business, that is still enough to do 720p version of 1080p xb1 games. It is also still 4 times more powerful than Wii U, without the need to display graphics on a second screen. Honestly Bayonetta 2's Starfox scene was pretty good, considering that can do 4x better with Bayonetta 3 or Starfox resurrection (because that is what it needs at this point) along with all the modern effects like god rays, and if the developers do use mixed precision, it's going to be a decent console.

Aren't a lot of XB1 games not running at 1080p either?
 
yeah, I understand what you are saying, but it is still 45% of XB1 without any real funny business, that is still enough to do 720p version of 1080p xb1 games. It is also still 4 times more powerful than Wii U, without the need to display graphics on a second screen. Honestly Bayonetta 2's Starfox scene was pretty good, considering that can do 4x better with Bayonetta 3 or Starfox resurrection (because that is what it needs at this point) along with all the modern effects like god rays, and if the developers do use mixed precision, it's going to be a decent console.



Yeah... More like between 40 and 45.
That is, if XB1 targets 1080p, which it does like 1 game for 10 ?
That is... For Switch docked to target 720p (which makes you wonder what handheld would be targeting).
Let's be fair, with the current hardware available, it could've been a better machine for these ports. Not that it matter for me. But itns clear Switch wasnt made with that in mind. It's a handheld which ups its clocks when docked to up the resolution on TV. Nth more, nth less.



As for the CPU cores, no, there's no hidden CPU cores or whatsoever to devs. Nintendo usually document that kind of stuff. They did for 3DS when 1core out of 2 was available, describing one as syscore and the other as appcore.
 
Not really. You already add up funny business in this.

Cool can you point them out so I can remove them?

Aren't a lot of XB1 games not running at 1080p either?

XB1 targets 1080p and uses dynamic resolution to keep performance from tanking, Switch COULD use that, or they can go with Checkerboard rendering to have a smoother look to the game. I'm trying not to dwell on features because some people can't handle it, they are way too broken over clock speed news, but Switch should handle XB1 games at a reasonable lower resolution without issue.
 


My point was that XB1 struggles to reach 1080p. So Switch will struggle to bring that to 720p. Which will drag down handheld res even further.



Cool can you point them out so I can remove them?



XB1 targets 1080p and uses dynamic resolution to keep performance from tanking, Switch COULD use that, or they can go with Checkerboard rendering to have a smoother look to the game. I'm trying not to dwell on features because some people can't handle it, they are way too broken over clock speed news, but Switch should handle XB1 games at a reasonable lower resolution without issue.



That is inaccurate. XB1 targets 1080p for a few games. And it often uses dynamic res to target 900p.
As for checkerboard rendering, this is no magic and you still hurt IQ.

Your point is that Switch should be able to handle XB1 games when XB1 targets a resolution it can't even handle most of the time. XB1 is a 900p/720p machine.
 
Status
Not open for further replies.
Top Bottom