• 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.

Confirmed: The Nintendo Switch is powered by an Nvidia Tegra X1

Status
Not open for further replies.

Hermii

Member
Well it is very close in terms of gflops. the switch is weaker than both the ps3 and the xbox 360 in portable mode if we believe the specs in the DF article ((256 * 2) * .3072 = 157.2864 gflops) compared to 192 gflops (ps3) and 240 gflops (xbox 360). The console is around twice as powerful as those consoles in dock mode (switch dock=393 gflops).

Sources: last gen specs: https://forum.beyond3d.com/threads/g...ent-gen.55710/
switch specs: http://www.eurogamer.net/articles/di...-spec-analysis

Yea in flops its also weaker than a Wii U when mobile, which Zelda and Fast shows isnt the case in real world. Maxwell flops >>>>>> last gen consoles flops.
 

AetherZX

Member
Yea in flops its also weaker than a Wii U when mobile, which Zelda and Fast shows isnt the case in real world. Maxwell flops >>>>>> last gen consoles flops.

It's not, apparently there's a higher clocked profile in portable mode which pushes it closer to 200 gflops.
 

KingSnake

The Birthday Skeleton
Would people be happy if all games on Switch hit 900p but with locked 30fps? And if so is it possible to get say Battlefield 1 to run at this on Switch?

I doubt Battlefield 1 would run decently above 720p without changes that it would make it not be Battlefield 1 anymore. Maybe EA would call it a "custom built Battlefield".
 

Hermii

Member
It's not, apparently there's a higher clocked profile in portable mode which pushes it closer to 200 gflops.

Forgot about that. We know Zelda uses that mode, but I doubt FAST does since it uprenders to 1080p when docked. Anyway my point about Maxwell still stands.
 

Pasedo

Member
I think it would require severe downgrades in addition to halving the framerate. Also I doubt it would be a cheap investment from EA, which means its not happening.

Yeh I'm just wondering if they can just use low to mid textures. Perhaps native resolution at 720p even and set every other feature down to a low to mid level. On a 7in screen I imagine low to mid settings would be perfectly fine. I hear its easy to port stuff over to Switch. Perhaps it's not that expensive.
 

Hermii

Member
Yeh I'm just wondering if they can just use low to mid textures. Perhaps native resolution at 720p even and set every other feature down to a low to mid level. On a 7in screen I imagine low to mid settings would be perfectly fine. I hear its easy to port stuff over to Switch. Perhaps it's not that expensive.

Most if not all of the comments we heard about how great the Switch is to develop for comes from indie developers. I doubt the same would apply to porting hardware taxing aaa games. Im going to be a cynic and say the reason we havent heard from those devs is because there are no games like that in development for Switch. Hope E3 proves me wrong.
 

KingSnake

The Birthday Skeleton
Or maybe is that Nintendo has a technical prowess so, so good that they are capable of reaching the maximum limits of their new hardware day one?

Zelda BotW is quite a formidable game though, far beyond what Nintendo has done so far.

Sure, it takes more time and practice to learn a new system and optimise the games better, but very few future Nintendo games will have the complexity and ambition of BotW. Out of the announced games only Xenoblade 2 has the potential to go beyond. Out of all Nintendo franchises I can't think of too many that could do that.

That's not to say that the games won't be very good looking. But looking at Zelda and saying "if this can be done at launch then surely most of the upcoming games will push even further" is disingenuous towards Zelda.
 

KingSnake

The Birthday Skeleton
There is nothing exotic about the X1 so there is nothing to learn or to get used to.

There are new tools that Nintendo developers have to learn. Plus building up "best practices" on what works better and what shouldn't be done needs some experience build up.

Because the architecture alone doesn't provide the full story about that, it matters a lot how it interacts with the OS and API.
 

2+2=5

The Amiga Brotherhood
Not really. The computing is all still done inside the Switch itself.
The dock help performance, so it fits the patent.
You should take into account that in any case games will be aimed to the undocked power, not to an eventual optional accessory, so in any case even if a new dock will give more power if would be used for better resolution and framerate and maybe some effects, nothing more.

Yea in flops its also weaker than a Wii U when mobile, which Zelda and Fast shows isnt the case in real world. Maxwell flops >>>>>> last gen consoles flops.

Let's stop with this myth, flops are flops, there aren't maxwell flops or whatever flops, they are like kilometers/hours(or miles/hours), a 2017 car can be more efficent, being more comfortable, less polluting etc than a 80s car, but if they do the same km/h then they do the same km/h, period.

Gflops are the number of floating point operations per second, why they are important? Because everything is a floating point in 3d, a 3d model of a big game is made of dozen or hundreds of thousands vertices that have at very minimum 3 floats for the position and 2 floats for the texture coordinates(but more realistically they have many more floats of data), also translation and rotation data are other 3 floats each that have to be multiplied with every single vertex and so on.

You can do floating point operations in a way or another but in the end is the gflops that count.

If games on a newer gpu look better than games on a older gpu with comparable gflops is probably because of newer shaders and the help of more ram, more powerful newer cpu and so on.
 

gundalf

Member
With just the USB port that Switch tablet has now the possibility of an SCD that helps in any significant way is very slim. There's just not enough bandwidth to support that. Not to mention that it would still face the bottlenecks in the RAM bandwidth and CPU processing power. A SCD for the current Switch sounds like an awful idea.

Would it free up enough bandwidth if the Switch doesn't output a DisplayPort signal over the USB? If the dock has its own GPU, theoretically the dock could output the HDMI from that instead of using USB bandwidth for the video signal.

There's not enough bandwidth for anything. As people already pointed out the only way it works is if the dock does everything, at which point it's just a seperate device and doesn't need the tablet in the first place

The Switchs USB-C alternative mode can go up to 10.2 Gbit/s bandwidth which is around PCIe x2, so If SCD would ever come real, then it would replace the whole Dock and from there breakout USB and HDMI.
 

Pasedo

Member
Most if not all of the comments we heard about how great the Switch is to develop for comes from indie developers. I doubt the same would apply to porting hardware taxing aaa games. Im going to be a cynic and say the reason we havent heard from those devs is because there are no games like that in development for Switch. Hope E3 proves me wrong.

I'm wishful that publishers will somehow do it but to manage my own expectations I believe it'll just be a 1st party, Indie, 3rd party basic graphical games e.g Lego, Sports, Dancing games machine lol.
 

Hermii

Member
The dock help performance, so it fits the patent.
You should take into account that in any case games will be aimed to the undocked power, not to an eventual optional accessory, so in any case even if a new dock will give more power if would be used for better resolution and framerate and maybe some effects, nothing more.



Let's stop with this myth, flops are flops, there aren't maxwell flops or whatever flops, they are like kilometers/hours(or miles/hours), a 2017 car can be more efficent, being more comfortable, less polluting etc than a 80s car, but if they do the same KM/H then they do the same KM/H, period.

Gflops are the number of floating point operations per second, why they are important? Because everything is a floating point in 3d, a 3d model of a big game is made of dozen or hundreds of thousands vertices that have at very minimum 3 floats for the position and 2 floats for the texture coordinates(but more realistically they have many more floats of data), also translation and rotation data are other 3 floats each that have to be multiplied with every single vertex and so on.

You can do floating point operations in a way or another but in the end is the gflops that count.

If games on a newer gpu look better than games on a older gpu with comparable gflops is probably because of newer shaders and the help of more ram, more powerful newer cpu and so on.

Newer shaders is part of the gpu architecture. Flops is a theoretical peak, how often and how close it can get to the theoretical peak in real world scenarios depends on the architecture among other things. A better featureset also allows for more efficient programming techniques.
 
so does this mean that we may see some of the NVdia Shield games ported over to the Switch more easily?

I am really unfamiliar what kind of games Nvidia Shield has but I got the impression that it got some PC games, no?
 

120v

Member
so does this mean that we may see some of the NVdia Shield games ported over to the Switch more easily?

I am really unfamiliar what kind of games Nvidia Shield has but I got the impression that it got some PC games, no?

the sheild is basically a mini PC that runs ps360 era stuff pretty decently
 
Capcom is apparently seeing a challenge to port RE7 over because of bandwidth. This already isn't looking great for Switch AAA 3rd party support lol. I'm wishful that publishers will somehow do it but to manage my own expectations I believe it'll just be a 1st party, Indie, 3rd party basic graphical games e.g Lego, Sports, Dancing games machine lol.

which handheld has been able to get ports of the highend console games of the same generation? I cant think of any. PSP and Vita was close but even on those platforms the games that sold well usually where handheld type games.

switch will mostly rely on nintendo games, japanese RPGs, mobile game ports and indie games. This already will make it a viable platform to some.

I dont see a 3rd Party rennaissance happening unless it sells Wii levels.

That being said, Id love to see games like Dark Souls 3 on Switch. Portability improves playability for me.
the sheild is basically a mini PC that runs ps360 era stuff pretty decently

ps360 era level portable sounds good to me.
No idea but I'm thinking that we might see Switch games on Shield, eventually.

probably not nintendo games but sure, why not.
 

2+2=5

The Amiga Brotherhood
Newer shaders is part of the gpu architecture. Flops is a theoretical peak, how often and how close it can get to the theoretical peak in real world scenarios depends on the architecture among other things. A better featureset also allows for more efficient programming techniques.

Newer shaders are more efficent than older ones, flops are teorethical peaks for both older and newer gpus and peaks are peaks, meaning that usually regular flops are lower for both old and new chips.
Again a featureset may be more efficient than a older featureset(old car vs new car) but if in the end their gpus do the same gflops(km/h) they do the same gflops, period.
 

Hermii

Member
Newer shaders are more efficent than older ones, flops are teorethical peaks for both older and newer gpus and peaks are peaks, meaning that usually regular flops are lower for both old and new chips.
Again a featureset may be more efficient than a older featureset(old car vs new car) but if in the end their gpus do the same gflops(km/h) they do the same gflops, period.

All Im saying is that a 157gflop maxwell gpu will most likely in most real world scenarios outperform much older GPUs at higher flop counts.

I guess you could compare it to acceleration / road properties of a car vs top speed.
 
There is nothing exotic about the X1 so there is nothing to learn or to get used to.
Well, 2x fp16 isn't exotic in the mobile world, but no other console provided an increase in fp16 except for the the PS4Pro. It also has a full dx12 feature set.

Capcom is apparently seeing a challenge to port RE7 over because of bandwidth. This already isn't looking great for Switch AAA 3rd party support lol. I'm wishful that publishers will somehow do it but to manage my own expectations I believe it'll just be a 1st party, Indie, 3rd party basic graphical games e.g Lego, Sports, Dancing games machine lol.
Where did you get your info on the first statement?
 

2+2=5

The Amiga Brotherhood
All Im saying is that a 157gflop maxwell gpu will most likely in most real world scenarios outperform much older GPUs at higher flop counts.
What you see it's the effect of better cpu, more ram, better shaders, optimized games and so on, if you run the same unoptimized game on the same hardware but with a old 157gflops gpu first and then with a new 157gflops gpu later the results would be similar, the newer gpu will probably do slightly better but not the big difference that people seem to believe.
 

Ninja Dom

Member
Capcom is apparently seeing a challenge to port RE7 over because of bandwidth. This already isn't looking great for Switch AAA 3rd party support lol. I'm wishful that publishers will somehow do it but to manage my own expectations I believe it'll just be a 1st party, Indie, 3rd party basic graphical games e.g Lego, Sports, Dancing games machine lol.

Where did you read this?
 

tkscz

Member
What you see it's the effect of better cpu, more ram, better shaders, optimized games and so on, if you run the same unoptimized game on the same hardware but with a old 157gflops gpu first and then with a new 157gflops gpu later the results would be similar, the newer gpu will probably do slightly better but not the big difference that people seem to believe.

Which explains why Zelda runs the way it does.

Newer shaders are more efficent than older ones, flops are teorethical peaks for both older and newer gpus and peaks are peaks, meaning that usually regular flops are lower for both old and new chips.
Again a featureset may be more efficient than a older featureset(old car vs new car) but if in the end their gpus do the same gflops(km/h) they do the same gflops, period.

You have to explain that a little better. Saying "It is what it is" won't really help.

A new GPU can outproduce an old one based on the other hardware it works with, along with having the ability to produce new and better/more optimized shaders, lighting effects, etc... however, it can't produce as much of them at once as something with a higher FLoPs can. So the Switch can produce graphics beyond what we've seen on the PS360U, but if you ported one of their games onto the Switch without optimization to it's hardware, it will run "as good" as it did on that console.
 
It's game engine being unoptimize, hardware has little ti do with it. Twin Snake was on a more powerful system using Mgs2 engine and yet had slowdown
 

Mokujin

Member
Zelda BotW is quite a formidable game though, far beyond what Nintendo has done so far.

Sure, it takes more time and practice to learn a new system and optimise the games better, but very few future Nintendo games will have the complexity and ambition of BotW. Out of the announced games only Xenoblade 2 has the potential to go beyond. Out of all Nintendo franchises I can't think of too many that could do that.

That's not to say that the games won't be very good looking. But looking at Zelda and saying "if this can be done at launch then surely most of the upcoming games will push even further" is disingenuous towards Zelda.

I didn't imply that "surely upcoming games will push further" but we surely could expect that given the peculiarities of BotW release (new hardware, day one release, port from a different architecture) more efficient use of the hardware can be achieved through different ways in the future, the point about none, few or a nice number of games with a similar scope graphics wise is purely speculative and in noway disingenuous towards Zelda.

If we were 2-3 years into Swith life and BotW was a game designed from scratch for it your argument would have a lot more weight, but right now its like looking at KZ: Shadow Fall back when Ps4 was released and stating that the game was the limit that the new playstation was going to achieve.

And to make it clear I really don't care if it goes one way or the other I'm really, really happy with my Switch and the only things that have annoyed me are pricing and some minor left joycon issues, but from a technial point of view I'm quite satisfied.

There is nothing exotic about the X1 so there is nothing to learn or to get used to.

That's quite the BS there, by your logic Ps4 and X1 were also at their limit on release because there is nothing further from exotic than AMD x86 + AMD Gpu.
 
Having that tech in their back pocket may have just been to hedge bets if Switch failed to take off or to introduce a mid-gen upgrade to the devices later. Since the Switch actually seems to be doing OK out of the gate I seriously doubt that they want to muddy the waters so early and confuse their base with a power add-on in the first year.

Yeah I agree with this, but I also really don't see why it makes business sense to chase high end graphics for the Switch at all, even as an upgrade option, seeing as it likely won't be getting most high end AAA games regardless. And I don't think a third target power level will entice third parties to actually bring more games if they have to also make it run on the first two.

If they're gonna use the SCD patent I think it might be more for the "local cloud" functions, possibly for storage or for supplementing online play. The whole reward system seemed very well thought out for something that might never get used.
 

Pasedo

Member
Well, 2x fp16 isn't exotic in the mobile world, but no other console provided an increase in fp16 except for the the PS4Pro. It also has a full dx12 feature set.


Where did you get your info on the first statement?


Tbh I can't remember exactly where I read it but here's an article related to it. There was some sort of dev conference in Japan and there was discussion with Capcom on Switch. Seems like memory was an issue. Not sure how legit info is from this site.

http://nintendoeverything.com/more-...ry-and-capcom-targeting-re-engine-for-switch/
 

blu

Wants the largest console games publisher to avoid Nintendo's platforms.

bomblord1

Banned
If you can't run your code on X1 then 5 years more of time will not fix that.

Of course you can optimize for any hardware given time stating otherwise is just silly. Change resolution, effects, assets, and change the code to play to the strengths of the system (ex use FP16 effects where possible on Switch or on Xbox One optimize for the ESRAM).

No one just hits a compile button without changing anything and then goes well it didn't run on this completely different hardware so it never will and dumps the project.
 

z0m3le

Banned
The dock help performance, so it fits the patent.
You should take into account that in any case games will be aimed to the undocked power, not to an eventual optional accessory, so in any case even if a new dock will give more power if would be used for better resolution and framerate and maybe some effects, nothing more.



Let's stop with this myth, flops are flops, there aren't maxwell flops or whatever flops, they are like kilometers/hours(or miles/hours), a 2017 car can be more efficent, being more comfortable, less polluting etc than a 80s car, but if they do the same km/h then they do the same km/h, period.

Gflops are the number of floating point operations per second, why they are important? Because everything is a floating point in 3d, a 3d model of a big game is made of dozen or hundreds of thousands vertices that have at very minimum 3 floats for the position and 2 floats for the texture coordinates(but more realistically they have many more floats of data), also translation and rotation data are other 3 floats each that have to be multiplied with every single vertex and so on.

You can do floating point operations in a way or another but in the end is the gflops that count.

If games on a newer gpu look better than games on a older gpu with comparable gflops is probably because of newer shaders and the help of more ram, more powerful newer cpu and so on.

Theoretical is the key word, Xbox 360 has a shader efficiency around 60%, and it's shader count is actually hindered to only 217gflops iirc, which means that you are only capable of extracting about 130gflops from the 360 per frame, at least without extreme care to GPU bottlenecks like low level cache and thread wave allocation.

VLIW5 in Wii U is capable of about 80% efficiency, and so it's 176gflops is only capable of about 140gflops per frame, again you can painstakingly extract a bit more, but threads will allocate through the architecture and run into issues where only 4 of the 5 available stream processors can process data.

Maxwell architecture has over 96% efficiency, so that 196 gflops in handheld mode is 188gflops or 377gflops when docked. since 2005, these processors have gotten better at handling effects more efficiently as well, and you shouldn't directly compare gflop numbers because it just doesn't work that way. This is why generally Nvidia gflops will perform 4/3 better than GCN, this is without heavy Async compute engine utilization, but that is only recently wide spread and hasn't been a focus for Nvidia's drivers where they could gain back ground here.

We also haven't touched on fp16, as many graphic operations do not need 32bit values, this is a game by game basis, but with some developers reporting a 70% utilization in their own games, it is safe to say that it can certainly gain ground as developers get more and more use to writing code with 16bit floating point values in mind.
 

blu

Wants the largest console games publisher to avoid Nintendo's platforms.
Theoretical is the key word, Xbox 360 has a shader efficiency around 60%, and it's shader count is actually hindered to only 217gflops iirc, which means that you are only capable of extracting about 130gflops from the 360 per frame, at least without extreme care to GPU bottlenecks like low level cache and thread wave allocation.

VLIW5 in Wii U is capable of about 80% efficiency, and so it's 176gflops is only capable of about 140gflops per frame, again you can painstakingly extract a bit more, but threads will allocate through the architecture and run into issues where only 4 of the 5 available stream processors can process data.

Maxwell architecture has over 96% efficiency, so that 196 gflops in handheld mode is 188gflops or 377gflops when docked. since 2005, these processors have gotten better at handling effects more efficiently as well, and you shouldn't directly compare gflop numbers because it just doesn't work that way. This is why generally Nvidia gflops will perform 4/3 better than GCN, this is without heavy Async compute engine utilization, but that is only recently wide spread and hasn't been a focus for Nvidia's drivers where they could gain back ground here.

We also haven't touched on fp16, as many graphic operations do not need 32bit values, this is a game by game basis, but with some developers reporting a 70% utilization in their own games, it is safe to say that it can certainly gain ground as developers get more and more use to writing code with 16bit floating point values in mind.
That an over-generalization. Different workloads will exhibit different ALU efficiency. The only thing the paper specs tell us is that there cannot exist a workload where the ALU rate is higher.

Of course newer architectures tend to achieve better utilization for the same workload.
 

z0m3le

Banned
That an over-generalization. Different workloads will exhibit different ALU efficiency. The only thing the paper specs tell us is that there cannot exist a workload where the ALU rate is higher.

Of course newer architectures tend to achieve better utilization for the same workload.

Pretty good articles on AMD architectures on anandtech that will give you those numbers, but yes, I said without optimizations that is what you'd get. I'm actually being generous with vliw5, which pushed 5 spus together and only saw an average of 3.4 utilized on average in games, and the 80% number I use is actually stated to be above what the architecture actually achieves generally. http://www.anandtech.com/show/4061/amds-radeon-hd-6970-radeon-hd-6950/4

It's a really good read if you haven't taken a look at it, the article explains why AMD started grouping 4 spus together in the 6000 series cayman+ cards. There is another article about GCN on there, but I haven't recently looked at it, iirc it could hit 92% efficiency in practice, which shows again how much performance gains these architectures achieve at a base level of code utilization.

Edit: here's the gcn article http://www.anandtech.com/show/4455/amds-graphics-core-next-preview-amd-architects-for-compute, for other people in this thread, ps4 and xb1 use this architecture.
 
That didn't make Knack look like Uncharted 4.

Yea, Knack looks better than UC4. It really shows how easy it is to exploit when launch games looks so good. Also, Knack is more fun than UC4. Knack Forever.

Being serious, the jumps haven't been that grand so far from individual devs. It makes no sense to compare Sony Japan to Naughty Dog as Naughty Dog and Guerilla are sonys most technically proficient studios.
 

Hermii

Member
Yea, Knack looks better than UC4. It really shows how easy it is to exploit when launch games looks so good. Also, Knack is more fun than UC4. Knack Forever.

Being serious, the jumps haven't been that grand so far from individual devs. It makes no sense to compare Sony Japan to Naughty Dog as Naughty Dog and Guerilla are sonys most technically proficient studios.

Ok, Killzone Shadow Fall to Horizon then. You can't say thats not a huge jump.
 

10k

Banned
I think the sweet deal Nintendo got was to use up the 20nm Tegra Nvidia has ordered and couldn't sell on their own and then ASAP switch to the 16nm Tegra (which is essentially Pascal Tegra/X2) for a later model, like a Switch Pro.
 
I think the sweet deal Nintendo got was to use up the 20nm Tegra Nvidia has ordered and couldn't sell on their own and then ASAP switch to the 16nm Tegra (which is essentially Pascal Tegra/X2) for a later model, like a Switch Pro.

OOrrrrr, x2 isn't ready for mass, multi-million production 6 months ago.
 

AmyS

Member
Nvidia Tegra line:

Tegra APX
Tegra 2 <-------- 3DS almost used this, before Nintendo decided to go with DMP's PICA 200
Tegra 3 - "Kal-El"
Tegra 4 - "Wayne" - original Shield handheld
Tegra K1 - "Logan" (Kepler architecture) - Shield Tablet
Tegra X1 - "Erista" (Maxwell architecture) - Shield TV, updated Shield TV, Switch
Tegra "Parker" (Pascal architecture)

*Xavier (Volta architecture)
 
Status
Not open for further replies.
Top Bottom