• 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.

Nintendo Switch: Powered by Custom Nvidia Tegra Chip (Official)

z0m3le

Banned
This is only true if you compare them on dx11. Both Xbox1 and PS4 use low level APIs.

@Skittzo0413 The 970 uses a newer architectur then the 390. 390 vs 780(ti) would be a much better comparison. But usually Nvidia needs fewer Gflops for equal performance.

You can also compare GTX 750ti 1.4tflops to the PS4 1.843tflops, this similar performance still happens, infact the 750ti is usually faster.
 

ggx2ac

Member
To be honest I don't think it would be that big of an issue. Nintendo's internal teams would have had Nvidia hardware to work on from pretty much the moment the deal was done (perhaps two years before launch), and they dealt with arguably a bigger architectural jump from Wii to Wii U (moving from a fixed-function to fully programmable graphics architecture).

Edit: And CPU ISA really wouldn't be that big of a deal outside of whoever's working on the compiler (if they're not using a third party compiler). Also keep in mind that Nintendo have continuously been developing for ARM-based handhelds since the GBA.

Saving this post, for when apparently someone claims Nintendo are having development troubles because of using "foreign ARM architecture" where apparently handhelds don't count because apparently the console software devs have never worked on handheld software development before.
 

Doctre81

Member
It's not just nintendo who have experience with ARM but all three console manufactures considering wiiu, xbox one , and ps4 all have ARM co procressors inside.
 
This is only true if you compare them on dx11. Both Xbox1 and PS4 use low level APIs.

@Skittzo0413 The 970 uses a newer architectur then the 390. 390 vs 780(ti) would be a much better comparison. But usually Nvidia needs fewer Gflops for equal performance.

The 390 also performs better in games and it's aging like wine.
 
This is only true if you compare them on dx11. Both Xbox1 and PS4 use low level APIs.

@Skittzo0413 The 970 uses a newer architectur then the 390. 390 vs 780(ti) would be a much better comparison. But usually Nvidia needs fewer Gflops for equal performance.

You can also compare GTX 750ti 1.4tflops to the PS4 1.843tflops, this similar performance still happens, infact the 750ti is usually faster.

So if I understand this correctly, the major advantage of Nvidia flops over AMD flops exists only on PC, but there still would be a smaller advantage on consoles?

Also Pascal being a brand new architecture might provide an advantage over the GCN 1.1 architecture in PS4/XB1 (flop for flop, not outright)?
 
If they're the same family of chips, then theoretically it shouldn't matter. You'd be able to keep using your old Shift, until it reaches a point where new games ONLY supported a min version. Kind of like how while you can install iOS on old iPhones, you can only go so far back. Shift software would be the same
That's actually a good idea. Console generations last much longer than phones, let's say a new Switch releases every 4 years, but depending on specs, some newer games can carry over at lower settings or not be compatible, but that guarantees backwards compatibility with new devices forever onward until Nintendo decides to leave the Switch brand. Kinda like what Microsoft plans to do with Xbox. All Xbox games play on all Xbox systems by going an upgrade route.
People are used to that now, so it seems logical to do that with game consoles at this point.
 
That's actually a good idea. Console generations last much longer than phones, let's say a new Switch releases every 4 years, but depending on specs, some newer games can carry over at lower settings or not be compatible, but that guarantees backwards compatibility with new devices forever onward until Nintendo decides to leave the Switch brand. Kinda like what Microsoft plans to do with Xbox. All Xbox games play on all Xbox systems by going an upgrade route.
People are used to that now, so it seems logical to do that with game consoles at this point.

Yeah its pretty much what MS has been saying for a while now. They want you to build a library inside their ecosystem. That makes the ecosystem much harder to leave when it comes time to upgrade. There may not be a generational reset like there was with Nintendo and MS. I am not sure about Sony yet.
 
You can't fudge late 2013's 200W of processing power into an early 2017's 10W chassis no matter how much mental gymnastics you do.

Stop doing this to yourselves. Just be happy with Wii U power in portable form factor.
 

MDave

Member
This is only true if you compare them on dx11. Both Xbox1 and PS4 use low level APIs.

@Skittzo0413 The 970 uses a newer architectur then the 390. 390 vs 780(ti) would be a much better comparison. But usually Nvidia needs fewer Gflops for equal performance.

I should of added more context to my post, hah. Richard Leadbetter did that video this year in June: a year after the release of the R9 390?

The video shows two games running DX12, Rise of the Tomb Raider runs faster on the 970 while the 390 runs Ashes of the Singularity DX12 faster. It depends on the game and the drivers, but on average the performance is roughly in the same area as each other.

Also didn't the R9 390 come out like 10 months after the 970?
 
You can't fudge late 2013's 200W of processing power into an early 2017's 10W chassis no matter how much mental gymnastics you do.

Stop doing this to yourselves. Just be happy with Wii U power in portable form factor.

I think I'm going to take your tag to heart. Good thing we have that as a warning
 

FStubbs

Member
That's actually a good idea. Console generations last much longer than phones, let's say a new Switch releases every 4 years, but depending on specs, some newer games can carry over at lower settings or not be compatible, but that guarantees backwards compatibility with new devices forever onward until Nintendo decides to leave the Switch brand. Kinda like what Microsoft plans to do with Xbox. All Xbox games play on all Xbox systems by going an upgrade route.
People are used to that now, so it seems logical to do that with game consoles at this point.

Given PS4 and Xbox are using more or less standardized x86 hardware and Switch is using more or less standardized ARM hardware, forward compatibility is a possibility.
 

Doctre81

Member
You can't fudge late 2013's 200W of processing power into an early 2017's 10W chassis no matter how much mental gymnastics you do.

Stop doing this to yourselves. Just be happy with Wii U power in portable form factor.

tegra-X1-Gaming-.jpg
 

Yes, a TX1 dev board ran the same UE4 demo that runs on XBox One, PS4, and PCs at lower frame rate, resolution and lower fidelity. That doesn't change the tiny chassis that the new chip has to work inside with much reduced wattage.

If we are discussing whether a new TX1 based plug in home set top box console could match XBox One performance, I'm right there with you guys. Swapping out a weak X86 CPU with a powerful ARM CPU and going with nVidia GPU instead AMD GPU, I can see near parity situation easily. But that's not what Switch is. It's stuck with a portable form factor that would be lucky to handle 10W of constant heat dissipation.
 

Thraktor

Member
You can't fudge late 2013's 200W of processing power into an early 2017's 10W chassis no matter how much mental gymnastics you do.

Stop doing this to yourselves. Just be happy with Wii U power in portable form factor.

Not that I expect Switch to hit XBO or PS4 performance levels, but it would be entirely technically possible for them to do so within a 10W envelope. The GP104 Pascal GPU (which is 20 SMs or 2560 "cores") consumes 36W at 1060 MHz. By that basis, an appropriately scaled down Pascal GPU with 6 SMs (768 cores) should be able to achieve ~1GHz within 10W for 1.5TF of FP32 or 3 TF of FP16. Easily the match of PS4 or XBO provided the CPU/RAM/etc are up to it.

As I say I'm certainly not expecting that (perhaps half the performance is plausible), but there's a lot to be gained from a wide application of an energy efficient architecture on a new node with a modest clock speed.
 

Neoxon

Junior Member
Yes, a TX1 dev board ran the same UE4 demo that runs on XBox One, PS4, and PCs at lower frame rate, resolution and lower fidelity. That doesn't change the tiny chassis that the new chip has to work inside with much reduced wattage.

If we are discussing whether a new TX1 based plug in home set top box console could match XBox One performance, I'm right there with you guys. Swapping out a weak X86 CPU with a powerful ARM CPU and going with nVidia GPU instead AMD GPU, I can see near parity situation easily. But that's not what Switch is. It's stuck with a portable form factor that would be lucky to handle 10W of constant heat dissipation.
You also have to consider that Pascal is much more power-efficient than Maxwell. Also, if I recall, Emily Rogers said that the Switch is at minimum 2x the Wii U in terms of power.
 

Locuza

Member
That would give Nvidia flops a 1.7x advantage over AMD flops which sounds way too high to be honest. I'd love if someone could clarify what causes these differences if they even truly exist.

(I didn't watch all of the video but I'm pretty sure neither of the chips were overclocked, right?)
It's like you would compare cars only looking at the horsepower and ignoring all other facts, like the weight of the car, the profile of the wheels, the underlaying underground where the cars are used on etc.

You would then claim that the horsepower of car A (200) is more worth than of the car B (300) because car A is faster.
What was car A? Car A was a damn tractor and car B a Porsche.
Why was the tractor faster? Well we compared the cars on a farmland pulling some heavy machinery.

Besides the theoretical floating-point performance we always so stupidly compare, there are effective pixel- and texel-rates, there are cache and register sizes for various logic, there are trianlge and shading outputs from various fixed-function hardware with many smart and also simpel scheduling policies, bad and good load-balancing of different workload and then there is of course the raw bandwith from all the different caches and the connection to the GPU memory.

Computerbase compared the different GCN generations, where the ALUs are practically the same, which are used for the Floating-Point-Calculations.
https://translate.google.de/translate?sl=de&tl=en&js=y&prev=_t&hl=de&ie=UTF-8&u=https%3A%2F%2Fwww.computerbase.de%2F2016-08%2Famd-radeon-polaris-architektur-performance%2F2%2F%23diagramm-the-witcher-3-1920-1080&edit-text=

And what do you see? You see some examples with 20-40% better performance on GCN Gen 4 vs. Gen 1.
Not because the ALUs got so much better but because AMD improved various points which are not contributing to the floating-point calculations at all, be it theoretical or practical performance.

Then there is of course the software side too, having bad compilers and drivers will cost every hardware performance.
You can get "easily" 10-30% more performance just by having better software.
This would be also a simple mistake you could do, claiming hardware A ist better than B and not even considering if the software for hardware A is any good or pure rubbish.
 

Portugeezer

Member
Eh, I don't expect it to match XB1, but if I am optimistic my guess is that it will be close enough so that multiplatform games shouldn't be much of a problem... and we have probably 3+ years of these consoles left so that is a lot of potential games if this sells well enough.

We know this gens consoles were weaker in 2013 for their time (especially XB1) than X360 was in 2005, which will benefit the Switch (given the rapid evolution of mobile tech) as far as being close in specs to the home consoles. When it comes down to it however, almost every game can be ported in some way, the most important thing is how well the Switch sells to warrant the ports.
 
When it comes down to it however, almost every game can be ported in some way, the most important thing is how well the Switch sells to warrant the ports.

No, the most important thing is whether Nintendo is making any actual effort to court the target demographic for third-party AAA titles. There's no real reason to think that they are.
 

Doctre81

Member
Yes, a TX1 dev board ran the same UE4 demo that runs on XBox One, PS4, and PCs at lower frame rate, resolution and lower fidelity. That doesn't change the tiny chassis that the new chip has to work inside with much reduced wattage.

If we are discussing whether a new TX1 based plug in home set top box console could match XBox One performance, I'm right there with you guys. Swapping out a weak X86 CPU with a powerful ARM CPU and going with nVidia GPU instead AMD GPU, I can see near parity situation easily. But that's not what Switch is. It's stuck with a portable form factor that would be lucky to handle 10W of constant heat dissipation.

Why wouldn't it? It has active cooling and the system is at least half as big as the WiiU. And NO disc drive.
 
I wonder if any dev can request a dev kit now or if they are waiting for the full reveal in January to do that. I remember the Ori devs and the unravel devs saying that getting a dev kit was impossible.
 

Instro

Member
Not that I expect Switch to hit XBO or PS4 performance levels, but it would be entirely technically possible for them to do so within a 10W envelope. The GP104 Pascal GPU (which is 20 SMs or 2560 "cores") consumes 36W at 1060 MHz. By that basis, an appropriately scaled down Pascal GPU with 6 SMs (768 cores) should be able to achieve ~1GHz within 10W for 1.5TF of FP32 or 3 TF of FP16. Easily the match of PS4 or XBO provided the CPU/RAM/etc are up to it.

As I say I'm certainly not expecting that (perhaps half the performance is plausible), but there's a lot to be gained from a wide application of an energy efficient architecture on a new node with a modest clock speed.

Worth mentioning that the XBO S has been measured at around 50W during gameplay. Not a stretch to think that a much more energy efficient GPU/CPU in hardware with no internal HDD or disc drive could do quite well in comparison. I imagine if that they push it in docked mode the hardware will punch above it's weight pretty well with the various efficiencies and architectural improvements with a new customized Tegra GPU. Not XBO matching power of course, but good none the less.
 
Not that I expect Switch to hit XBO or PS4 performance levels, but it would be entirely technically possible for them to do so within a 10W envelope. The GP104 Pascal GPU (which is 20 SMs or 2560 "cores") consumes 36W at 1060 MHz. By that basis, an appropriately scaled down Pascal GPU with 6 SMs (768 cores) should be able to achieve ~1GHz within 10W for 1.5TF of FP32 or 3 TF of FP16. Easily the match of PS4 or XBO provided the CPU/RAM/etc are up to it.

As I say I'm certainly not expecting that (perhaps half the performance is plausible), but there's a lot to be gained from a wide application of an energy efficient architecture on a new node with a modest clock speed.

What if the SCD is planned, with a similar chipset as the one in the hybrid?
 

Durante

Member
Why wouldn't it? It has active cooling and the system is at least half as big as the WiiU. And NO disc drive.
Well, for one, at 10W sustained dissipation just for the SoC, if you assume a very low 2W for the entire rest of the system, you are looking at a 44 minute battery life with a battery capacity like 3DS XL. A bit under 2 hours with the same battery capacity as the 8" shield tablet.

Are we looking at Wii U power as the bare minimum for portable mode?
It would be very strange to go lower than that in most performance metrics. I'd say at least 2x Wii U in terms of GPU compute is a realistic lower bound.
 
Are we looking at Wii U power as the bare minimum for portable mode?
I'd think the performance in portable mode hinges entirely on how much energy Nintendo wants it to use when running on battery. Which would mean a very wide range of speculation.

I've heard that at a bare minimum the Switch would be several times as powerful as the Wii U, but I'm not sure if that's assuming running at full power or not.
 

Easy_D

never left the stone age
Didn't Epic confirm Unreal Engine 4 for Switch? We all know how Epic feels about Nintendo hardware so UE4 support should be a good sign of where it might be on the power scale.

I think that's more about the hardware supporting the featureset of the engine rather than it having X much power, with them going Nvidia there's not really any point in the thing not being able to use modern shaders in some form, so getting UE4 to run on it is probably not a big deal at all.

Well, for one, at 10W sustained dissipation just for the SoC, if you assume a very low 2W for the entire rest of the system, you are looking at a 44 minute battery life with a battery capacity like 3DS XL. A bit under 2 hours with the same battery capacity as the 8" shield tablet.

It would be very strange to go lower than that in most performance metrics. I'd say at least 2x Wii U in terms of GPU compute is a realistic lower bound.

I'm liking that this is the general sentiment tbh, because 2x Wii U in a portable is an enormous step up from the 3DS and it's not like the system wouldn't be limited by what's possible for a handheld anyway, so no use judging it as a console either
 

Oregano

Member
Well, for one, at 10W sustained dissipation just for the SoC, if you assume a very low 2W for the entire rest of the system, you are looking at a 44 minute battery life with a battery capacity like 3DS XL. A bit under 2 hours with the same battery capacity as the 8" shield tablet.

It would be very strange to go lower than that in most performance metrics. I'd say at least 2x Wii U in terms of GPU compute is a realistic lower bound.

Well the battery life is apparently very poor. ;-)

In all serious I'm expecting somewhere around the smack dab mid point of Wii U to XBO. I think that's a somewhat safe assumption considering the fact it has some form of cooling.
 

Thraktor

Member
Well, for one, at 10W sustained dissipation just for the SoC, if you assume a very low 2W for the entire rest of the system, you are looking at a 44 minute battery life with a battery capacity like 3DS XL. A bit under 2 hours with the same battery capacity as the 8" shield tablet.

I don't believe anyone claimed that the SoC would draw 10W while operating in portable mode, but there shouldn't be anything preventing an internal fan from dissipating 10W when running off mains power, given the thickness of the device.
 

blu

Wants the largest console games publisher to avoid Nintendo's platforms.
I think that's more about the hardware supporting the featureset of the engine rather than it having X much power, with them going Nvidia there's not really any point in the thing not being able to use modern shaders in some form, so getting UE4 to run on it is probably not a big deal at all.
The whole 'UE4 equates power' fiasco (and its episode 'wiiU? haha!') started with Sweeney's talk right before the start of the gen how Epic were planning for SVOGI for next-get consoles, and how that takes 2TF as a minimum. Of course we all know how that story went.
 

Zedark

Member
Well the battery life is apparently very poor. ;-)

In all serious I'm expecting somewhere around the smack dab mid point of Wii U to XBO. I think that's a somewhat safe assumption considering the fact it has some form of cooling.

If they can reach a power level that is above roughly half the Xbox One in portable mode, they should be set, right? Multiplats running at 1080p should scale down to 720p pretty much effortlessly, unless I am neglecting some other aspect. If they can supplement it with enough RAM and a good cpu it should be more than fine.
 

Schnozberry

Member
If they can reach a power level that is above roughly half the Xbox One in portable mode, they should be set, right? Multiplats running at 1080p should scale down to 720p pretty much effortlessly, unless I am neglecting some other aspect. If they can supplement it with enough RAM and a good cpu it should be more than fine.

Half the Xbox One would be 650GFlops of power. I would guess we're looking at somewhere between 350-400GFlops during portable mode.
 

Oregano

Member
If they can reach a power level that is above roughly half the Xbox One in portable mode, they should be set, right? Multiplats running at 1080p should scale down to 720p pretty much effortlessly, unless I am neglecting some other aspect. If they can supplement it with enough RAM and a good cpu it should be more than fine.

I'm by no means a tech expert but considering a whole lot if XBO games run at resolutions lower than 1080p it would still have a lot of issues. Its certainly much closer than Wii was to 360 or Vita was to PS3.
 

KingSnake

The Birthday Skeleton
Realistically, unless there is some miracle FP32/FP16 balancing for the most demanding AAA 3rd parties the native resolution will most probably be 720p in docked mode and 540p in handheld mode.
 

Oregano

Member
Realistically, unless there is some miracle FP32/FP16 balancing for the most demanding AAA 3rd parties it will be 720p in docked mode and 540p in handheld mode.

From what we know it should be quite a bit above Wii U even in handheld mode so it should be able to hit 720p easily. Plus Nintendo probably would have gone with a 540p screen if that was the case.

EDIT: oh wait AAA third party games you're probably right. Though I'd expect scene complexity to take more a hit.
 

KingSnake

The Birthday Skeleton
From what we know it should be quite a bit above Wii U even in handheld mode so it should be able to hit 720p easily. Plus Nintendo probably would have gone with a 540p screen if that was the case.

EDIT: oh wait AAA third party games you're probably right. Though I'd expect scene complexity to take more a hit.

Yeah, I was talking about 3rd parties AAA. 1st party games I expect to run at 1080p docked and 720p in handheld mode.
 
I wonder if the device will have a gsync LCD, and use adaptive vsync when docked. That way, handheld mode will always feel smooth and lower input lag in both modes.

Edit: Adaptive Vsync is what I meant. Vsync turns on at 60fps or higher, off on 59 and lower.
 

TheWarlord

Neo Member
Hey guys, I wanted to make a new thread, but it seems that I don't have enough privileges. I'm not very active, I mostly lurk. So I thought I could post this here.

I'm from Romania, and our biggest electronics/IT & gaming retailer, Media Galaxy, just put up the Switch pre-order page. It already has a price attatched, but it could be just a placeholder (the date is also listed as a "default" 31st of March 2017). The price is 1349 Lei/RON, which is roughly equivalent to 299 euros, which usually means that it would be listed at $299 in the US.

I'm not really sure if there are shops with prices attached to the Switch in other countries, so I thought it's an interesting bit of information. 299 is not out of the ordinary considering Nintendo is basically selling a 6" tablet with a dock and some physical controllers. The price should be around that mark anyway..

Here is the link:

http://mediagalaxy.ro/consola-nintendo-switch
 

MDave

Member
Realistically, unless there is some miracle FP32/FP16 balancing for the most demanding AAA 3rd parties the native resolution will most probably be 720p in docked mode and 540p in handheld mode.

In regards to FP16/FP32 : I did some digging and found a few posts on beyond3d forums about it:

Picked the most relevant quotes:

Sometimes it requires more work to get lower precision calculations to work (with zero image quality degradation), but so far I haven't encountered big problems in fitting my pixel shader code to FP16 (including lighting code). Console developers have a lot of FP16 pixel shader experience because of PS3. Basically all PS3 pixel shader code was running on FP16.

It is still is very important to pack the data in memory as tightly as possible as there is never enough bandwidth to lose. For example 16 bit (model space) vertex coordinates are still commonly used, the material textures are still dxt compressed (barely 8 bit quality) and the new HDR texture formats (BC6H) commonly used in cube maps have significantly less precision than a 16 bit float. All of these can be processed by 16 bit ALUs in pixel shader with no major issues. The end result will still be eventually stored to 8 bit per channel back buffer and displayed.

One example where 16 bit float processing is not enough: Exponential variance shadow mapping (EVSM) needs both 32 bit storage (32 bit float textures + 32 bit float filtering) and 32 bit float ALU processing.

FP16 is more than enough for post processing (DOF, bloom, motion blur, color correction, tone mapping). As FP16 makes post processing math 2x faster on Rogue (all the new iDevices), it will actually be a big thing towards enabling console quality graphics on mobile devices. Obviously FP16 is not enough alone, we also need to solve the bandwidth problem of post processing on mobiles. On chip solutions (like extending the tiling to support new things) would likely be the most power efficient answers.

They will, most likely, use FP32 for some pixel shader code. But the parts that need FP32 aren't actually that frequent. Some developers will want to play it safe and use FP32 wherever there is any doubt, but even then there are shader parts that are obviously fine with FP16.

I don't think it is particularly challenging to write pixel shader code for a "console quality", "high fidelity" mobile game that is something like 70% FP16 and 30% FP32.

I think this sounds really good for potentially squeezing as much performance as possible via the use of FP16 as Nvidia have obviously made it a focus in their Maxwell and Pascal architectures. Not to mention the tools and support they will have for Nintendo and other developers.
 

Donnie

Member
Half the Xbox One would be 650GFlops of power. I would guess we're looking at somewhere between 350-400GFlops during portable mode.

Conservatively 400gflops is reasonable. Could easily be more but I suppose if they're going for true 1080p in docked mode then 400gflops at 720p would require around 900gflops docked to run the same.
 

Donnie

Member
I'm by no means a tech expert but considering a whole lot if XBO games run at resolutions lower than 1080p it would still have a lot of issues. Its certainly much closer than Wii was to 360 or Vita was to PS3.

Well most Xbox One games are 900p or above. So lets say we're porting a 900p game and making the assumption that the 1.3tflop max is actually the reason that game isn't hitting a better resolution on Xbox One (which it might not be).

To run that game at 720p with the same effects and frame rate you're going to need just over 800gflops. Though that ignores the newer Tegra architecture and running some shaders in fp16 which could partially bridge the gap (lowering the theoretical number required). Also as I mentioned the theoretical processing performance of Xbox One's GPU may not be the reason it can't run that particular game above 900p anyway.

So basically its near impossible to answer this at the moment.
 
Top Bottom