• Hey Guest. Check out your NeoGAF Wrapped 2025 results here!

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

Status
Not open for further replies.
It's actually a PS4 or PS3 port (they're likely basically the same), there is not even a Vita DQH 1 version. Regardless, it uses Omega Force's musou engine, and DQ11 PS4 is being made using Unreal Engine 4.

Given how FE Heroes looks so far, it's clear the Musou engine runs pretty ok.
 
As expected.

On par at best, up to ~60% behind in optimized multi-threaded workloads.

The CPU is the most competitive aspect of the Switch hardware.
That's good. Im looking forward to 2k18. Which runs at 1080/60 on Xbox/ps4.

Could the switch keep that performance if the resolution is lowered to 720P, and some of the graphics settings are turned down?

I'd be amazing if I could get 720/60 Though I'd be fine with 720/30 locked.
 
Gpu comparison
Switch flops handheld: 0.2tflops dockmode: 0.39tflops
PS4 pro 4.2tflops ( 21x switch handheld 10.76x switch docked)
PS4 flops 1.84tflops ( 9.2x switch handheld 4.17x switch docked)
Xbox one 1.31tflops ( 6.6x switch handheld 3.36x switch docked)


Bandwidth Switch 25.6 GB/s
PS4 pro 218GB/s 8.5x nintendo switch
Ps4 176 GB/s 6.87x nintendo switch

xbox one 102GB/s 4x ninntedo switch

Those three numbers aren't actually comparable, as PS4 is a standard full scene renderer vs Switch's tile based rendering. Also XBox One's main memory bandwidth is 68GB/s not 104GB/s

Flops comparison may be correct in fp32 (haven't worked it out) though fp16 usage can significantly narrow that difference down for Switch games.
 
Those three numbers aren't actually comparable, as PS4 is a standard full scene renderer vs Switch's tile based rendering. Also XBox One's main memory bandwidth is 68GB/s not 104GB/s

Flops comparison may be correct in fp32 (haven't worked it out) though fp16 usage can significantly narrow that difference down for Switch games.

I think it remains to be seen whether fp16 actually does anything or not. It's certainly not a "press button for double performance" situation. People might want to avoid getting ahead of themselves with their expectations of it. Also, the Pro has the capability as well doesn't it?
 
That's good. Im looking forward to 2k18. Which runs at 1080/60 on Xbox/ps4.

Could the switch keep that performance if the resolution is lowered to 720P, and some of the graphics settings are turned down?

I'd be amazing if I could get 720/60 Though I'd be fine with 720/30 locked.
CPU limits are largely independent of resolution. Can't say if 60fps are realistic. Might be. If other consoles can do it, remain hopeful.

Is that NBA 2k18 or ...?
 
CPU limits are largely independent of resolution. Can't say if 60fps are realistic. Might be. If other consoles can do it, remain hopeful.

Is that NBA 2k18 or ...?
Yeah, the bball game. I wouldn't oppose removing the fans from the stands, or replacing them with paper cutouts. It's not like I'm paying attention to them when I'm playing. :P
 
Those three numbers aren't actually comparable, as PS4 is a standard full scene renderer vs Switch's tile based rendering. Also XBox One's main memory bandwidth is 68GB/s not 104GB/s

Flops comparison may be correct in fp32 (haven't worked it out) though fp16 usage can significantly narrow that difference down for Switch games.

It's true that the numbers aren't directly comparable, but any way you slice it, Switch is an order of magnitude less powerful than PS4, and even Xbox One.
 
I question whether the average consumer can even recognize that Zelda BOTW is a step down from PS4/X1 in the graphics department.

Obviously more power is always better, but I think the Switch has enough power to do what Nintendo wants it to do and the average consumer doesn't realize it's well below PS4/X1.

Yes, they would see the difference. Zelda BotW is really great, but there is a big visible difference between what it can push (image quality wise for one) and what Xbox One and PS4 can push.
 
I think it remains to be seen whether fp16 actually does anything or not. It's certainly not a "press button for double performance" situation. People might want to avoid getting ahead of themselves with their expectations of it. Also, the Pro has the capability as well doesn't it?

Its not some kind of theoretical feature that may or may not be able to do anything.. Its just using 16bit precision instead of 32bit where 32bit precision isn't necessary. With hardware that supports fp16 (like Switch and unlike PS4/XBox One) that then allows you to perform two 16bit operation at the same speed as one 32bit operation. Nowhere did I say anything about being able to just turn it on or doubling performance either. I said usage of fp16 precision could narrow the difference in flops between Switch and XBox One/PS4, and it absolutely can.

The only question is how many 32bit ops can safely be expressed in 16bit precision. We've seen developers claim 70%, but of course it will depend on the game. Some games may not be able, or the developer not willing, to use it for significant performance improvements. While others no doubt will. Its something AMD has added to both Scopio and PS4 Pro, so its certainly something developers will use.
 
What do you mean if it does anything? Its not some kind of theoretical feature.. Its just using 16bit precision instead of 32bit where 32bit precision isn't neccesary. With hardware that supports fp16 (like Switch and unlike PS4/XBox One) that then allows you to perform two 16bit operation at the same speed as one 32bit operation.

Its obviously not a question of wether that does anything, of course it does, the only question is how many 32bit ops can safely be expressed in 16bit precision. We've seen developers claim 70%, but of course it will depend on the game. SOme games may not be able, or the developer not willing, to use it for significant performance improvements. While others no doubt will. Its something AMD has added to both Scopio and PS4 Pro, so its certainly something developers will use.

"If it does anything" meant if it actually had an appreciable effect on game performance. That depends on to what extent it can be implemented in a given game, which is uncertain, and then on top of that whether anyone even does implement it to any large degree, which is also uncertain. Those are completely open questions at this point. There are plenty of technologies that could have theoretical benefits but aren't ever realized for a lot of reasons. It is silly to think fp16 is a panacea that will close a very large hardware gap.
 
The Switch doesn't need to compete horse-power-wise. Great art-direction = Great graphics. This has been true since the dawn of art and will continue to be true for the rest of eternity. Having a lot of horsepower is great of cource, but nintendo is a mature company that has enough talent to know how to prioritize. And though you shouldnt always trust the "critics-consensus" just take a look at gamerankings.com


This is kind of a BS argument. Nintendo could always use more power to making more stunning games. People said the same thing about the Wii U and yet Nintendo had to sacrifice a lot of performance to bring their vision of the game to Wii U.

If horse power is no big deal, Nintendo should just stay with Wii level graphics.

Horsepower also means computations. These are things that affect (or limit) game design. (See: Nemesis System).

It's not just about graphics.
 
Those three numbers aren't actually comparable, as PS4 is a standard full scene renderer vs Switch's tile based rendering. Also XBox One's main memory bandwidth is 68GB/s not 104GB/s

Flops comparison may be correct in fp32 (haven't worked it out) though fp16 usage can significantly narrow that difference down for Switch games.

Yes xbox one have sram with 104gb/s and standard memory 68gb/s but all programmers benefit from sram memory. Fp16 has also ps4pro but don't know if anybody use this mode. Tile based renderer helps little but still tegra x1 bandwidth is huge bottleneck for 1080p docked mode. And to be honest tegra x1 is much closer to wii u and ps3/xbox 360 than xbox one.
 
It's true that the numbers aren't directly comparable, but any way you slice it, Switch is an order of magnitude less powerful than PS4, and even Xbox One.

Well no its not as it goes, even the inaccurate comparison I quoted doesn't show any single area where its a order of magnitude less powerful. But either way I'm not interested in arguing how powerful it is overall in comparison because there are two many variables. Was only making sure people don't take that comparison literally. as its really apples to oranges in some way.
 
Yes xbox one have sram with 104gb/s and standard memory 68gb/s but all programmers benefit from sram memory. Fp16 has also ps4pro but don't know if anybody use this mode. Tile based renderer helps little but still tegra x1 bandwidth is huge bottleneck for 1080p docked mode. And to be honest tegra x1 is much closer to wii u and ps3/xbox 360 than xbox one.

All developers benefit from tile based rendering, the effect on bandwidth saving is very similar to using on chip SRAM as XBox One does. Either way you can't make comparisons of different parts of a system that don't directly relate to each other no matter what you think. For example Switch's main memory does not have the same purpose as XBox One's on chip SRAM, so comparing the bandwidth of the two is just wrong. If you're going to make comparisons then do so apples to apples.

BTW yes PS4 Pro also supports fp16, is Switch really competing with PS4 Pro though? The fact that PS4 Pro and Scorpio both support FP16 will likely be a plus for Switch, as more developers will be looking into working with that precision.
 
The Switch only has 26 GB of space available. Even if you buy your games retail and only buy digital exclusives on the Eshop between patches, DLC and Eshop exclusive digital games you're going to burn through that 26 GB very quickly.

Not really. Not at launch. Most of the games are under a gig and all of the patches have been under a gig.
 
Dunno about 2K, but the Fifa switch version will be be PS3/360 one.

We don't know this so stop saying it as fact.

Earlier you yourself said we can't infer anything from 5 seconds of FE Warriors footage. We don't have any FIFA 18 footage. What was shown in the Japanese Switch ad is FIFA 17. There is no FIFA 18 footage anywhere, for any system.
 
We don't know this so stop saying it as fact.

Earlier you yourself said we can't infer anything from 5 seconds of FE Warriors footage. We don't have any FIFA 18 footage. What was shown in the Japanese Switch ad is FIFA 17. There is no FIFA 18 footage anywhere, for any system.


That sounds painfully obvious it'll be the PS360 version. Come on, it's EA lol. EA on Switch. Not only the console may not cut it out ot get the Frostbite Engine version of FIFA, but even if it did, it'd be easier for EA to just port the PS360 version.
 
BTW yes PS4 Pro also supports fp16, but so what?, is Switch really competing with PS4 Pro?..

Yes, Switch at 300 dollars is competing for people's money with the Pro at 400 dollars. People's money and time for gaming is limited and they purchase based on what provides the most value in their opinion. I thought that would be pretty obvious.
 
All developers benefit from tile based rendering, the effect on bandwidth saving is very similar to using on chip SRAM. Either way you can make comparisons of different parts of a system no matter what you think. If you're going to make comparisons then do so apples to apples.

BTW yes PS4 Pro also supports fp16, but so what?, is Switch really competing with PS4 Pro?..

So you say that tile rendering reduce bandwidth usage several times? :D Maxwel used it and Kepler doesn't, bandwidth compression of Maxwell was around 20% better than Kepler co 1.2x 25.6 gb/s = 30.7gb/s still very poor.
 
Once footage of the games started to come out it became really evident that the Switch isn't a graphical powerhouse compared to other consoles. I don't understand how people expected better hardware to be powering those games.
 
Yes, Switch at 300 dollars is competing for people's money with the Pro at 400 dollars. People's money and time for gaming is limited and they purchase based on what provides the most value in their opinion. I thought that would be pretty obvious.

So what is your point here? Switch has fp16 functionality but its not worthy of note in a comparison between Switch, XBox One, PS4 and PS4 Pro merely because PS4 Pro alone also supports it?
 
That sounds painfully obvious it'll be the PS360 version. Come on, it's EA lol. EA on Switch. Not only the console may not cut it out ot get the Frostbite Engine version of FIFA, but even if it did, it'd be easier for EA to just port the PS360 version.

Sure, but we shouldn't go around stating things as fact when we simply do not know.

Also who knows, the fact they call it a custom version may mean it's something else entirely.
 
So what is your point here? Switch has fp16 functionality but its not worthy of note in a comparison between Switch, XBox One, PS4 and PS4 Pro merely because PS4 Pro alone also supports it?

The point is that f16 is for now as magical as cloud compute power for xbox one ;)
 
So what is your point here? Switch has fp16 functionality but its not worthy of note in a comparison between Switch, XBox One, PS4 and PS4 Pro merely because PS4 Pro alone also supports it?

I don't particularly think fp16 functionality is worthy of note on either the Switch or the Pro at this point, it has to be shown that it makes a meaningful difference.

But if you're going to bring it up, you can't also handwave away the Pro as somehow "not competing" with another gaming device that sells at a similar price.
 
So you say that tile rendering reduce bandwidth usage several times? :D Maxwel used it and Kepler doesn't, bandwidth compression of Maxwell was around 20% better than Kepler co 1.2x 25.6 gb/s = 30.7gb/s still very poor.

What are you talking about? Its nothing to do with compression. Go and google the differing methods of rendering a frame. I really don't feel like explaining what tile based rendering is again when its easily found on the internet through a search engine. But to very quickly give you a hint here.

The purpose of a large pool of embedded memory like XBox One's SRAM is to keep the bandwidth required to render a frame inside the chip and away from main memory. Hence is reduces the bandwidth hit on main memory.

The purpose of tile based rendering is to keep the bandwidth required to render a frame inside the chip and away from main memory. Hence is reduces the bandwidth hit on main memory.
 
What are you talking about? Its nothing to do with compression. Go and google the differing methods of rendering a frame. I really don't feel like explaining what tile based rendering is again when its easily found on the internet through a search engine. But to very quickly give you a hint here.

The purpose of a large pool of embedded memory like XBox One's SRAM is to keep the bandwidth required to render a frame inside the chip and away from main memory. Hence is reduces the bandwidth hit on main memory.

The purpose of tile based rendering is to keep the bandwidth required to render a frame inside the chip and away from main memory. Hence is reduces the bandwidth hit on main memory.

Yes, I saw how magically it reduced bandwidth usage on Zelda in 900p where docked mode is slower than handheld 720p ;) Give me a break
 
I don't particularly think fp16 functionality is worthy of note on either the Switch or the Pro at this point, it has to be shown that it makes a meaningful difference.

But if you're going to bring it up, you can't also handwave away the Pro as somehow "not competing" with another gaming device that sells at a similar price.

It can certainly make a difference. Its simply using lower precision for ops that don't need 32bit. The only question is the magnitude of the difference. I agree it can't be quantified in a comparison, and I didn't say it should be. I just mention that while the flops comparison is accurate its worth noting, for whoever is reading that comparison, that Switch can support a doubling in flops performance for some effects when supported by developers. This is something already fully supported by UE4 BTW.
 
Wow you really don't understand the stuff you're posting, seriously stop it.

Give me one proof that developers use fp16 in games. Or even so still tegra x1 is weak old hardware and nintendo lost chance too have some 3rd party games on handheld because of not using parker version (double bandwidth, 16nm, better architecture).
 
I've explained everything you need to know. You aren't listening and frankly getting more and more nonsensical (a secret sauce mythical feature that's fully supported by Unreal Engine 4, I mean really come back to earth...) I'm not wasting my time anymore, you're on ignore.
 
Give me one proof that developers use fp16 in games. Or even so still tegra x1 is weak old hardware and nintendo lost chance too have some 3rd party games on handheld because of not using parker version (double bandwidth, 16nm, better architecture).
First TX2 is more expensive than TX1 so thats means more expensive Switch, The only real advantage that the Pascal Tegra have over the Maxwell is power consumption and bandwidht (so more expensive DDR modules too) also the Pascal architecture is the same as Maxwell but shrinked and looks like both have "almost" the same performance becuase the Cuda Cores count... Also the more important thing... Not aviable at the moment... Did you really think that is a good idea that Nintendo release the Switch in december at $350? Basically the TX1 is the better option right now... Is a fact.
 
First TX2 is more expensive than TX1 so thats means more expensive Switch, The only real advantage that the Pascal Tegra have over the Maxwell is power consumption and bandwidht (so more expensive DDR modules too) also the Pascal architecture is the same as Maxwell but shrinked and looks like both have "almost" the same performance becuase the Cuda Cores count... Also the more important thing... Not aviable at the moment... Did you really think that is a good idea that Nintendo release the Switch in december at $350? Basically the TX1 is the better option right now... Is a fact.

How its not available as Nvidia sells it in jetson ? :)
 
It doesn't run at 20fps in many areas, it has occasional drops to 20fps... And I, for one, Im happy to get all Nintendo games on a hanheld.

It is pretty bad in the forest, and a lot of other areas especially on TV, but I can overlook zeldas frame rate shortcomings for the enjoyment and freedom the game gives.
 
Yes but how this translates to usage in console games?
Texture coordinates can be half precision, useful for example when rendering many millions of particles on the GPU. It prevents a lot of expensive cache misses. A significant part of the lighting engine can be done in half precision, colors, textures, AI can benefit from half-precision, the audio engine.
 
Status
Not open for further replies.
Top Bottom