Nintendo Switch Tech Speculation discussion

[Entropy]

That very same tweet says, "2GHz Maximum CPU, 1 GHz Maximum GPU". That's the maximum the chip can do, the specs nVidia provided Nintendo no doubt because that's what the chips can do if you provide enough cooling. The fact they say maximum shows they're not the operating clocks - contrast that with Sony and MS leaks where we're given clockspeeds as target clockspeeds.

On the flip side, where do you think DF's numbers are coming from? If Switch is clocked that much faster, where the heck is 300 MHz coming from?

Only they know where their numbers came from. The clock speeds are very precise but their wording show both confidence and ambiguity.
Once hardware is available we'll know more obviously. Perhaps even in three weeks. But from a pure tech point of view, it would be more interesting if we are dealing with a bespoke SoC, regardless if Nintendo would have chosen to adjust its hardware resources up or down from the X1.

 

[MrFox]

The first leak was only that the dev kit is using an X1.

Someone pasted the X1 specs as the rumor. Hence the weird maximum clock specs.

DF recently got more info to add, like the actual clocks.

 

[Eteric]

How do you guys think the Switch will compare to the GPD Win? That little thing can actually run most modern games (you have to fiddle with stuff like Doom or Witcher 3, but it works) while running Windows 10.

Honestly, looking at stuff like LowSpecGamer, it might not be as hard to make Switch ports as we're thinking it'll be. There will be a lot of concessions of course, but we all knew that from the beginning. It'll especially be easy if the game uses Vulkan.

 

[Deleted member 13524]

That very same tweet says, "2GHz Maximum CPU, 1 GHz Maximum GPU". That's the maximum the chip can do, the specs nVidia provided Nintendo no doubt because that's what the chips can do if you provide enough cooling.

If the clocks in the twitter leak were so far off from the devkits' own clocks, then why did Eurogamer say they heard developers saying those specs were "uncannily similar" to the specs they had in their own devkits?
And then they say "oh but the clocks are much lower in the retail version".
So not that much uncannily similar after all.

On the flip side, where do you think DF's numbers are coming from? If Switch is clocked that much faster, where the heck is 300 MHz coming from?

Oh I'm willing to believe the clock speeds are correct. It's just that the rest doesn't seem compatible with said clocks.
If it's just a TX1 then it doesn't make sense that docked clocks for GPU are that much lower than Shield TV. The Shield TV has a tiny low-rpm fan with a tiny heatsink and the CPU clocks are almost twice as fast, so why would the docked Switch need to scale the GPU clocks back ~23% compared to the Shield TV? There's obviously a higher thermal headroom and power consumption while docked shouldn't be a concern (we're obviously talking sub-25W here).
And then Emily Rogers (who has been rock solid so far, with the Switch at least) claimed the devkits' fans were annoyingly loud. If it's a TX1 with CPU and GPU clocked much lower than Shield TV's, why would the devkits be loud?


To me, this means either:

1) The Switch SoC is made on an older and less efficient process than TX1's 20nm. For example, a density-optimized 28nm process to make the chip as small and as cheap as possible. First devkits were already using engineering samples of said chip that were pushing more power than the final version.
2) The GPU in the Switch is bigger than TX1's and it's made on the same 20nm process (or a more advanced one).



And then there's the news of yesterday: Unreal Engine 4's latest reference guide form has a preset for docked and mobile Switch (SwitchConsole and SwitchHandheld respectively, in the previous versions known as WolfSea and WolfAir).
The SwitchConsole preset has the quality set to 2 (meaning high). The SwitchHandheld preset has the resolution set to 0.666 (1080 * 0.666 = 720) and the quality is set to 1 (medium).
This makes sense if you look at the clocks between Handheld and Docked mode.

Here's the thing though: the Xbox One quality preset is extremely similar to SwitchConsole. How could this be?
This guy made a video using the UE4's Elemental Demo with SwitchConsole's quality presets at 900p:

He's getting rather crappy framerates, and he's using a Core i7 3930k 6-core 3.2Ghz SandyBridge and a GTX 650 Ti, which has 4 Kepler SMX at 928MHz, a ~1.4 TFLOPs GPU.
How on earth would a 2 SM Maxwell GPU at 768MHz (< 400 GFLOPs FP32, <800 GFLOPs FP16) be able to run something like that?

 

[BRiT]

I still don't think there's a nvidia SoC in the final console. AFAIK, there's absolutely nothing on nvidia's official statements, financials, roadmaps etc. suggesting such a high-profile design win for a semi-custom SoC and Parker seems to be an afterthought to anything but automotive.
I definitely believe there might be a Tegra X1 in the handheld's development kits (perhaps even just a Jetson TX1 plus their own PCB with I/Os), but I think the final SoC will either be made by Nintendo with a DMP GPU or made by AMD.

@ToTTenTranz just how long are you going to hold onto old beliefs in the face of evidence? Even when evidence of using Nvidia was around you still refused to believe it. Now there is evidence of real clock speeds and you still refuse to believe it. It's time for you to take a deep breath and embrace the reality.

 

[Eteric]

I think there's something we don't know, honestly. Nothing really makes sense tbh.

 

[Shifty Geezer]

If the clocks in the twitter leak were so far off from the devkits' own clocks, then why did Eurogamer say they heard developers saying those specs were "uncannily similar" to the specs they had in their own devkits?

Same numbers (because they were silicon peak).

Oh I'm willing to believe the clock speeds are correct. It's just that the rest doesn't seem compatible with said clocks.

So you think it's much bigger silicon with more shader cores?

If it's just a TX1 then it doesn't make sense that docked clocks for GPU are that much lower than Shield TV. The Shield TV has a tiny low-rpm fan with a tiny heatsink and the CPU clocks are almost twice as fast, so why would the docked Switch need to scale the GPU clocks back ~23% compared to the Shield TV? There's obviously a higher thermal headroom and power consumption while docked shouldn't be a concern (we're obviously talking sub-25W here).

That's one for Nintendo to answer.

And then Emily Rogers (who has been rock solid so far, with the Switch at least) claimed the devkits' fans were annoyingly loud. If it's a TX1 with CPU and GPU clocked much lower than Shield TV's, why would the devkits be loud?

Maybe Nintendo were running faster clocks before settling on these new ones?

He's getting rather crappy framerates, and he's using a Core i7 3930k 6-core 3.2Ghz SandyBridge and a GTX 650 Ti, which has 4 Kepler SMX at 928MHz, a ~1.4 TFLOPs GPU.
How on earth would a 2 SM Maxwell GPU at 768MHz (< 400 GFLOPs FP32, <800 GFLOPs FP16) be able to run something like that?

The logic here clearly points to the UE profile being wrong. If that spec PC can't run it at a decent framerate, there's no way a portable is! Again, how do you get XB1 performance from a skinny tablet running a mobile SOC on a handful of watts?

For me, I'd rather take the evidence of a trusted insider source than an unexplained .ini file that contradicts the simpler, straight-forward explanation. Not that I'd say anyone who disagrees with the TX1 @ 700 MHz theory
should be burnt at the stake, because the specs don't matter at this juncture other than a pastime puzzle. But if pressed to make a call to hang my reputation on (which I'm not), I'd be backing DF here rather than the ToTTenTranz horse.

 

[BRiT]

@Eteric what to you doesn't make sense? It makes sense when you take into account Battery Life, Thermal Management, Longevity, and Price.

 

[AlNom]

Again, how do you get XB1 performance from a skinny tablet running a mobile SOC on a handful of watts?

"A handful of shaders on a leaky chip are going to save Nintendo." :|

"EXACTLY!":sly:

 

[Shifty Geezer]

I think there's something we don't know, honestly. Nothing really makes sense tbh.

Why doesn't a TX1 clocked at the rumoured clockspeeds make sense? It fits the needs for a tablet and scales for TV. Perhaps the reason it's running slower than a Shield TV is because Nintendo want to cap the games at mobile quality and thought to just scale the GPU to change from 720p to 1080p? Or maybe Shield TV's solution doesn't work with a tablet? Or maybe Nintendo are just bonkers cautious in wanting to make a device that'll work for 50 years?? And the UE4 specs are premature or somehow unrepresentative. From where I'm sitting it all first as nicely as every other Nintendo hardware investigation we do, where every single time it's been the lowest possible situation that has come true. All those questions about 'is it really this hardware when it doesn't fit the evidence 100%' were answered with an emphatic 'yes' eventually. We hoped Nintendo were going to change their spots, and at least Switch looks a bit cool which is something, but otherwise they're the same animal operating according to the same DNA. One really has to reach a bit to see a scenario in which these pieces don't comfortably align 80%.

 

[Deleted member 13524]

I don't even want to share what I have in mind. This forum has become horrible for people making guesses that happen to come out wrong. I still get thrown in the face stuff I predicted wrong from years ago, some people here are [censored] eager to use strawmans at the first opportunity.
(...)
(BTW what the hell do these people do in their lives that's so boring that they get off on searching/keeping/whatever years-old quotes from unimportant anonymous users in internet forums?!)

[QUOTE=ToTTenTranz]
I still don't think there's a nvidia SoC in the final console. AFAIK, there's absolutely nothing on nvidia's official statements, financials, roadmaps etc. suggesting such a high-profile design win for a semi-custom SoC and Parker seems to be an afterthought to anything but automotive.
I definitely believe there might be a Tegra X1 in the handheld's development kits (perhaps even just a Jetson TX1 plus their own PCB with I/Os), but I think the final SoC will either be made by Nintendo with a DMP GPU or made by AMD.
[/QUOTE]
@ToTTenTranz just how long are you going to hold onto old beliefs in the face of evidence? Even when evidence of using Nvidia was around you still refused to believe it. Now there is evidence of real clock speeds and you still refuse to believe it. It's time for you to take a deep breath and embrace the reality.

Theeere we go. Here it is.
Out of curiosity, how long did it take to find that specific 4 month-old post on page 116 of a closed thread?
Was it worth it, at least? Is there that much satisfaction from engaging in strawmans? You were wrong before so you must be wrong again! Haha you're discredited!!
I don't mind having been wrong about the SoC supplier. I was the first user to bring the nvidia news to the forum, it's the very first post in this thread, lol.

Why not engage in my arguments instead of pulling strawmans? How does that benefit discussion? Bonus internet points for you, I guess?
I mean you already have your own thread where you made your own rules of prohibiting people from speculating on speculation. What more do you want?

Same numbers (because they were silicon peak).

If there is such a thing (nominal clocks can be surpassed through multiplier configuration and higher voltage and/or current), I'm positive TX1's silicon peak wouldn't be at 2GHz CPU and 1GHz GPU.
Get access to nvidia's custom kernel and a competitive overclocker working on it and I'm positive that by feeding up to 50W into TX1 while attaching a CPU cooler to it then both the GPU and CPU could be clocked significantly higher than what the Shield TV can do with 22W (pulled from the wall BTW).

Which, in the end, means that those clocks wouldn't be silicon peak as you say. At best, they would be the maximum granted by the specific devkit that was handed to the developers.
But then why would the devkit even allow 1GHz GPU and 2GHz CPU if the maximum target clocks were so much lower?

So you think it's much bigger silicon with more shader cores?

The "much" part is your own deduction. It could even have substantially more shader cores and be smaller.
1 - Cortex A53 cluster stripped out
2 - Video encoder can be much simpler/smaller as no one is going to use the Switch to record 4K HEVC videos
3 - Same goes for image processor, it doesn't need to handle JPEGs at 600MPixel/s because the Switch won't be used to take e.g. 40MPixel pictures at 15 frames-per-second.
4 - PCI-Express interface and some other I/O are useless and can be stripped out too.
5 - if Cortex A57 cluster is running at low clocks then they can be printed for higher density / smaller footprint.

Take all of these into account and maybe it fits an extra SM with the left over space.
Plus, if this is a custom chip, there's no reason to assume the Streaming Multiprocessors as a fixed and known value.
It could be each SM with more ALUs like Kepler's SMX as a closed and hardware-specific development environment could allow for higher utilization from the get go.
Pascal consumer chips for example have no 2*FP16 throughput, but they do have dedicated FP16 ALUs. We could be seeing one or more smaller SM units with FP16 ALUs only.

These are just ramblings and each of these tweaks has a very low chance of being true. I'm aware of that.
The only message I want to convey is there's a lot of missing pieces besides the clocks. SM count and SM configuration are just an example of missing info.

The logic here clearly points to the UE profile being wrong.

You could assume the Unreal Engine 4 developers made the Switch performance profile wrong, or you could just look at it as yet another piece of the puzzle that doesn't fit.

If that spec PC can't run it at a decent framerate, there's no way a portable is!

It doesn't seem to be that far away from 30 FPS. Console implementations punch above PCs with similar theoretical specs. The 650 Ti is a 4 year-old GPU and the Maxwell already boosted performance-per-GFLOP by a lot.
According to nvidia, a Maxwell 1 SM has 90% of the performance of a Kepler SMX. So 4 Maxwell SMM at 1GHz (512 ALUs, 1TFLOP) would have 90% the real performance of 4 Kepler SMX at 1GHz (768 ALUs, 1.5 TFLOPs).

I'd be backing DF here rather than the ToTTenTranz horse.

Lol I'm practically on the same horse as DF, which is there's a lot of missing info.

 

[BRiT]

@ToTTenTranz
It didn't take me any time at all, it was shared with me from others. I'm just trying to figure out why you're so dead set against reality when it comes to the Nintendo Switch. You've been fighting it every step of the way. I don't see what you have to gain from fighting so hard against the inevitable.

 

[Shifty Geezer]

I don't see the need to 'give in' to a theory. ToT's position is simply that the confidence factor in the evidence isn't high enough to satisfy his threshold, and he's willing to question the unknowns in search of evidence. Given that the whole purpose of this thread and other speculation threads is the joy of the debate and not to prove anyone right or wrong, I see nowt wrong with that. For those who have made their minds up and don't want to argue the minutiae, they can enjoy Christmas with the peace of mind that comes from knowing Nintendo will be providing another hardware turkey. For the others, they must pace restlessly in the uncertainty that maybe, just maybe, Switch isn't a vanilla TX1 underclocked and coupled with a dock as boring to the point of being little more than a USB charging station. There might yet be something exciting about Switch's hardware.

Either way it doesn't matter if anyone's right or wrong, so long as the arguments are reasonable and logically debatable.

 

[Goodtwin]

The logic here clearly points to the UE profile being wrong.

Based on what? What UE4 games have you seen running on Switch? I personally think both @BRiT and @ToTTenTranz are making valid arguments, but this is definitely a topic left to open discussion rather than dealing in absolutes. We do not have complete specs, we have a couple of clock speeds. We have Dark Souls 3 running on Switch. We have Unreal 4 development profiles would be what Epic deems is a good place to start. These profiles aren't outdated, they were previously named Wolf to disguise the profiles, and more recently changed them to Switch profiles. These seem unlikely when your talking about a standard TX1 at 760Mhz console and 300Mhz portable. People are making assumptions, and in a few weeks things will look a lot clearer, either from actually specs going public from Nintendo or Nividia, or from software demonstrations. Who knows, maybe a TX1 can run modern games on low to medium settings just fine. We do not have a clear picture yet, and I think that's what @ToTTenTranz is really arguing. I guess the real question is why is it so important to pass judgment prior to reveal. For me I actually do think the specs are a TX1 clocked as rumored, and I will get a kick out of people finding out that such a modest processor can infact run most modern games at reduced settings. I could be wrong, but that's what I speculate. Games aren't that much different from the 360/PS3 era, and resources seem to get eaten up very quickly for a modest gain in fidelity. No matter what, Mario will be once sexy bitch on Switch.

 

[Entropy]

Come on, of course Tottentranz will be burned on a stake here if the innards of the Switch is a TX1 at the clocks eurogamer reported.
He won't even get the benefit of getting tossed in a lake to see if he floats first.

 

[Shifty Geezer]

Based on what?

The fact that they are the specs that require a 50 W SOC to achieve. And by 'wrong', I just mean they aren't presenting a black-and-white indication that Switch must be as powerful as other platforms using the same settings. That is, if those settings are there for Switch, we can expect Switch to run the same game with the same settings as any other platform with the same profile. For example, Switch might still need massive downgrades in quality for the same title and it's just UE4's in-built adaptation that we're seeing here. A game created and balanced to run at the mobile profile will scale up with zero dev effort to docked quality by using the higher quality settings relative to the mobile assets.

either from actually specs going public from Nintendo or Nividia.

Now that's wishful thinking!

 

[psurge]

Only they know where their numbers came from. The clock speeds are very precise but their wording show both confidence and ambiguity.
Once hardware is available we'll know more obviously. Perhaps even in three weeks. But from a pure tech point of view, it would be more interesting if we are dealing with a bespoke SoC, regardless if Nintendo would have chosen to adjust its hardware resources up or down from the X1.

One thing I've been wondering is whether it would make sense to put a large amount of last-level cache into a hypothetical bespoke SoC. Something like 16MiB, which seems like it could soak up practically all frame and z-buffer traffic at 720p. Would that be a large perf/W win?

 

[Goodtwin]

The fact that they are the specs that require a 50 W SOC to achieve. And by 'wrong', I just mean they aren't presenting a black-and-white indication that Switch must be as powerful as other platforms using the same settings. That is, if those settings are there for Switch, we can expect Switch to run the same game with the same settings as any other platform with the same profile. For example, Switch might still need massive downgrades in quality for the same title and it's just UE4's in-built adaptation that we're seeing here. A game created and balanced to run at the mobile profile will scale up with zero dev effort to docked quality by using the higher quality settings relative to the mobile assets.

Now that's wishful thinking!

Who is arguing that Switch will match X1/PS4? Even the Unreal Profile specifies that the Switch profiles default to medium on Switch when they default to high on PS4/X1. I know this isn't conclusive, as running a game like Mario 3D World with Unreal 4 on medium settings is very different than running Batman AK, but to outright dismiss the information as meaningless because it doesn't fit your narrative is....well, it shows significant bias in your opinion. Not saying you wont ultimately be correct, but you are still filling in the blanks with assumptions none the less.

 

[function]

One thing I've been wondering is whether it would make sense to put a large amount of last-level cache into a hypothetical bespoke SoC. Something like 16MiB, which seems like it could soak up practically all frame and z-buffer traffic at 720p. Would that be a large perf/W win?

Transfers on chip should - according to my copied knowledge from genuinely smart people on B3D - almost always burn less power than off chip transfers. So it could be a way to increase performance while lowering power consumption. However, that would require a potentially significant increase in engineering time and cost (nvidia may not have a unified L3 cache complete with controller logic etc ready to roll) and would require a significant increase in chip size (32 MB on 28 nm was around 80 mm2). If Nintendo haven't told developers about it yet, it's extremely unlikely a game changer like that would pop up at the last minute.

Who is arguing that Switch will match X1/PS4? Even the Unreal Profile specifies that the Switch profiles default to medium on Switch when they default to high on PS4/X1. I know this isn't conclusive, as running a game like Mario 3D World with Unreal 4 on medium settings is very different than running Batman AK, but to outright dismiss the information as meaningless because it doesn't fit your narrative is....well, it shows significant bias in your opinion. Not saying you wont ultimately be correct, but you are still filling in the blanks with assumptions none the less.

The Switch profile - if accurate or meaningful - may be for the docked mode (significant difference in performance between docked and undocked) and may or may not be based on final clocks.

I would expect - though this is just a guess - that you'd ideally be looking at a single advised profile but with resolution changing between modes. That would make the suggested Unreal settings for NX seem ... optimistic? Though it's hard to decipher what practical impact they would have, and it's important to remember that developers will be expected to fine tune settings themselves.

 

[Shifty Geezer]

Who is arguing that Switch will match X1/PS4? Even the Unreal Profile specifies that the Switch profiles default to medium on Switch when they default to high on PS4/X1. I know this isn't conclusive, as running a game like Mario 3D World with Unreal 4 on medium settings is very different than running Batman AK, but to outright dismiss the information as meaningless because it doesn't fit your narrative is....well, it shows significant bias in your opinion. Not saying you wont ultimately be correct, but you are still filling in the blanks with assumptions none the less.

What's the purpose of the UE profile evidence? What do you get from it?