Hardware Nintendo Switch Technical Specifications Discussion

[sarkwalvein]

Near xb1 power for $250 is achievable. Near xb1 power at much lower wattage and stays cool in a handheld form factor is not.

Actually, as a fact...
Near xb1 power for $250 (or even cheaper) is called Xbox One.
Near xb1 power for near $250 and lower wattage is called Xbox One S.
Cool and in a handheld form factor............ Well, it depends how you define "near".

 

[Obveron]

I'm talking a mobile TDP here, the XB1s is not in the same TDP category as a mobile device like the the Switch. Get Xb1 performance down to 20 watts at full load and now we're talking. There's just no hardware that exists that is that efficient. Certainly not Tegra X1.

 

[sarkwalvein]

Well, well. No idea what is the TDP of the Switch, because everything is rumours, but the Xb1S drains ~50W while gaming, and the unreliable rumours for the Switch said something like 39W docked, so it is not that much apart. Of course, rumours are BS.

 

[Yil]

Snakebyte knows the screen size of the switch by now.

Well, well. No idea what is the TDP of the Switch, because everything is rumours, but the Xb1S drains ~50W while gaming, and the unreliable rumours for the Switch said something like 39W docked, so it is not that much apart. Of course, rumours are BS.

XBox one S while using new production tech does not feature the full set of newer tech like a new engine. Tegra on the other hand is superior than Desktop counterpart at ultra low voltage.

 

[sarkwalvein]

Snakebyte knows the screen size of the switch by now.

XBox one S while using new production tech does not feature the full set of newer tech like a new engine. Tegra on the other hand is superior than Desktop counterpart at ultra low voltage.

I'm not sure exactly what are you implying but you seem like you're endorsing my post.
Anyway I have to say that, even if I haven't checked for the Xb1S, almost everything in the late years, specially notebooks and most probably consoles, use ULV.

 

[Yil]

I'm not sure exactly what are you implying but you seem like you're endorsing my post.
Anyway I have to say that, even if I haven't checked for the Xb1S, almost everything in the late years, specially notebooks and most probably consoles, use ULV.

XB1S uses a new production node but still uses older architecture, and ROP is likely redesigned for 4k HDR.
And Tegra is more mobile focused despite lacklustre performance per watt (940mx 15w).

 

[TheDarkGreninja]

Well, well. No idea what is the TDP of the Switch, because everything is rumours, but the Xb1S drains ~50W while gaming, and the unreliable rumours for the Switch said something like 39W docked, so it is not that much apart. Of course, rumours are BS.

BS until proven, LKD just got proven correct with USB-C (she said it would be on october)

--------------------- MERGED ---------------------------



Well, we've just seen what the X2 is being used in, such a sad missed opportunity for Nintendo (not the car stuff, the fact switch wont be using the X2 but a custom maxwell chip.) Would an X2 even fit in the switch? Die definitely seems larger than X1.

 

[Yil]

BS until proven, LKD just got proven correct with USB-C (she said it would be on october)

--------------------- MERGED ---------------------------



Well, we've just seen what the X2 is being used in, such a sad missed opportunity for Nintendo (not the car stuff, the fact switch wont be using the X2 but a custom maxwell chip.) Would an X2 even fit in the switch? Die definitely seems larger than X1.

There is no point though. Denver is meant to handle dgpu.

 

[TheDarkGreninja]

There is no point though. Denver is meant to handle dgpu.

What do you mean?

 

[Yil]

What do you mean?

Tegra X2 is meant to drive a gp106/ gtx 1060 for auto drive applications, not used as a standalone chip. It's not really that good for gaming. x1 refresh with 16nm would do better.

 

[TheDarkGreninja]

Tegra X2 is meant to drive a gp106/ gtx 1060 for auto drive applications, not used as a standalone chip. It's not really that good for gaming. x1 refresh with 16nm would do better.

Thats the thing a variant of the X2 would make sense on the switch. Although Denver might not fit inside such a small chip.

 

[Xzi]

Near xb1 power for $250 is achievable. Near xb1 power at much lower wattage and stays cool in a handheld form factor is not.

Sure it is. XB1 is running on a 2011 architecture, and it's GCN-based. With Maxwell-based architecture you only need to push ~750GFlops to be in the same performance range as XB1. Your stock X1 already does 500GFlops on 11W and doesn't require active cooling. On 22W with double the SMs and cores it'd push 1TFlop. That's assuming it's not underclocked like the Eurogamer rumor, but still. If it's normal X1 clocks with 4SMs and 512cores, it would actually outperform XB1 occasionally at just 22W. The 39W on Switch patents would leave plenty of overhead for charging and other things. It has active cooling but it really won't put out much heat.

 

[sarkwalvein]

In order to match X GFlops of a given precision, you have to pull out X GFlops, that is the benchmark metric.
You could say it is just peak performance, and that would be true.
You could also say that another architecture may provide better cache hit algorithm, better inter thread communication and less parallelisation overhead, but in the end if it can't even reach the X/2 GFlops peak it will not have a better average.
Why you still say blah/2 GFlop of architecture A is better than blah GFlop from architecture B?!
If it can do less floating operations per second, it has less number crunching power, and it performs worse.

PS: I'm curious, if there is a logic or reason for this I would like to know.

 

[Xzi]

In order to match X GFlops of a given precision, you have to pull out X GFlops, that is the benchmark metric.
You could say it is just peak performance, and that would be true.
You could also say that another architecture may provide better cache hit algorithm, better inter thread communication and less parallelisation overhead, but in the end if it can't even reach the X/2 GFlops peak it will not have a better average.
Why you still say blah/2 GFlop of architecture A is better than blah GFlop from architecture B?!
If it can do less floating operations per second, it has less number crunching power, and it performs worse.

PS: I'm curious, if there is a logic or reason for this I would like to know.

Maxwell consistently outperforms GCN by about 30% in all benchmarks. Regardless of FLOPs calculations. So at ~750GFlops the relative performance of Maxwell to GCN is ~1TFlop, with XB1 sitting at 1.3TFlops GCN. At 1TFlop Maxwell relative performance to GCN would be ~1.3TFlops, so the same or a little better than XB1 in actual performance. Just not on paper.

 

[Obveron]

Well, well. No idea what is the TDP of the Switch, because everything is rumours, but the Xb1S drains ~50W while gaming, and the unreliable rumours for the Switch said something like 39W docked, so it is not that much apart. Of course, rumours are BS.

39W in that small enclosure? yikes, sounds hot. I hope the fans in the dock can keep up.
I just don't see how Nintendo can cram XB1 performance into a package that is perhaps 10% the volume, and throw in a display at the same time, and keep it running cool. Maxwell is more efficient than GCN, I completely agree with that. but the leap in efficiency required seems out of reach. I'm inclined to believe the downclock rumours, simply because they seem to be realistic adjustment to keep temperature under control.

 

[TheDarkGreninja]

39W in that small enclosure? yikes, sounds hot. I hope the fans in the dock can keep up.
I just don't see how Nintendo can cram XB1 performance into a package that is perhaps 10% the volume, and throw in a display at the same time, and keep it running cool. Maxwell is more efficient than GCN, I completely agree with that. but the leap in efficiency required seems out of reach. I'm inclined to believe the downclock rumours, simply because they seem to be realistic adjustment to keep temperature under control.

Right now its all speculation, once we get our hands on it we'll find out a bit more. Well the switch was never going to be more powerful than XBONE in terms of GPU power, CPU wise it couldve easily demolished it with a few A73's. Reason being it would have less cores and would have to run at a lower clock, however, those specs that EG released make no sense next to the UE4 presets, im inclined to believe the UE4 presets more than EG's report.
Edit: Forgot to say that USB-C can do up to 100 watts in power.

 

[Yil]

Right now its all speculation, once we get our hands on it we'll find out a bit more. Well the switch was never going to be more powerful than XBONE in terms of GPU power, CPU wise it couldve easily demolished it with a few A73's. Reason being it would have less cores and would have to run at a lower clock, however, those specs that EG released make no sense next to the UE4 presets, im inclined to believe the UE4 presets more than EG's report.
Edit: Forgot to say that USB-C can do up to 100 watts in power.

20V 5A is way too much for a mobile controller.
A73 is massively more efficient than jaguar and should be able to reach twice the performance, but still not demolishing.
So did you see UE4 Preset? Was it on normal branch too?

 

[TheDarkGreninja]

20V 5A is way too much for a mobile controller.
A73 is massively more efficient than jaguar and should be able to reach twice the performance, but still not demolishing.
So did you see UE4 Preset? Was it on normal branch too?

Nope, not on normal branch yet i dont think. I exaggerate when I say things

 

[Yil]

Nope, not on normal branch yet i dont think. I exaggerate when I say things

Do you think it's possible that the switch can match Geforce 940mx (found in Surface Book) in mobile mode? 16nm vs 28nm.

 

[TheDarkGreninja]

Do you think it's possible that the switch can match Geforce 940mx (found in Surface Book) in mobile mode? 16nm vs 28nm.

940mx is maxwell right? If so, yeah, most likely. Although when it comes to gaming, Epic have definitely optimised switch massively for ue4. So comparisons are a little muddy when basing it on just the presets.