Nintendo Switch Tech Speculation discussion

[Goodtwin]

The XB1S SoC costs more despite being much smaller. It was estimated at $100 versus $75 for for the last 28nm version they made.

But for a portable device, power consumption is extremely important, I agree it makes little sense to not use 14/16nm.

Do you have a source for this? I have always been had the understanding that smaller processes reduce cost, and find it hard to believe that that the dye shrink increased cost by $25 per APU on the Xbox One. I could see if its a bleeding edge process that came out last night, but we are talking industry standard process that is being used for millions of devices currently in production.

http://www.neogaf.com/forum/showthread.php?t=1325344

So we have a ton of info from the recently filed patents by Nintendo. The fan does run when in portable mode, but at a lower speed. The overclocking in docked mode is specifically mentioned as a means to facilitate 1080p rendering. There is a ton of info there for anyone who is interested.

 

[AlNom]

But for a portable device, power consumption is extremely important, I agree it makes little sense to not use 14/16nm.

On the other hand, Nintendo isn't intending to compete with computing power, so nothing holds them to clocking a hypothetical 20nm part particularly high. If they were looking at a low enough power point, perhaps switching to FF wouldn't have gained them as much as a much larger, more power consuming chip ala desktop GPUs & the other consoles. In that case, TSMC would have had to have offered a discount of sorts for keeping the 20nm fab line occupied, where they have plenty of utilization for their 16nmFF lines & would price it accordingly to supply/demand.

Hypothetically speaking.

 

[ninelven]

The overclocking in docked mode is specifically mentioned as a means to facilitate 1080p rendering.

There is actually significant information contained here if people pause and use their brains for a bit .

 

[Deleted member 13524]

The XB1S SoC costs more despite being much smaller. It was estimated at $100 versus $75 for for the last 28nm version they made.

The cost was higher in H1 2016. For all we know, it could be similar now and actually lower come mid 2017.
These chips aren't designed to be produced during a 6-month period, they're designed with waffer price evolution in mind.

 

[Goodtwin]

On the other hand, Nintendo isn't intending to compete with computing power, so nothing holds them to clocking a hypothetical 20nm part particularly high. If they were looking at a low enough power point, perhaps switching to FF wouldn't have gained them as much as a much larger, more power consuming chip ala desktop GPUs & the other consoles. In that case, TSMC would have had to have offered a discount of sorts for keeping the 20nm fab line occupied, where they have plenty of utilization for their 16nmFF lines & would price it accordingly to supply/demand.

Hypothetically speaking.

Sure, but we now know from the patents that even in portable mode, its actively cooled, and the fan ramps up considerably when docked. The Switch is about twice as thick as a tablet, and there are two vents on the bottom and a large vent on top. We also know that Dark Souls 3 runs on Switch with performance the developer is happy with. This thing is going to clock high. You don't need three vents and a fan running targeting 720p in portable mode if this thing doesn't clock high.

@MrFox

I found an article that suggested the cost of the Xbox One S Finfet 16nm APU being $99, but this was about six months ago, and the same article mentions that the cost will likely come down significantly as the process matures, and I think its safe to assume that this has happened. An article for the launch Xbox Ones showed a cost of $110 for the APU, and I am sure that came down quite a bit over the 3 years leading to the new Xbox One S.

A dye shrink is pretty much the cheapest way to increase performance. If you were to take a Tegra X1, do nothing but dye shrink it to Finfet 16nm, you would be able to clock it at or near 1.5Ghz. Scrap the 64bit memory bus for a 128bit memory bus, and you have a chip that greatly outperforms a stock Tegra X1 for about the same price it cost to manufacture the standard 20nm Tegra X1 back in early 2015.

 

[Esrever]

I called maxwell 3 months ago. Now I am going to call 20nm. I can't wait to be proven wrong but nintendo is nintendo.

 

[Lalaland]

http://www.neogaf.com/forum/showthread.php?t=1325344

So we have a ton of info from the recently filed patents by Nintendo. The fan does run when in portable mode, but at a lower speed. The overclocking in docked mode is specifically mentioned as a means to facilitate 1080p rendering. There is a ton of info there for anyone who is interested.

There is actually significant information contained here if people pause and use their brains for a bit .

http://appft.uspto.gov/netacgi/nph-...d=PG01&s1=15/178984&OS=15/178984&RS=15/178984

Hmmm always running fans in a portable device? Sounds like a really crappy idea to me to be frank.

Beyond saying that they're going to turn the fan up when docked there really isn't much in the patent filing, given the 2.25 diff in the number of pixels to be driven by docking it.

[0551] (a) Setting for Limiting Processing Power of Main Unit 2

[0552] In the present embodiment, in the portable mode, the clock frequency range over which the CPU 81 is allowed to operate is limited. In the present embodiment, the clock frequency at which the CPU 81 is allowed to operate can be specified within a predetermined range by the program executed on the main unit 2. In the portable mode, the range over which a clock frequency can be specified by the program is limited as compared with that in the console mode. For example, the range over which a clock frequency can be specified in the console mode is X1 [Hz] or less, whereas it is limited to X2 (<X1) [Hz] in the portable mode. Note that if the main unit 2 includes a GPU (Graphics Processing Unit) in addition to a CPU, the range of processing power (i.e., clock frequency) may be limited for the CPU and/or for the GPU.

[0553] In the present embodiment, in the portable mode, the image rendering capacity (which can also be referred to as the image generating capacity) is limited. Specifically, in the portable mode, the resolution (in other words, the number of pixels) of images generated by the main unit 2 is lower than that in the console mode.
...
[0558] On the other hand, the console mode is a mode in which the information processing device 1 is used as a console-type device. In the console mode, images obtained or produced by the information processing device 1 are displayed on the TV 6. The sound obtained or produced by the information processing device 1 is output from the speaker of the TV 6. In the console mode, the function suppressions in the portable mode are lifted. That is, in the console mode, the limitations (a) and (b) described above are lifted. Therefore, in the console mode, a program running on the information processing device 1 can make better use of the processing power of the CPU 81. The information processing device 1 can display, on the TV 6, images of a higher resolution than in the portable mode. The information processing device 1 can be cooled by the cooling fan 96 more effectively than in the portable mode.

Note In [0558] b) is a fan

 

[AlNom]

Sure, but we now know from the patents that even in portable mode, its actively cooled, and the fan ramps up considerably when docked. The Switch is about twice as thick as a tablet, and there are two vents on the bottom and a large vent on top.
...
This thing is going to clock high. You don't need three vents and a fan running targeting 720p in portable mode if this thing doesn't clock high.

You might design it that way if you're not planning on throttling, which plenty of devices do, over the course of a few hours.

 

[ninelven]

Beyond saying that they're going to turn the fan up when docked there really isn't much in the patent filing, given the 2.25 diff in the number of pixels to be driven by docking it.

Which is why one needs to use their brain (really not that much though)....

 

[turkey]

Sure, but we now know from the patents that even in portable mode, its actively cooled, and the fan ramps up considerably when docked. The Switch is about twice as thick as a tablet, and there are two vents on the bottom and a large vent on top. We also know that Dark Souls 3 runs on Switch with performance the developer is happy with. This thing is going to clock high. You don't need three vents and a fan running targeting 720p in portable mode if this thing doesn't clock high.

@MrFox

I found an article that suggested the cost of the Xbox One S Finfet 16nm APU being $99, but this was about six months ago, and the same article mentions that the cost will likely come down significantly as the process matures, and I think its safe to assume that this has happened. An article for the launch Xbox Ones showed a cost of $110 for the APU, and I am sure that came down quite a bit over the 3 years leading to the new Xbox One S.

A dye shrink is pretty much the cheapest way to increase performance. If you were to take a Tegra X1, do nothing but dye shrink it to Finfet 16nm, you would be able to clock it at or near 1.5Ghz. Scrap the 64bit memory bus for a 128bit memory bus, and you have a chip that greatly outperforms a stock Tegra X1 for about the same price it cost to manufacture the standard 20nm Tegra X1 back in early 2015.

On release IHS Market said in a press release

Slightly higher component costs, lower R&D

The new CPU device in the Xbox One S, supplied by AMD is now fabricated by TSMC (Taiwan Semiconductor Manufacturing Company) using a 16 nanometer (nm) “FinFET” process, rather than the previous standard of 28nm technology. While this process resulted in a “die shrink” at this early stage it’s typical to have a lower yield rate, which has been built into the estimate of $99 versus the previous device cost of $76. “As the production of this device matures, and as yields rise, we can expect commensurate cost erosion,” Keller said.

_____________________

The current value in the bom estimate has dropped nearly 15%,

 

[Picao84]

There is actually significant information contained here if people pause and use their brains for a bit .

Well, it does open the possibility for the dock unit to have an additional GPU.. Such a GPU, being dedicated whitout a CPU block could still be passively cooled, hence the thin dock?

 

[Goodtwin]

The increase in performance when docked likely isn't sufficient to go from 720p to 1080p. With that said, we are probably looking at it wrong. Basically, if the developer targets 1080p when docked, then the performance will be great in portable mode. I would bet this will be how Nintendo themselves handle development. Get the game running in 1080p 60fps docked, and 720p 60fps portable mode is a lock. Im sure some developers will push the Switch to the breaking point, and docked will give 720p and portable 600p, but regardless, the extra clocks are there to facilitate higher resolutions when gaming on the TV.

 

[turkey]

The increase in performance when docked likely isn't sufficient to go from 720p to 1080p. With that said, we are probably looking at it wrong. Basically, if the developer targets 1080p when docked, then the performance will be great in portable mode. I would bet this will be how Nintendo themselves handle development. Get the game running in 1080p 60fps docked, and 720p 60fps portable mode is a lock. Im sure some developers will push the Switch to the breaking point, and docked will give 720p and portable 600p, but regardless, the extra clocks are there to facilitate higher resolutions when gaming on the TV.

What if the dock actively cools or powers enough to boost compute to do a pspro and reprojet upto 1080.

The dock may boost gfx but it might not be a massive GPU increase but a smart software / hardware mode change.

It is quite a technically challenging design, given it seems Nvidea have been tasked with a lot I find it hard to think they would not think up clever tricks to achieve the results. They already been have their own smart AA and other tech, they are not lacking in software R&D

Pascal and compute or Maxwell with improved compute would fit better with this hypothesis.

TLr Nvidea own the software stack, not all boosts need to be 1:1 to hardware performance.

Disclaimer , Xmas party tonight, spelling be dammed

 

[orangpelupa]

Is the 720p screen resolution has been mentioned in the patent?

If not, maybe switch actually have 1080p screen but normal games are forced to render to 720p while vr games in 1080p?

 

[Lalaland]

Which is why one needs to use their brain (really not that much though)....

I can't guess what your speculation is here, from the docs we have a dumb as a box of rocks charger/video out box that also tells the switch to clock itself higher. Nothing about that suggests an automatic boost from 720p on the go to 1080p docked, if anything I would guess devs will target 720p on the go and upscale from to 1080p from whatever res they can hit at the higher clocks when docked (some will be native 1080p some will be 900p like on PS4 etc).

 

[fehu]

http://arstechnica.com/gaming/2016/...-suggest-possibile-head-mounted-vr-accessory/

I was right confirmed u.u

 

[Lalaland]

Given the low res of that large screen I think we'll need to start talking about the chain link fence rather than screen door effect. I'd see this as a 360 vid viewer or party trick in much the same way Google glass is

 

[AlNom]

Not the Link fans want from Nintendo.

 

[milk]

Party trick is all they need. Its a good move, comercially.

 

[Shifty Geezer]

That patent is dumb in sooooo many ways. The patent is for a supporting device with rails for two controllers either side. And then they try and hang a headset off that patent as an embodiment despite the headset not conforming in any way to the patent. Not to mention the headset idea is clearly unpatentable due to prior art.

I think this was just a practice patent for some trainee patent lawyers to let them stretch themselves in thinking up related claims and clauses, hence a bazillion paragraphs.