"New" Nintendo Switch (OLED Model) [2021-07-06]

Steam Deck is interesting from a Switch perspective because it shows what is possible this autumn/winter inside a 15-20W power envelope. RDNA2 is pretty much as power efficient as it gets, and TSMC 7nm is arguably the best readily available process.
Mariko draws roughly 3W undocked and in the Switch Lite.

For those that want a substantially more powerful Switch, Steam Deck should serve as a sanity check.
 
Steam Deck is interesting from a Switch perspective because it shows what is possible this autumn/winter inside a 15-20W power envelope. RDNA2 is pretty much as power efficient as it gets, and TSMC 7nm is arguably the best readily available process.
Mariko draws roughly 3W undocked and in the Switch Lite.

For those that want a substantially more powerful Switch, Steam Deck should serve as a sanity check.

It also shows that Nintendo won't be able to get as good of a deal on HW out of Nvidia compared to Valve/AMD and that full backwards compatibility has a moderate hidden cost since die space used on redundant logic to provide compatibility can't be easily reused to make the HW more powerful ...

Assuming that Samsung's 8nm process scaling is anywhere near as optimistic as TSMC's 7nm process, Nvidia would have anywhere between 60-90 mm^2 of die size for adding new HW to the existing Mariko design which could hypothetically take up as much as 60mm^2 if it was ported to the new process as well. Or Nintendo could just as well decide to dump the whole Mariko design itself and have potentially as much as 150mm^2 to play around with to make more powerful HW ...

Either way their options are limited. If they want to retain full compatibility then making significantly more powerful HW becomes intractable since their partner doesn't like doing discounts or extending/enhancing existing design. If they decide to instead opt-in for another vendor then they suddenly have access to more discounts and superior process technology thus opening up options to make more efficient or cheaper HW at the cost of dropping compatibility ...
 
Hopefully it will use TSMC 5nm process or Samsungs new 4nm process.

Samsungs 8nm was not great in 2020 given its pretty terrible power efficiency. It would not be suitable in such a power constrained device at all and given it will likely release in 2023, anything higher than 5nm would just be bad.
 
Nintendo doesnt need TFlops. Advanced upscaling techniques are more usefull. So with Ampere they can render at 720p and even upscale to 2160 with DLSS-UP and get better image quality than render at 1080p without DLSS.
 

Steam deck starts at $400 so its just $50 more than the oled switch.

For Nintendo they don't need a huge ssd. They could just support the carts and micro sd again
 
Aye? I thought the M1 absolutely destroyed it, though of course we aint gonna see apple chips in a switch

That picture was certainly painted when the M1 released last year with a couple of gaming benchmarks (or rather.. one benchmark, i.e. Tomb Raider?), but the notebookcheck comparisons tell a different story. On most of the games that were tested it doesn't seem to get that much better gaming performance than Tiger Lake U and Cezanne U, both at 15W.
It's certainly not the "discrete graphics killer!!!" that we saw suggested over and over at the time.
 
That picture was certainly painted when the M1 released last year with a couple of gaming benchmarks (or rather.. one benchmark, i.e. Tomb Raider?), but the notebookcheck comparisons tell a different story. On most of the games that were tested it doesn't seem to get that much better gaming performance than Tiger Lake U and Cezanne U, both at 15W.
It's certainly not the "discrete graphics killer!!!" that we saw suggested over and over at the time.
all those game benchmarks are running translated code not native https://www.macgamerhq.com/apple-m1/native-mac-m1-games/ <- bugger all native titles

benchmark borderlands 3 1080p medium
UHD Graphics Xe 750 32EU 11fps 45 watts
M1 26 fps 15 watts

more than double the performance at a third of the power and its not even running native to rub salt in the wounds, mate its not even close,
 
ARM CPUs are significantly more power efficient than X86 CPUs.
It's true that this is a factor. I doubt it's that big though, mainly because I can't see the CPU block being the main power drain in a game console. So the difference in power draw due to CPU architecture seems as though it should be modest.

Who said Next Switch will be using Samsung 8N process?
The rumour mill. But from a performance/Watt point of view it is not too dissimilar from the process Mariko is already on (albeit twice as dense), and significantly worse than TSMC 7nm that Van Gogh uses. I can't really see it as a viable node for a Switch console, unless Nintendo got a low enough price per wafer that they could save money on it simply scaling Mariko. As Dampf points out, it's power characteristics makes it problematic for low power design, if you are looking at even modest increases in performance over Mariko.
Nintendo doesnt need TFlops. Advanced upscaling techniques are more usefull. So with Ampere they can render at 720p and even upscale to 2160 with DLSS-UP and get better image quality than render at 1080p without DLSS.
Not true. While DLSS is a low cost process on a discrete desktop GPU, its demands get problematic on low power devices. If, for example, you look at a desktop Ampere 3090 GPU, you need to divide its speed by 20 (to simulate 4SMs) and that still represents 15+W for the GPU alone (and you need to factor in CPU and memory in the 3W of Mariko) so you need to divide clocks by three or so to get anywhere close to where you need to be. And of course you still need time for the game logic to do its thing, and of course the rendering at the native resolution. No matter what I do, I can't get the numbers to fit when you take target SoC limitations into account. Not even close. (And given the nature of DLSS, all post processing would go after the DLSS step, at a full 4k.)
Now, I don't know to what extent a future Nintendo SoC could change the variables around. But in its present incarnation the approach is problematic within the Switch design constraints. Something more light weight is needed.

Basically, if you want to have a Switch that performs like a Steam Deck, you have to somehow reduce power draw by a factor of five. That's ... quite a challenge.
 
GTX980 was a 165W card, Tegra X1 used 10 to 15W for 512 GFLOPs. Why wouldnt it be possible to do a 1TFLOPs SoC within 10 to 15W?
3090 is a bad comparision because GDDR6X uses more power than GDDR6. The quadro version has a 300W TDP.
 
GTX980 was a 165W card, Tegra X1 used 10 to 15W for 512 GFLOPs. Why wouldnt it be possible to do a 1TFLOPs SoC within 10 to 15W?
3090 is a bad comparision because GDDR6X uses more power than GDDR6. The quadro version has a 300W TDP.

Steam deck already claims more than 1TF with 15w max. So I think Nvidia should be able to achieve similar things too. Although I'm not sure with their expertise in the cpu département...
 
GTX980 was a 165W card, Tegra X1 used 10 to 15W for 512 GFLOPs. Why wouldnt it be possible to do a 1TFLOPs SoC within 10 to 15W?
Oh, it is! That’s roughly what Steam Deck does. But since it uses an architecture and process which are both currently pretty much as energy efficient as you can get unless you’re Apple, that’s what it costs. Or putting it in other words, doing it in 3W can’t be done without major strides in process technology.
The Switch Lite has a 3570mAh battery, and has a 3-4h battery life when new. Subtract networking, subtract screen power draw and other smaller drains and losses and you are left with (possibly a little less than) 3W for SoC and memory. The same goes for the original only it has a 4310mAh battery. Also, if you’re comparing to pure GPU:s, you also have to subtract the power draw of the CPU block. Andrei Frumusanu published interesting power vs. frequency curves and some conclusions regarding Samsung 5nm vs TSMC 7nm here. Which also shows that even these efficient cores actually draw a substantial chunk of power once we’re down at these target levels.

Mariko doesn’t get much credit for what it achieves within its power constraints, (nor Nintendo for what it delivers at $199).
The Steam Deck, much as I like it for what it is, isn’t a design that is particularly attractive as a new Nintendo Switch. Expectations for a successor model need to be tempered accordingly.
 
benchmark borderlands 3 1080p medium
UHD Graphics Xe 750 32EU 11fps 45 watts
M1 26 fps 15 watts

Why would you compare the 5nm M1 to a desktop 14nm Rocket Lake with a GT1 GPU?

The 10nm 1165G7 Tiger Lake at 15-28W does 25 FPS on the same game and same settings.

https://www.notebookcheck.net/Acer-...GPU-attacks-Entry-Level-GeForce.496234.0.html


Get something like Van Gogh is the Steam Deck and it should be a lot faster than that at the same 15W TDP, despite using an older 7nm process. And Intel isn't exactly known for their GPU efficiency prowess yet.
I'm pretty sure Nvidia could do something faster than the M1 at 15W too.




more than double the performance at a third of the power and its not even running native to rub salt in the wounds, mate its not even close

Please do spare us the bolded hyperbolic statements.. It sounds fanboyist..

The M1's GPU efficiency is great. It's just not as great as many painted it to be. When it released in late 2020 on TSMC's brand new 5nm, it was mostly compared to the Polaris 11 GPUs from 2016 made on GlobalFoundries' 14nm. Of course it was going to show much better power efficiency than those.
 
Why would you compare the 5nm M1 to a desktop 14nm Rocket Lake with a GT1 GPU?

The 10nm 1165G7 Tiger Lake at 15-28W does 25 FPS on the same game and same settings.

https://www.notebookcheck.net/Acer-...GPU-attacks-Entry-Level-GeForce.496234.0.html
sorry for the BS I thought that was a tiger lake, I see the one you link to is 28W (though does more)

the apple m1 -
The peak performance of the high end variant is 2.6 teraflops, the M1 GPU uses under load approximately 10 Watt (11.5 Watt package power including the RAM)
10 Watts!, how much does that slower performing tiger lake GPU do again :)
Remember Intel just BSs with their power ratings
https://www.anandtech.com/show/16084/intel-tiger-lake-review-deep-dive-core-11th-gen/5

Mate I'm definitely not an apple fanboy, I just are stating the reality that per watt they are destroying the competition. Im hoping this embarrassment of intel / amd forces them and windows to advance their game, as I have no desire to use the apple OS again, I wanna stick with windows
 
Last edited:
Btw anyone compared switch oled speakers to Nintendo new 3ds xl?

I compared my launch model switch to new 3ds xl and turns out

  1. Switch is quieter
  2. New 3ds xl have much richer sound
  3. New 3ds xl have much better stereo seoarati
 
I have always thought the speakers in the Switch were "ok", OLED improved them a bit, but as a rule, anytime I am really into a game while playing portably, I use headphones. Playing Metroid Dread on my new Oled Switch with headphones on is a great experience.
 
Back
Top