Fair enough. I guess, since I see a Switch 2 being rather far out, I should only really speculate about what would be in a Switch Pro or something
-
Ever wanted an RSS feed of all your favorite gaming news sites? Go check out our new Gaming Headlines feed! Read more about it here.
Does Nvidia have any chip that we know of that would be ideal for Switch 2 whenever it releases?
- Thread starter Hero of Legend
- Start date
I am just commenting on how impossible it is to talk about technology 5 years away, I mean the switch will be 6+ years old in 2023-2024 and we went from vita to switch in just a 5 year period. A 35gflops console from Sony with 750GB ram to a 400gflops console from Nintendo with 4GB of ram. 5 years down the line, with solid state batteries, process nodes that will easily handle 5 to 10 billion transistors in a similar die size to the current Switch and even 10x the memory bandwidth of the current Switch doesn't seem hard to imagine.
Keeping the technology to what can be released in the next year or two is about the only way to focus speculation imo.
You can bet the most demanding games are using pretty much every trick possible to squeeze out the most from the hardware.
Another thing supporting an iterative model coming next year, is Iwata's quote about NX adopting iOS and Android's rapid hardware turn around, and moving away from hardware platforms, and going forward with a hardware agnostic approach to the NX platform.
Frankly if Wii U didn't exist, Switch might have launched holiday 2015.
The PS4 and XB1 use 2011 GCN even though they came out in 2013. The Switch was also originally suppose to launch Holiday 2016, but as the former president of Nintendo said, they missed the holiday launch because they didn't have the software ready to launch the console.
Frankly if Wii U didn't exist, Switch might have launched holiday 2015.
Maybe I'm being too optimistic here. But I'm expecting the Switch 2 to have undocked graphics performance somewhere between Xbox One and PS4 and docked graphics performance to be close to PS4 Pro but with significantly better CPU performance.
Isn't the iPad Pro basically as powerful as the Xbox One graphically but with a much better CPU? That came out in 2018. I remember that Anandtech mentioned that the iPad Pro draws 8 watts of power in total. That is very similar to the Switch. A Switch 2 in 2022/2023 should at least match that and more right? If Nintendo targets 720p and 1080p for the Switch 2, I expect ports of PS5/Scarlet games to fare much better than what we're getting for the Switch now.
Isn't the iPad Pro basically as powerful as the Xbox One graphically but with a much better CPU? That came out in 2018. I remember that Anandtech mentioned that the iPad Pro draws 8 watts of power in total. That is very similar to the Switch. A Switch 2 in 2022/2023 should at least match that and more right? If Nintendo targets 720p and 1080p for the Switch 2, I expect ports of PS5/Scarlet games to fare much better than what we're getting for the Switch now.
Apple is years ahead of the rest of the mobile tech industry when it comes to cpus, but when it comes to gpu architecture/ featureset nvidia is on the forefront (at least in theory if they bring their latest architecture). Nintendo will not use a state of the art fabrication node though, as Apple will.
I do think this is a reasonable expectation.
GCN reached the market in early 2012,
the December 2011 reviews were pre-launch
also it looks like the PS4 or XB1 have improvements like from GCN 1.1 from 2013
the switch Tegra X1 was launched in late 2015 (with devices like https://www.gsmarena.com/google_pixel_c-7826.php ) and uses tech from late 2014 (Maxwell 2.0)
I remember
I'm not exactly sure how it works, but I keep hearing that Nvidia tile rendering with mobile RAM has some leeway on bandwidth usage. I reckon just doubling our bandwidth would be enough to remove the bottleneck the current switch has and make it on par with xbone performance wise (assuming we got a CPU and GPU clocked 2.5x higher to match xbone base also). I think 128 bus bandwidth with LPDDR5 can give us something close to 100GB/s, which should be more than enough to match PS4...
10x the bandwidth in raw on paper numbers on a handheld does seem hard to imagine for me on a mobile device using mobile RAM like LPDDR4 or LPDDR5. Is that even physically possible on , assuming we get something crazy like 1-2 TFLOPs handheld and 4 GFLOPs docked in switch. We will definitely need at least 128 bit buswidth. Maybe 256. But I don't know if that will physically fit on a handheld.. unless we can stack the buswidth somehow (Ram on its own is possible to stack vertically).
I'm not exactly sure how it works, but I keep hearing that Nvidia tile rendering with mobile RAM has some leeway on bandwidth usage. I reckon just doubling our bandwidth would be enough to remove the bottleneck the current switch has and make it on par with xbone performance wise (assuming we got a CPU and GPU clocked 2.5x higher to match xbone base also). I think 128 bus bandwidth with LPDDR5 can give us something close to 100GB/s, which should be more than enough to match PS4...
Last edited:
They are probably using 16fp over the NVN api i guess. Also the answer to your question is still no. They don't use all the power of the X1 as it's underclocked. And the Mariko Revision even more.
They seem to not care about performance boosts for the 2019 model but that also means they don't see it enough of a boost to call it Pro or even a upgrade (which seems to be a good sign for a real Pro if they decide to make one).
The 720p aren't a problem in my eyes, the Switch is no Smartphone which is used in other ways. The problem is that many games can't achieve the native 720p resolution, even most Nintendogames don't or are at least variable in resolution which does make the image blurrier. We won't see anything like Xbox One X level of power in a handheld/hybrid from Nintendo in the next years. I would be more than happy with PS4 levels, we are still far away from the 1teraflops border with the Switch.
Yeah, i'm not even thinking about ports too much (it would be great to get more bigger japanese games though), my biggest reasons to want a stronger Switch are Nintendogames. Zelda BotW2, the next Xenoblade, Mario etc.
that is actually one of my biggest issues. Like games can look nice (not great but not terrible!) on that screen at 720p because of its size but games not even hitting that is always incredibly disappointing.
That being said have a similar sized screen, some more power and make it so that some hit native 1080p while others drop to 720p or 900p and that issue would be pretty small i feel. If they keep the screen size the same obviously.
I agree, depends on how strong they make it though. When a sizable amount of games can reach 1080p in undocked mode it would be worth it to include a 1080p screen. Also i read that a 1080p screen could be cheaper in the future than a 720p screen. I guess a 1080p screen is more energy hungry though.
Yeah you are right, they weren't commercially available until January 9th 2012. The chip inside the base PS4 launched on January 9th.
Tegra X1 is Maxwell 3.0 or Pascal 0.9. It was a new chip with features not seen in other maxwell cards like half precision (fp16). It is also the only Nvidia chip to use 20nm.
The Tegra X1 was at worst, the same age as GCN when their respective consoles launched. The Tegra X2 wasn't even consumer available until the very end of 2016 iirc.
Wii dragging for 9 years would be terrible. Lol
Until we get PS4 specs on handheld mode, 1080p is something we will always be chasing because devs (especially 3rd party) will continue to push as much out if the system as possible in fidelity. But 1 TFLOP xbone specs on handheld is more than enough (with majority of games reaching 1080p)and I think when we reach that, having a 1080p screen would be nice and should happen. 720p screen on 2021-2022 feels off. If switch pro ends up being 400 GFLOPs and 1 TFLOP, then sticking to 720p would be necessary.
Nvidia Shield refresh using the die shrunk TX1 chip is only 25% faster than the previous one. Hmm
I'm pretty sure whatever's next will have to be semi-custom. If both other new consoles have raytracing, it'll need it too if it wants ports. Otherwise, it'll be an excuse to skip the platform.
OP
OP
It's a possibility. Mariko in 7nm. If they do that i will beginn doubting the decades long future of their relationship.
Mariko on 7nm would be a pretty poor product. even by nintendo standards
A Tegra chip roughly equivalent to the Apple A13X (coming this fall/early next year) is what I'd expect.
The Tegra X1 was roughly equivalent to an Apple A9X processor.
The Apple A12X is supposedly in line with an XBox One S (1.4 TF) with a better CPU, so an A13X would be above that.
That said I'd expect a far more modern architecture, Ampere-based with hardware ray-tracing possibly.
So it wouldn't just be a "PS4 portable", it would be able to handle ports of some PS5/XB2 games at reduced resolution. Just as looking at the current Switch it should basically just be an XBox 360 tier hardware+ but it's able to handle PS4/XB1 ports, the far more modern architecture makes a big difference.
The Tegra X1 was roughly equivalent to an Apple A9X processor.
The Apple A12X is supposedly in line with an XBox One S (1.4 TF) with a better CPU, so an A13X would be above that.
That said I'd expect a far more modern architecture, Ampere-based with hardware ray-tracing possibly.
So it wouldn't just be a "PS4 portable", it would be able to handle ports of some PS5/XB2 games at reduced resolution. Just as looking at the current Switch it should basically just be an XBox 360 tier hardware+ but it's able to handle PS4/XB1 ports, the far more modern architecture makes a big difference.
PS4 level GPU (but more modern) or a bit better with a better CPU than the PS4 and Xbox One is honestly what I expect, but no hardware ray-tracing support.
I think in general the barriers to port games down to the Switch 2 from PS5 and Xbox Two will be much smaller than what we have now, especially thanks to the resolution differential between 900p/1080p that the Switch will likely target and the 4k target of next gen games as well as dumping any ray-tracing features to save on performance. The games that won't be able to come over will be the ones that use the CPU so heavily that there isn't anything they can pull back on and the games that absolutely HAVE to have the SSDs that will be in the PS5 and Xbox Two to function.
Hard to know, maybe Nvidia can impliment hardware based ray tracing into a Tegra chip more easily than we know.
I think Nintendo realizes the Tegra chip they had now wasn't actually that far off from being able to handle current gen games pretty well, it was the heat issue and the inefficiency of the chip that way that really hamstrung them.
If they had used Mariko from the start, likely you would see a good number more PS4/XB1 ports on the current Switch.
For Switch 2 I think they will make that a bigger priority. Switch is a hit concept and developers like it, they can push a little harder to ensure they can get a bigger slice of the current gen software pie, especially those bigger Japanese games I think they want those. They're going to want to retain Dragon Quest XII and be able to run future Monster Hunter and Final Fantasy games I think.
Agree with you on the resolution thing ... for PS5/XB2 tier ports, even 600p undocked and 720-900p docked would be acceptable, and then maybe you can add the option of a 4K dock for higher docked resolution. That's fair.
The issue with hardware accelerated ray-tracing is that it isn't a magic bullet, it still has an impact on performance. I don't think Nintendo will see the performance, heat and power draw impact of it to be worth it for Switch 2, but by its successor the technology will likely be in a good enough place to implement. Also not including it will actually help keep the gap between it and the PS5/XboxTwo smaller since that will instantly be some performance savings when porting games down.
For Nintendo I think if they can keep
Dragon Quest XII, NBA 2K, FIFA, DOOM, Wolfenstein IP and add
Final Fantasy
Kingdom Hearts
Resident Evil
Monster Hunter World
Madden NFL
Call of Duty
Maybe throw in a handful more Western IP ...
For Switch 2, they'd be pretty happy with that. It's not about getting every single 3rd party IP, it's about adding a few more key IP. Especially the bigger Japanese IP, the goal should be to own the Japanese market entirely.
I think another thing they might want to consider for Switch 2 is having an official portable battery pack/stand solution that allows the Switch 2 to run docked performance while portable for 3-4 hours. That would even the scales a lot.
Dragon Quest XII, NBA 2K, FIFA, DOOM, Wolfenstein IP and add
Final Fantasy
Kingdom Hearts
Resident Evil
Monster Hunter World
Madden NFL
Call of Duty
Maybe throw in a handful more Western IP ...
For Switch 2, they'd be pretty happy with that. It's not about getting every single 3rd party IP, it's about adding a few more key IP. Especially the bigger Japanese IP, the goal should be to own the Japanese market entirely.
I think another thing they might want to consider for Switch 2 is having an official portable battery pack/stand solution that allows the Switch 2 to run docked performance while portable for 3-4 hours. That would even the scales a lot.
well ray racing on mobile devices aren't all that new, and with the shrink to 7nm, they could possibly fit more RT cores on the die
Nintendo *might* introduce a slight speed-bumped version of the switch that smoothes performance somewhat, but I don't expect a Switch Pro or anything like that for another 5-7 years.
Nintendo's strategy has always been to milk their portable gaming hardware for as long as humanly possible, in order to take advantage of the massive install base. I can't see them breaking compatibility, and I sure as hell don't see them trying to get developers on-board with an "enhanced on Pro" approach. Too complicated.
Instead they'll just continue to drive the price down, and they'll develop games that work well within the technological boundaries of their HW, like they always have. It's fine.
Nintendo's strategy has always been to milk their portable gaming hardware for as long as humanly possible, in order to take advantage of the massive install base. I can't see them breaking compatibility, and I sure as hell don't see them trying to get developers on-board with an "enhanced on Pro" approach. Too complicated.
Instead they'll just continue to drive the price down, and they'll develop games that work well within the technological boundaries of their HW, like they always have. It's fine.
Oh shit. Things just got interesting... Nvidia is releasing a smaller version of Xavier called the Nvidia Jetson Xavier NX, next year on March 2020! It can run 10-15 watts and it's gonna be be using a 128 bit 8GB RAM running at 51 GB/s for the bandwidth. It is using Carmel cores for the CPU. GPU is using 384 cores and something else (I have no idea what that means in power). It's going to cost $400.
So now I wonder...
1. What chip node could it possibly be using to get 10-15 watts? Will it use the 7nm+ EUV Samsung chip?
2. How does it compare to regular Nvidia Xavier in power?
3. How large is the die?
3. If it could be on a switch pro next year...?
So now I wonder...
1. What chip node could it possibly be using to get 10-15 watts? Will it use the 7nm+ EUV Samsung chip?
2. How does it compare to regular Nvidia Xavier in power?
3. How large is the die?
3. If it could be on a switch pro next year...?
What do you think?
Last edited:
- 12nm
- cpu, hard to say since nothing consumer uses Caramel and Nvidia killed it. gpu, probably not much more than Maxwell. there are 128 more cores, so there's that. but Volta never made it to consumer hardware because it was geared for AI
- who knows
- no
Just saw the wiki page.. 845 GFLOPs for GPU. Ufff. They pulled a Tegra nano on this. Still interesting though as it offers more than 2x the power of OG switch. Could be as powerful as a base xbone or at least within 90% of it. And mixed precision mode should easily surpass base xbone if it's used.
en.m.wikipedia.org
The TX2 was also used for A.I. but then it ended up being produced on the Magic VR set. Of course this doesn't mean it will be used on Switch, especially with the Carmel cores
Tegra - Wikipedia
The TX2 was also used for A.I. but then it ended up being produced on the Magic VR set. Of course this doesn't mean it will be used on Switch, especially with the Carmel cores
the fact they never put it in the new shield tells you enough about this product. it's not meant for devices like the SwitchJust saw the wiki page.. 845 GFLOPs for GPU. Ufff. They pulled a Tegra nano on this. Still interesting though as it offers more than 2x the power of OG switch. Could be as powerful as a base xbone or at least within 90% of it. And mixed precision mode should easily surpass base xbone if it's used.
Tegra - Wikipedia
The TX2 was also used for A.I. but then it ended up being produced on the Magic VR set. Of course this doesn't mean it will be used on Switch, especially with the Carmel cores
That's likely because of the Denver cores. I never said would ever be on the switch either. I did say it was unlikely with Carmel cores. Updated TX1 is identical to TX2 in architecture minus double the Ram and buswidth, and some changes in cache.
Maxwell Tegra was a precursor to Pascal. You would t see much improvement there.
Knowing Nintendo, we would be lucky if Switch 2 GPU did 1 TFlop in docked mode, and probably no way in hell would it do that in portable mode.
It sucks that Nintendo will probably never again have another pure home console that is comparable to PlayStation and Xbox.Or anything that is really impressive as far as hardware, other than more gimmicks.
Nintendo will always have their first party games, but we won't get another Super Famicom/SNES, N64 or Gamecube. I would love to be proven wrong by the mid 2020s....
It sucks that Nintendo will probably never again have another pure home console that is comparable to PlayStation and Xbox.Or anything that is really impressive as far as hardware, other than more gimmicks.
Nintendo will always have their first party games, but we won't get another Super Famicom/SNES, N64 or Gamecube. I would love to be proven wrong by the mid 2020s....
Last edited:
Nintendo already experienced the sweet smell of third party support with games like Doom, The Witcher, Mortal Kombat, or upcoming games like The Outer Worlds on Switch. I'm sure this will encourage Nintendo to deliver the most powerful portable they can make at that time, of course won't be in the same league as Xbox and Sony, but I can see a big step over the actual console, enough to have a decent third party support.
This Nvidia chip will be used for mobile though. Interesting to see what it's capable of still. At 845 GFLOPs and double the RAM and bandwidth, it's 2x the power of the current Switch and on the same node, albeit takes 15 watts, but it's good. It's a gutted down Nvidia Xavier.
Still fascinating to see what it's capable of with the improved Volta architecture. Carmel cores could rule it out completely of ever being used, but next gen ampere is likely to use more CPU friendly architecture (it's a given to use on gaming), and would be cool to see on mobile...
We'll get it. The technology is already there to surpass xbone/ps4 specs if you count the latest iphones and ipad pros. With Nvidia Xavier's NX TFLOPs numbers of 845 GFLOPs, we're within 90-95% of xbone base GPU power in raw power after counting the newer and more superior nvidia tech efficiency vs xbone/ps4 AMD tech. With mixed precision mode, we'll actually surpass it.
with 7nm + EUV node, I wouldn't be too surprised if we can get the full 1.4 TFLOPs GPU on this chip actually. and that should easily surpass ps4 base specs. Ampere is a successor to Turing, with the latter being more efficient than the tech in Xavier and it is going to be significantly more powerful. A handheld ampere in 2021-23 on a hybrid should pass ps4 without too much effort.
I'd like a console based system to match them too.. I really do. I play docked 99.999% of the time.
Theoretically, they can "shrink" Xavier or Orin onto one of the more advanced production lines and get its power into mobile envelope without rebuilding the SoC itself.
But that would be a wasted opportunity for everyone.
1. 12nm FFN
2. Only 384 out of 512 shader units active, lower clocked CPUs
3. 350mm² (so 3x as big as Switch's chip), it is the same chip as standard Xavier
4. No, makes no sense economically
Even if it would be feasible (which it from my POV it is not), it only would give 2x the performance of the original Switch, which I feel is a too small step
Yep. I think in order to appeal to the current fan base and to keep Switch sales up we will likely get a Switch+ that is more akin to a Switch 2 in many ways within a couple years that will replace the standard model. Switch Lite will remain the budget console.
But their goal is another revolution in gaming. Which likely means VR and super duper controllers. That will be the real next-gen Nintendo system. Whichever Switch model comes next is likely more of a stop-gap. They want to sell as many as they can.
I can see them trying to balance between VR and standard games, but it will be difficult.
I wonder if you could get ~1024 or more Turing cores on a 5nm EUV die (or better process if available) in 2022-2023 or so in a cost effective way. When docked I'd want it to clock even to a higher degree than the current switch proportionally - like, the difference between docked and handheld clocks would be larger than on the current switch (esp when HH clocks have been increasing while docked has been stable). Maybe even with some kind of active or semi-active cooling in the dock itself to really boost it up more.
As discussed prior, increasing bandwidth proportionally is kind of a complicated problem though.
As discussed prior, increasing bandwidth proportionally is kind of a complicated problem though.
1,2,3, and 4 have been covered before your post. Where did you get 350mm^2? Sounds like the size of the original. This smaller Xavier chip is the same size as the Jetson Nano.
Because it will break API compatibility and they need to start from 0 again.
That's not for certain. How did MS get BC working for 3 different generations of consoles.