Is the Switch 2 Getting ORIN (2022) or ATLAS (2025) Nvidia Tegra chip for a 2025 release?

[Pemalite]

Leynos said:

They better stick with Nvidia so all my cart games work in BC.

There is no guarantee of that even if they did stick with nVidia.

Going from Maxwell to Ampere is a significant one.
Going from ARM A57 to ARM A78 is also a night and day difference.

There is a reason why Nintendo didn't opt for Tegra X2 at 16nm for the chip shrink, but rather got nVidia to do it to the older Tegra X1... And they are almost identical chips except on the GPU side of the equation (which are almost identical being Maxwell vs Pascal.)
Nintendo is using some extremely low level abstraction to get as much out of the Switch as possible... And some of the games are absolutely showing that, but it does potentially bring issues bringing that code forwards to newer hardware.

Going from Maxwell to Ampere will mean a new GPU instruction set, just like how AMD went from VLIW to GCN to RDNA was all new instruction sets every time. (With extra extensions/variations with parts like Vega and Rapid Packed Math)
But GPU's are abstracted in such a way on the PC where it's not an issue... On a console, that has not always historically been the case. - As games are sometimes built to the metal.

Sony and Microsoft have migrated to monolithic software stacks these days where they can side-step those issues (I.E. They learned a few PC tricks), plus they have a few hooks in the software stacks so they can emulate/abstract certain functions depending on the code that is being run. - Not saying Nintendo won't do this approach, but they tend not to have a very flexible OS in the Switch, it's pretty bare bones.

eddy7eddy said:

I wish Apple's M Series would be an option, would be a monster console if you develop games specifically for that chip.

I kind of remember a partnership between Nintendo and Nvidia for 10 years, what I remember is that before Switch release, Nvidia Expected a 20-Year relationship with Nintendo, I guess they are expecting 3 generations of Switchs with Nintendo. (Switch, Switch U & Switch One)

But true, I can see the Next console beating PS4, getting results closer to PS4 Pro, thanks to the new arquitectures and developed capacities for the new generation processors.

And TFlops...

Apple M would be a brilliant chip, expensive, but brilliant.

Don't try and bait me with irrelevant TFlops. :P

Alistair said:

Orin is ready now. Can launch any time now, the economic slowdown means nothing is in short supply now. 5nm is available, ram is available, everything is becoming abundant.

Also Nintendo could still just release a 5nm revision of the original Tegra.... fanless design.

There is still a semiconductor shortage.

eddy7eddy said:

Doctor_MG said:

Nintendo wants to ensure that they transition well into the next generation. Your suggestion is not the way to do that. That's way too sudden. 

True, but maybe Alistair didn't mean about Next Gen? We can get a Switch V3 with Orin, Fanless and an amazing battery life. Or a Pro that I doubt.

If we get a better Switch with Odin, I'm pretty sure that Atlas would be the Next Gen, but Odin architecture could be a problem, so maybe a Tegra X3 7nm.

They *could* take Tegra X2 and increase the Switch's performance by 50% with relative ease... It would also be built at the same 16nm process as the current Tegra, so the logistics of competing on a more constrained process is not going to be difficult as they have an idea of quantities they can obtain as-is.

h2ohno said:

What are the specs of the Atlas chip? Have those even been released?

Looking at the comparisons between Orin and the Tegra X1 Maxwell I'm sure they'll go with Orin. It's 7-8 times more powerful and that is a generational leap for a console. If they are smart they'll repeat the hardware decisions they made with the Tegra X1 and slightly underclock it for battery life while adding more RAM than the chip normally comes with. If they release the next-gen Switch at any point between 2023 and 2025 that's almost certainly what they'll do. If they wait too long it'll be too close to the release of the PS6 and XB5.

I think they're currently aiming for a March 2024 release, giving the OG Switch a full 7 years before being replaced and also releasing 3.5 years into the PS5's life-cycle like they did with the PS4. Now that the Switch has passed its peak it's time to start seriously preparing for the next generation, but Switch is still selling too well to replace this fiscal year. That may not give Switch enough time to catch up to the DS and PS2, but Nintendo is more interested in profits than sales records.

Atlan. And nVidia revealed it last year at a GPU technology conference.
We know it's going to use the new ARM A715 cores, with variable length vector extensions, boosted SIMD.. And will use nVidia's Ada Lovelace GPU architecture.

Other aspects like LPDDR5X support just makes blatant sense for the release time frame... And will likely re-use the same 256bit bus... So probably reach about 256GB/s of memory bandwidth.
But a handheld would likely use a cutback memory bus for battery life concerns, so probably along the lines of 128bit/128GB/s which is perfect for targeting 1080P.

Bofferbrauer2 said:

Over on the AMD side of things, Rembrandt has shown that AMD can now make really high performing APUs on a tight power budget. A low-power spinoff/successor from there could be used both in an upcoming Switch successor and AMD could use them in their ultra-low-power line (which to this day is still on 14/12nm chips, go figure)

A Rembrandt derivative would probably come with less CU, probably 6 or maybe 8, as 12 would need too much power to feed through. Also possible would be less cores, six cores would be better to keep consumption down without bottlenecking, but I'm not sure if this is an option with AMD and Zen 3. Disabling Hyperthreading is the more likely option in my opinion. Even then, CPU clock speeds would probably go down to somewhere between 2-2.3Ghz and also the GPU clock speeds would need to get cut down somewhat, possibly ~1.6Ghz (No idea what the RDNA2 sweet spot is, so this is just wild guessing).

Since the release is still years away, some Zen4/RDNA3 chip could also be very possible. How that one would stack up in terms of power consumption is still unknown so far however, so we'll have to wait and see what will be possible with those chips in a handheld format.

15-35W mark is where modern processors are at their most efficient (Due to fabrication characteristics dictating some design rules oddly enough).
Rembrandt can suck down about 15w and achieve 1.9Ghz on all 8 Core/16 Threads.
At 20w you start to look at 2.5Ghz... And from there efficiency starts to decline.

That is just all the CPU cores loaded.
At 20w you will start to look at 1.8ghz with all 16 threads loaded and the GPU pegged at the same time.

But you can cut that CPU in half and only have 2/3rds of the CPU functional... And you start to get a very potent and energy efficient chip.

For a handheld, you probably dont need anymore than 4x competent CPU cores... Or maybe even 5 if you wish to dedicate one to the OS/Background so you can have voice/video chat.