Switch 2 Speculation

[orangpelupa]

Curious, does anyone have performance comparison between the original tx1, New tx1, and various snapdragons?

I only found notebookcheck's charts but it doesn't have many GPU bench. It does have tons of CPU bench tho.

 

[Sigfried1977]

If I were betting on a Switch 2, I'd guess it would be a max of 2X perf over the current Switch. A72 x 4 + double the GPU performance on a cheap 7nm process around ~2022. It would be modest boost in performance and in line with what they usually deliver.

A 2x jump is only in line with the jump from the GC to the Wii really. And if we're looking at the Switch as a portables, the jumps between those were always much, much bigger. GB to GBA to DS to 3DS and now the Switch - those certainly aren't times-two steps.

The Wii was an exception so far. Not the rule.

 

[Deleted member 13524]

IIRC - the licensing agreement only prohibits products including AMD IP that would directly compete in markets that AMD is currently in. That basically opens up the mobile/portable gaming market avenue for Samsung using AMD GPU IP.

Yes. Samsung can do whatever SoCs they want with RDNA GPUs as long as it's lower than 15W.

Which fits rather nicely with AMD declaring they'd stay out of the ultra-mobile business, but still finding a way to put their GPU architecture into handhelds.

A 2x jump is only in line with the jump from the GC to the Wii really.

IIRC the Wii is exactly a Gamecube with 50% higher clocks and an external 64MB single-channel GDDR3 chip for low-priority stuff like audio, home screen, social stuff, user interface and texture buffering.
Unless the gamecube was severely bottlenecked by memory amount, I doubt the Wii ever reached 2x its performance.

But to be fair, in the handheld space Nintendo has had mid-gen upgrades for quite some time.
The DSi had a 2x faster CPU and 4x more RAM than the DS, and the New 3DS had twice the amount of ARM11 cores clocked 3x higher (6x more powerful CPU!) and twice the FCRAM at twice the bandwidth (they didn't touch the DMP GPU though).
So if Mariko lets the A57 cores clock at ~2GHz without burning too much power, I think Nintendo might release a New Switch with that enabled, as it speeds up UX and other QoL enhancements.

 

[Mihailjones]

Curious, does anyone have performance comparison between the original tx1, New tx1, and various snapdragons?

I only found notebookcheck's charts but it doesn't have many GPU bench. It does have tons of CPU bench tho.

That would be interesting to see, I have seen some benchmarks but havent found one with multiple SoCs.

What I know is that even 2016 Snapdragons&other high end mobile chips are like 2-3x faster on Cpu side and gpu is about the same as tegra.

So estimating from that I would guess that Snapdragon 855 + other high end models must be 6-10x faster on cpu side and also multiple times times faster on gpu.

Nintendo didnt choose tegra because it were the best, they took it it because it were cheapest option with bare minium specs i guess.

But from switch 2 I wont hope much, probably another weak machine with high price tag as nintendo does want to maximise profits.

"Vita 2" could be epic device, if Sony wanted to make one. 3-5x switch perf = no stupid compromises and ancient low res games like on the switch

 

[Sigfried1977]

But to be fair, in the handheld space Nintendo has had mid-gen upgrades for quite some time.
The DSi had a 2x faster CPU and 4x more RAM than the DS, and the New 3DS had twice the amount of ARM11 cores clocked 3x higher (6x more powerful CPU!) and twice the FCRAM at twice the bandwidth (they didn't touch the DMP GPU though).
So if Mariko lets the A57 cores clock at ~2GHz without burning too much power, I think Nintendo might release a New Switch with that enabled, as it speeds up UX and other QoL enhancements.

Oh sure. Too bad they never really did anything with it. I think the new3DS has a total of 2 exclusive games to its name as well as a handful ones which ran a little better. And I think the sole purpose of the DSi was that it had wifi in order to connect to the eshop.

 

[Entropy]

Nintendo is interesting because they are difficult to predict.
If I had been Nintendo, I would have informed the major publishers "we are going to introduce a Switch 2 around late 2022, and it will offer capabilities roughly on par with the PS4" (but with faster secondary storage (UFS 3) and CPUs).
That would ensure that any games in development that would like to access the customer base of a Switch 2, would know roughly to what point it would have to scale down. Which would also happen to be the level to which it would have to scale down to target the current generation of stationary consoles.
It would be doable piggybacking on mobile SoC development on TSMC 5nm.

 

[Ika]

Xavier is a 30W SoC

IIRC Xavier is not a 30W SOC, that's only the max configurable TDP.

 

[Entropy]

IIRC Xavier is not a 30W SOC, that's only the max configurable TDP.

Xavier is 350mm2 on TSMC 12nm. It could be adapted to, say, TSMC 5nm, and probably do the job for Nintendo at a reasonable die size.
The question is whether it would make sense for the parties involved.

 

[Rootax]

if devs are happy with the nvidia dev tools / language, It's a strong point for them (if they are still interested doing business with Nintendo of course)

 

[Ika]

Xavier is 350mm2 on TSMC 12nm. It could be adapted to, say, TSMC 5nm, and probably do the job for Nintendo at a reasonable die size.
The question is whether it would make sense for the parties involved.

I didn't say that Xavier would be the best SOC to go with, I was just pointing out that the 30W claim was missleading.
I agree, the bussnies side of such deal is questionable. Nvidia makes the best graphics chips (for a long time now), but they are expensive, so perhaps the profit they could make (after going 5nm with the SOC) simply wouldn't be enough.

 

[Deleted member 13524]

Why is the 30w claim misleading? Isn't the Xavier in Jetson AGX and Drive PX set to 30W TDP by default?

What else would you expect of a 8 big-core CPU with 512sp GPU, dedicated sensor cores and 256bit MCU, 350mm^2 big with 9 billion transistors?

 

[Goodtwin]

Xavier is 350mm2 on TSMC 12nm. It could be adapted to, say, TSMC 5nm, and probably do the job for Nintendo at a reasonable die size.
The question is whether it would make sense for the parties involved.

I know it could be all PR talk, but they did make comments that suggested the Nvidia partnership was intended to last for a long time. If the die size of the Xavier cold be reduced enough thanks to a newer smaller process, then perhaps its a candidate for the next Switch. There is certainly a lot of extra fluff hardware on the Xavier chip that would go unused, but that was also a thing with the Tegra X1 chip. Having Nvidia's help with development tools has been a huge benefit to game development on Switch. If the tools were as poor as they had been in the past, there is no way we would have seen near the third party support that has come on the Switch.

I think Switch has shown that it doesn't need, nor will it have parity with other consoles when it comes to third party games. It will thrive on exclusive content, and ports of previous gen games, with some current gen games sprinkled in for those developers who are up for the challenge.

 

[AlNom]

https://devblogs.nvidia.com/jetson-xavier-nx-the-worlds-smallest-ai-supercomputer/

hm..................

Jetson Xavier NX achieves up to 15X higher performance than Jetson TX2, with the same power and in a 25% smaller footprint.
---
Jetson Xavier NX defines default power modes for 10 and 15W.
---
The CPU is arranged in three clusters of 2 cores each, and has a maximum operating frequency of 1400MHz in 4/6-core mode and up to 1900MHz in dual-core mode for applications that may require greater single-threaded vs. multi-threaded performance.

 

[swaaye]

Aren't the newer Tegras loaded with automotive functionality? ie lots of useless transistors for a game machine resulting in a less efficient and more expensive die.

Though Nintendo is of course frugal! Seems like the Switch X1 shrink was partly subsidized by letting Nvidia use it in a new Shield TV too.

 

[Deleted member 11852]

Aren't the newer Tegras loaded with automotive functionality? ie lots of useless transistors for a game machine resulting in a less efficient and more expensive die.

Economies of scale come into play. Some larger/more-complex semiconductors produced in volumes of tens of millions per month can be cheaper than something smaller/less-complex produced in volumes of hundreds of thousands or single-digit millions.

 

[orangpelupa]

Economies of scale come into play. Some larger/more-complex semiconductors produced in volumes of tens of millions per month can be cheaper than something smaller/less-complex produced in volumes of hundreds of thousands of single-digit millions.

And if those can be easily disabled for Nintendo batches, maybe it gobbles less electricity?

Anyway, it would be funny (awesome?) if Nintendo use the full fledge SoC, and actually used the automotive stuff for a new game. Something new, currently unthinkable, something genuinely Nintendo

 

[Deleted member 13524]

Though Nintendo is of course frugal! Seems like the Switch X1 shrink was partly subsidized by letting Nvidia use it in a new Shield TV too.

TSMC might be planning to drop 20nm alltogether too, as they retrofit the 20nm fabs to support finfet. That could have been an additional factor too.

Economies of scale come into play. Some larger/more-complex semiconductors produced in volumes of tens of millions per month can be cheaper than something smaller/less-complex produced in volumes of hundreds of thousands or single-digit millions.

I doubt there's any economy of scale that justifies using a 350mm^2 big 9B transistor chip on a handheld just to use some 30-40% of what's in the SoC.

 

[DuckThor Evil]

I could see this NX chip being in a hypothetical Switch Pro, not in a Switch 2. Switch Pro would be lower volume higher priced SKU like the PS4 Pro and it could house a more expensive chip and here we have a chip that already exists. For a high volume and more price sensitive market where the Switch 2 will be, this is quite far from ideal.

 

[stiftl]

^ This chip is 3x as big as the one in the original Switch, no way this will be used.
I really only could see a custom chip for a successor, 8 core A77 (stock or nV derivate) and a 768 shader units GPU could be realistically done on 7nm EUV on roughly the same die size. Together with a 128bit LPDDR5 memory interface this could reach PS4 performance levels docked / XBone undocked while having a much more powerful CPU part (important for next-gen ports).

 

[AlNom]

The marketing implies a footprint less than the Tegra X2, I think, while being pin compatible with the Jetson Nano. My impression was a new SKU on 7nm.

so..... ¯\_(ツ)_/¯

Or I’m crazy :V

—-

Reading aside, the blog mentions availability in Spring 2020.

Given the cost of 7nm and the impending glut of next gen volumes and the recent revision to Switch, I’d hazard a 2021 launch regardless of what the design is.