Switch 2 Speculation

[OlegSH]

VRS uses tensor core?

VRS has nothing to do with Tensor Cores. GTX Turing GPUs support VRS as well as other non ray-tracing features.

 

[OlegSH]

There's better power/performance hardware with ray tracing from a formerly british company...

Better power/performance based on what?
I thought imagination just walked in the completely opposite direction with the 128-wide warp in their latest architecture, sounds to me like a no go for RT and modern consoles workloads. Even AMD has finally moved to 32 wide warp with RDNA.

 

[Rootax]

Better power/performance based on what?
I thought imagination just walked in the completely opposite direction with the 128-wide warp in their latest architecture, sounds to me like a no go for RT and modern consoles workloads. Even AMD has finally moved to 32 wide warp with RDNA.

Maybe he had GR6500 in mind ?

 

[Goodtwin]

Doesn't the RSX include northbridge functionality?
Regardless, at that time the nForce efforts were still ongoing.


Of course it wasn't. Do we really need to come back to this every other month?
The Snapdragon 820 that was ready in 2015 would have run circles around the Tegra X1 with a sub-500MHz GPU. During 2015 even Samsung's Exynos 8890 would have been available for sampling to partners, and it too would have provided much better performance at the same power consumption. Either company could have been able to develop a radio-less and ISP-less solution with CPUs optimized for lower frequencies for Nintendo had they paid enough money for it, or either company would have gladly sold Nintendo their high-volume SoCs.

Do you really need to make these broad generalizations about Tegra X1 being OMG the best possible SoC for 2017 handheld release when it's very obviously and factually false? During all the ~8 years they were active in the mobile SoC market, nVidia never hit any power/performance jackpot. Their inability to compete is the reason they left that market.

What nvidia did provide to Nintendo (and the others could not) was a vertical stack for software development and optimization, an already existing development team that was specializing in porting PC games to ARM SoCs and a close presence next to 3rd party developers.
I.e. what nvidia could provide in the overall package was money savings in software development / optimization efforts plus a sizeable discount on an objecticely failed existing SoC.
nVidia did not a better hardware solution than others could, for a handheld console. The benchmarks exist, the power consumption comparisons exist and everything was made by multiple outlets and experts. Let's stop pretending they don't.

https://www.slashgear.com/qualcomm-snapdragon-820-gfx-benchmarked-against-tegra-apple-10426267/

The Tegra X1 looks to me like its on par with the Snapdragon 820 in the graphics benchmark. Certainly not a "runs circles around the Tegra X1" scenario.

Keep in mind I said one of the best solutions, not that it was the absolute best performer on the market. There are things to consider beyond just performance and energy consumption. The X1 was a relatively inexpensive option. I have to believe the 14nm FinFet Snapdragon in 820 was going to bring a higher price tag than Nintendo paid for the X1. We have to keep in mind that Nintendo was building a $300 product. Just because there were better SOC's out there doesnt mean much if they were in $600+ cell phones and tablets. If Nintendo hadn't partnered with Nvidia, they would not have gotten the development environment that we have now. So even if they had gotten a bit more performance from another chip, that could easily be squandered by being less developer friendly. It was the whole package that made the X1 a great solution. Performance, price and software support are the reasons it made so much sense.

As for the Switch successor, Nvidia may not have all the favorables that they did with the X1. X1 was an off the shelf chip that I am sure was sold at low margins because of its inability to penetrate the mobile market. Nvidia wont have an off the shelf product to sell, so Nintendo would have to flip the bill for a customer processor. It will depend on just how much Nintendo values the partnership with Nvidia.

Switch paved the way for what to expect out of their hybrid setup. If Switch were to release in 2022, I would expect a performance target to be one that surpasses base PS4 performance, opening up the door for previous gen ports. Nintendo's exclusives will continue to be the driving force, but Switch has shown plenty of enthusiasm for playing 360/PS3 era games on the go.

 

[Karamazov]

those benchmarks have to deal with android OS and other apps always opened, guess the swtich could bench higher

 

[Esrever]

https://www.slashgear.com/qualcomm-snapdragon-820-gfx-benchmarked-against-tegra-apple-10426267/

The Tegra X1 looks to me like its on par with the Snapdragon 820 in the graphics benchmark. Certainly not a "runs circles around the Tegra X1" scenario.

Thats only when the Tegra was probably using twice the power as the snapdragon 820. In the switch, the X1 is running at way below the the clocks of the normal X1 even in dock mode. In portable mode, it runs at 1/3rd the speed. If the switch came with a snapdragon 820, it could probably clock near full speed in mobile mode and probably OC in docked mode. That by itself would probably double the performance than what they ended up with and probably still increased battery life.

The main advantage of the X1 was probably how cheap it was for Nintendo to buy and the support Nvidia threw into the contract. But this is more of a deficiency in Nintendo's ability to make a modern development stack than anything to do with the Tegra. Nintendo's terrible dev tooling and hardware support is what made the Tegra attractive, not that the Tegra was very good to begin with.

 

[Deleted member 13524]

The Tegra X1 looks to me like its on par with the Snapdragon 820 in the graphics benchmark.

That's using the Pixel C, which AFAIK had the GPU running at 850MHz and the CPU at 1.9GHz.

You do the math on how much performance the Switch can do in comparison, especially in mobile mode.

those benchmarks have to deal with android OS and other apps always opened, guess the swtich could bench higher

And so would any other SoC without having to run Android services..?

 

[Benetanegia]

Thats only when the Tegra was probably using twice the power as the snapdragon 820. In the switch, the X1 is running at way below the the clocks of the normal X1 even in dock mode. In portable mode, it runs at 1/3rd the speed. If the switch came with a snapdragon 820, it could probably clock near full speed in mobile mode and probably OC in docked mode. That by itself would probably double the performance than what they ended up with and probably still increased battery life.

I don't believe that to be entirely true honestly. The X1 as seen in the Pixel C is far from its "normal clocks" that you're probably talking about.

Performance of the Pixel C is more than 20% lower than that of the Android TV, so it stands to reason that the clocks are also below 800mhz. And it wouldn't surprise me if clocks were even lower than that based on how most GPUs tend to offer a better perf/clock ration when "underclocked".

https://gfxbench.com/device.jsp?ben...(R) Tegra(R) X1&did=28156617&D=Google Pixel C
https://gfxbench.com/device.jsp?ben...R) X1&did=26084812&D=NVIDIA Shield Android TV

 

[mpg1]

It probably makes sense to launch in the same year as Sony and Microsoft. As long as they stay within a certain range power wise they have the chance to get the occasional port like they do with the existing Switch.

 

[Entropy]

It probably makes sense to launch in the same year as Sony and Microsoft. As long as they stay within a certain range power wise they have the chance to get the occasional port like they do with the existing Switch.

What they need to do is to communicate with the major publishers about their product plans. If they say that they will produce a new switch like device to be launched two years after PSXBox, with approximate capabilities, the publishers can have that in mind for their own product plans. If they go with a hybrid approach again, it is likely to be able to run anything the PS4/XB1 can, so generation crossing titles (for the tethered consoles) would fit beautifully.

The Switch sells well. The publishers would be foolish not to include a future version in their product plans.

 

[AlNom]

If they go with a hybrid approach again, it is likely to be able to run anything the PS4/XB1 can, so generation crossing titles (for the tethered consoles) would fit beautifully.

That's actually an interesting point with regards to the discussions of cross-gen multiplatform development, but I digress.

*cough*

 

[Svensk Viking]

I've played with the thought if Nintendo would release a Switch Pro and treat it like the Gameboy Colour, for a time releasing games that work on both systems, but later making them Pro only

 

[Goodtwin]

The main advantage of the X1 was probably how cheap it was for Nintendo to buy and the support Nvidia threw into the contract. But this is more of a deficiency in Nintendo's ability to make a modern development stack than anything to do with the Tegra. Nintendo's terrible dev tooling and hardware support is what made the Tegra attractive, not that the Tegra was very good to begin with.

That is precisely the basis for why I have said the Tegra X1 was the perfect choice for Nintendo. It might not be flattering to Nintendo, but they needed to make big improvements to their development tools, and Nvidia was able to offer that. They needed a SOC that would fit the budget of a $300 retail product that fit the performance profile they were looking for, and the X1 fit the bill. The X1 was pretty power hungry, but how much of that can be chalked up to the 20nm process? The new X1 Mariko version at 16nm FinFet is significantly less power hungry, offering nearly twice the battery life as the original Switch model. If Nintendo has used this for higher clock speeds instead of better battery life, the X1 easily matches or exceeds the Snapdragon 820, and its still on a larger process than the 820 was back in 2016. So to say the Tegra X1 as a design was bad is being pretty harsh, as we can now see it really just needed to move to the 16nm FinFet process to reach its potential.

 

[Frenetic Pony]

That is precisely the basis for why I have said the Tegra X1 was the perfect choice for Nintendo. It might not be flattering to Nintendo, but they needed to make big improvements to their development tools, and Nvidia was able to offer that. They needed a SOC that would fit the budget of a $300 retail product that fit the performance profile they were looking for, and the X1 fit the bill. The X1 was pretty power hungry, but how much of that can be chalked up to the 20nm process? The new X1 Mariko version at 16nm FinFet is significantly less power hungry, offering nearly twice the battery life as the original Switch model. If Nintendo has used this for higher clock speeds instead of better battery life, the X1 easily matches or exceeds the Snapdragon 820, and its still on a larger process than the 820 was back in 2016. So to say the Tegra X1 as a design was bad is being pretty harsh, as we can now see it really just needed to move to the 16nm FinFet process to reach its potential.

Would have to agree, the Switch SOC was probably a convenience for Nintendo at the time. A convenience that might be coming back to bite them, Nvidia doesn't seem interested in continuing its SOC business as such and is notoriously fickle about working with vendors for consoles (why both Sony and MS haven't used them for 2 and 3 generations respectively). But backwards compatibility will probably be much hoped for with a Super Switch or whatever it's called circa 2021/22.

I'd also expect a "PS4+" kind of target, while not necessarily what Nintendo will go for, to be an incredibly useful target. I'm still shocked MS didn't go for it with their next console. Sony hit it big with the PS4, so "more of the same" is a common enough tactic. But "most" games made with a kind of dualshock controller/accomplishable today could easily be run on a "PS4 with somewhat more ram/faster CPU", thus the specs of the PS5/Xs are just plain excessive for actually getting 95%+ games that are going to be on them to run. I'd figure the vast, vast, vast majority of customers would be satisfied with lower visual settings and less pixels if the tradeoff is "it runs on a mobile tablet I carry around". Nintendo has the opportunity for "best selling console of all time" laying right in front of them, will be interesting to see if they can hit it.

 

[DavidGraham]

According to a rumor circulating online, a new Nintendo Switch model could be released in the final quarter of the year. This new model will not adopt the new TegraX1+, but new custom processors Nintendo is developing with NVIDIA. The GPU will be based on the Volta architecture.

As the rumor comes from unspecified sources, we should take it with a grain of salt. The leaker, however, got a good reputation, as he correctly revealed information on mobile products before, so there may be some truth in what has been revealed today.

https://wccftech.com/nintendo-switch-pro-gpu-volta-4k/

 

[stiftl]

The Volta GPU part of Xavier (512 CUDA cores) has a die size of 89mm² (around 1/4 of the complete chip), at 7nm this should come out at 50-55mm².
The real problem with Xavier is the huge amount of unneccessary transistors for a gaming specific application, also the Carmel CPU cores are fucking HUGE (62mm² compared to ~10mm² in modern 8 core mobile chips like A76/A55 combos). The only solution would be a custom chip, for example with stock Cortex-A75 or A76 cores and without all the PVA / Deep-Learning portions of Xavier, this should also come out around the size of the current Tegra X1, at 7nm though.

 

[Goodtwin]

https://wccftech.com/nintendo-switch-pro-gpu-volta-4k/

I don't see a scenario where the Switch Pro would use anything other than the newer Tegra X1 Mariko chip. It has plenty of thermal head room to turn up the clock speeds by a significant margin. Newer LPDDR memory will bring a modest boost to memory bandwidth. We have seen from hackers overclocking their standard Switch units that they can get some signficant performance advancements from overlocking the CPU to 1.35Ghz, and a decent boost from overclocking the GPU to 1Ghz. The newer Mariko chip could go higher still.

As for Switch's true successor, is anyone aware of a mobile processor that has surpassed the Xbox One in performance yet? I feel like Nintendo will want to hold off until they can surpass the Xbox One and PS4 in the form factor of a Switch.

 

[Deleted member 13524]

As for Switch's true successor, is anyone aware of a mobile processor that has surpassed the Xbox One in performance yet? I feel like Nintendo will want to hold off until they can surpass the Xbox One and PS4 in the form factor of a Switch.

Contrary to popular belief (among TX1 fans), Qualcomm does semicustom solutions, and the Snapdragon SQ1 for the Surface Pro X has 8-channel LPDDR4X with 68GB/s, a 3GHz 4-Core Cortex A76 + 1.8GHz 4-Core Cortex A55 and GPU with 1.05 TFLOPs FP32 / 2.1 TFLOPs FP16 with full Vulkan 1.1 compliance.

The ~7W SQ1 is already really close to XBone, and despite not having eSRAM I'm guessing they could show very similar real-life performance, considering the FP16 throughput advantage and the considerably better CPU.

My guess is next-gen 5nm snapdragon for Windows (late 2020 / early 2021?) might be head and shoulders above the XBone. LPDDR5 would drive the same 8 channels towards 102GB/s, and getting a ~30% increase in GPU performance isn't out of the ordinary.

But if Qualcomm were to design a handheld gaming console SoC, they'd strip the Hexagon DSP, Spectra ISP and LTE Modem. Then they'd probably remove the Cortex A55 module and use 4 or 8 Cortex A76 cores, using density-optimized transistors because they wouldn't need more than 2GHz out of those. Take the left over power and area and implement a wider and lower clocked GPU, and they could probably do something more powerful than XBone right now, on 7nm.

 

[milk]

I think one strength that was on Switch's side was that despite its large preformance diferential to the bone and ps4, it had feature parity with them. With that in mind, I think it would be highly desireable for switch 2 to have a feature-set that matches ps4 and xbsx, even if the brute horsepower is more like that of the base one.

 

[Esrever]

But if Qualcomm were to design a handheld gaming console SoC, they'd strip the Hexagon DSP, Spectra ISP and LTE Modem.

If it wasn't a nintendo console, I'd think they would want to keep these. LTE tablets and laptops are already very wide spread. No reason a handheld can't have it as a feature.