Switch 2 Speculation

It depends if I'm on the go or not. Some weeks it's mostly mobile then there are other weeks where I keep it locked like 80% of the time I'm playing it. My son also uses his switch the same way. Plays it in the car and in the bus but will play it docked when he arrives home. So maybe that wide range of figures is just that, a wide range where people find different use cases for it at varying levels.
 
There's a rumor that docked performance of Switch 2 is roughly a PS4 Pro/ Series S. The thing I couldn't understand is why limit handheld mode to lower performce instead of giving the choice entirely to the user?. if it's about battery life then you can easily connect your Switch to a power bank or something like that. Could there be another reason why Nintendo is limiting power usage on handheld mode?, like the battery output is not enough do deliver full power?. I really hope that we get the option to play with docked clocks in handheld mode this time around.
 
There's a rumor that docked performance of Switch 2 is roughly a PS4 Pro/ Series S. The thing I couldn't understand is why limit handheld mode to lower performce instead of giving the choice entirely to the user?. if it's about battery life then you can easily connect your Switch to a power bank or something like that. Could there be another reason why Nintendo is limiting power usage on handheld mode?, like the battery output is not enough do deliver full power?. I really hope that we get the option to play with docked clocks in handheld mode this time around.

Give the average consumer a choice like that and they'll turn it to max, run the battery out in half an hour each time, hate the product and tell people not to buy it. People complain about not being able to do things with Apple products; but Apple got to be the world's most valuable brand by not letting customer own dimness disappoint themselves.
 
Give the average consumer a choice like that and they'll turn it to max, run the battery out in half an hour each time, hate the product and tell people not to buy it. People complain about not being able to do things with Apple products; but Apple got to be the world's most valuable brand by not letting customer own dimness disappoint themselves.
I partially agree but they can add it deep in the options then a warning message before you can unlock the full clocks indicating that this mode is not optimal for battery life. This shouldn't be an advertised feature just a fan service option hidden deep in settings.
 
Maybe this time around the dock is designed to help with heat dissipation and it would warm too much in portable mode with full clocks..
 
There's a rumor that docked performance of Switch 2 is roughly a PS4 Pro/ Series S. The thing I couldn't understand is why limit handheld mode to lower performce instead of giving the choice entirely to the user?. if it's about battery life then you can easily connect your Switch to a power bank or something like that. Could there be another reason why Nintendo is limiting power usage on handheld mode?, like the battery output is not enough do deliver full power?. I really hope that we get the option to play with docked clocks in handheld mode this time around.
I can believe it'll be '4TF', but that'll be via 2xFP32 or something and not directly comparable with the 4TF of the PS4 Pro/Series S, plus it'll still be heavily bandwidth constrained. No way will it really perform like those systems' GPUs.

As for Nintendo not giving people the option to run full blast in portable mode, it's cuz it's a console and designed around prescribed experiences. Plus they wouldn't want the negativity of complaints surrounding battery life and the unit being uncomfortably hot to hold and all. So they just take the safe route and lock it down. Wouldn't expect anything otherwise.
 
I feel another consideration would the inevitable temptation for devleopers to end up targeting the "full blast mode" if you could manually toggle it.
 
I can believe it'll be '4TF', but that'll be via 2xFP32 or something and not directly comparable with the 4TF of the PS4 Pro/Series S, plus it'll still be heavily bandwidth constrained. No way will it really perform like those systems' GPUs.

As for Nintendo not giving people the option to run full blast in portable mode, it's cuz it's a console and designed around prescribed experiences. Plus they wouldn't want the negativity of complaints surrounding battery life and the unit being uncomfortably hot to hold and all. So they just take the safe route and lock it down. Wouldn't expect anything otherwise.

100gbs is a match for MX570, rated at 4.7tf at 15w. I'd expect that to be docked mode, comparing to RDNA2, I'd expect this to be just below a Series S for the most part docked. Mobile mode, 3tf? (2/3rds as fast?) Pretty fast, fast enough that anything on a Series S at 60 will run on this at 30 in mobile, maybe. If it's 50gbps that's obviously much lower, but Switch had multiple clockspeeds that were all selectable by devs.

Of course games like Dragon's Dogma 2 are aiming for 30fps on Series S at all (worse it's capped by CPU), so I'd say this isn't positioned terribly off what the Switch was last time, though a bit better. I.E. I'd expect more games from Series S and up to also go onto the Switch 2 this time around than the Switch last time.

Edit- let's compare 3dmark Solar Bay, Exynos 2400 vs this (likely). Exynos scores around 8k, while this is expected to reach 13k, but at a much higher tdp.
 
Last edited:
What's the current consensus on GPU specs?

I was randomly thinking about this, and it feels like NVIDIA would want to optimise for "minimum work" on their end especially physical design, so that would point towards exactly 1/3 AD106 on TSMC N4 (1 GPC, 12 SMs, 1536 cores, 48 texture units, 16 ROPs) which would be 4 teraflops at 1.3GHz docked and that would match rumours quite nicely. But the rumours have been all over the place so that doesn't mean much ofc...
 
I just hope that we don't get miserably low clocks thanks to Samsung 8nm. It would be cool if developers could launch third party games simultaneously on all consoles and PC, including the switch 2.
 
What's the current consensus on GPU specs?

I was randomly thinking about this, and it feels like NVIDIA would want to optimise for "minimum work" on their end especially physical design, so that would point towards exactly 1/3 AD106 on TSMC N4 (1 GPC, 12 SMs, 1536 cores, 48 texture units, 16 ROPs) which would be 4 teraflops at 1.3GHz docked and that would match rumours quite nicely. But the rumours have been all over the place so that doesn't mean much ofc...
Think that's mostly the general consensus but it's 2 GPCs and so 32 ROPs. Frequency I have a feeling it's going to be lower than what a lot of people expect like the Switch's clocks were for power and thermal reasons. I'd be pleasantly surprised if it's >1000MHz docked, I'm in the 900 <= x <=1000MHz +/- a bit here and there although a lot of people disagree on that
 
Think that's mostly the general consensus but it's 2 GPCs and so 32 ROPs. Frequency I have a feeling it's going to be lower than what a lot of people expect like the Switch's clocks were for power and thermal reasons. I'd be pleasantly surprised if it's >1000MHz docked, I'm in the 900 <= x <=1000MHz +/- a bit here and there although a lot of people disagree on that
How many SMs per GPC if it’s 2 GPCs? 8 like AD107 so 16 total? That seems like a lot of SMs to me, while 6 SMs per GPC feels unbalanced with probably a significant area penalty…

Whether 32 ROPs even makes sense perf-wise depends on the expected memory bandwidth and L2 cache size. I would expect 128-bit LPDDR5 for cost reasons (+maybe a 12-16MiB L2, vs 24MiB on AD107, if we’re *very* lucky) which would be ~100GB/s vs ~256GB/s for AD107, so I guess it’s in the range where 16 ROPs is less than gaming Ada but 32 ROPs would be noticeably more relatively speaking.

So I’d still bet on 1 GPC and 16 ROPs personally but hard to say.
 
What's the current consensus on GPU specs?

I was randomly thinking about this, and it feels like NVIDIA would want to optimise for "minimum work" on their end especially physical design, so that would point towards exactly 1/3 AD106 on TSMC N4 (1 GPC, 12 SMs, 1536 cores, 48 texture units, 16 ROPs) which would be 4 teraflops at 1.3GHz docked and that would match rumours quite nicely. But the rumours have been all over the place so that doesn't mean much ofc...
so far, there isn't much reason to believe in any other specs than the one from the Nvidia leak and github commits. that being 1536 Ampere cores
 
How many SMs per GPC if it’s 2 GPCs? 8 like AD107 so 16 total? That seems like a lot of SMs to me, while 6 SMs per GPC feels unbalanced with probably a significant area penalty…

Whether 32 ROPs even makes sense perf-wise depends on the expected memory bandwidth and L2 cache size. I would expect 128-bit LPDDR5 for cost reasons (+maybe a 12-16MiB L2, vs 24MiB on AD107, if we’re *very* lucky) which would be ~100GB/s vs ~256GB/s for AD107, so I guess it’s in the range where 16 ROPs is less than gaming Ada but 32 ROPs would be noticeably more relatively speaking.

So I’d still bet on 1 GPC and 16 ROPs personally but hard to say.
There's a potential benefit in going with 2 GPCs with half the amount of SMs since rasterizer/ROP output scales with GPC count over SM count but at high clocks this configuration will manifest bottlenecks in memory bandwidth. For comparison, other systems like the Series S with 2 SEs (= 2 rasterizers/32 ROPs) clocked at ~1.5GHz has over 200 GB/s of memory bandwidth and AD107 with 2 GPCs (= 2 rasterizers/32 ROPs) clocked at ~2.5GHz with similar memory bandwidth has to make up the difference with 32MB of L2 cache ...

1 GPC is probably a realistic expectation if their plan is to keep a similar form factor and much like their current platform, unable to implement the full set of SMs for it ...
 
I just hope that we don't get miserably low clocks thanks to Samsung 8nm. It would be cool if developers could launch third party games simultaneously on all consoles and PC, including the switch 2.
Switch 2 is going to be a mobile device on non-cutting edge hardware. Switch 1 came out years after XB1/PS4 and uses TSMC, but that's still not enough for it to play AAA games of that gen without heavy efforts. They will need to play things cautious for reasons of performance consistency and longevity as well. And it needs to be affordable.

Better reconstruction might help a fair bit in terms of getting the GPU demands down(and memory somewhat), and one saving grace could be if devs really do keep making games 60fps, meaning a 30fps Switch 2 version becomes easier. But in the end, it's still gonna be a limited device, even if it's on TSMC 7/6nm. It would certainly be preferable to be TSMC and I very much hope it is, I just dont think that's gonna make the difference between 'capable of next gen games' and not.
 
How many SMs per GPC if it’s 2 GPCs? 8 like AD107 so 16 total? That seems like a lot of SMs to me, while 6 SMs per GPC feels unbalanced with probably a significant area penalty…

Whether 32 ROPs even makes sense perf-wise depends on the expected memory bandwidth and L2 cache size. I would expect 128-bit LPDDR5 for cost reasons (+maybe a 12-16MiB L2, vs 24MiB on AD107, if we’re *very* lucky) which would be ~100GB/s vs ~256GB/s for AD107, so I guess it’s in the range where 16 ROPs is less than gaming Ada but 32 ROPs would be noticeably more relatively speaking.

So I’d still bet on 1 GPC and 16 ROPs personally but hard to say.
Orin's 8 SMs per GPC, 32 ROPs because with Ampere I believe they made it 16 ROPs per GPC and I'm assuming it's the same with Orin. Maybe it's a bit more custom I'm assuming it's not

 

Attachments

  • orin ampere gpu.png
    orin ampere gpu.png
    118.1 KB · Views: 13
  • orin ampere soc block diagram.png
    orin ampere soc block diagram.png
    39.2 KB · Views: 13
Back
Top