I'm started to wonder if Switch will be sold as two separate devices
1) A handheld
2) A console
They can have a "cheaper" handheld while simultaneously going significantly more powerful with a more expensive dock
This seems like a winning strategy to me, rather than combining the two devices and forcing consumers to pay for both.
I suspect many audiences play either entirely handheld or docked, but seldom overlap both
So, do you think they would go for something that's mainly the same spec across two distinct devices, with the difference being the form factor? And that otherwise, the performance difference would be the console variant could run at higher clocks with better cooling?
That could work. Wonder what the stats are in terms of how many people use their Switches docked vs. undocked, particularly primarily. If it came down to something like 60% owners go undocked 90% of the time, and 30% owners go docked 90% of the time, then maybe Nintendo would have good enough data and business case to put the same specs in a portable and a console (which you could think of as an Apple TV-like).
The handheld would just run everything at lower clocks, and games would target it as the baseline, so anything on the console would be getting natural performance boosts. They might be able to go $299 for the handheld (just like the Switch) but maybe only $199 for the console, making 2x the former vs the latter. It'd require a scale of production logistics management like Microsoft's had to do with the S and X. But unlike those systems, Nintendo wouldn't have any explicit overlap. Demand for either could be more easily met, they could more easily adjust the production pipeline (everything's on the same process node APU-wise, for example). Really in the event of any SKU adjustments to meet demand, they're just changing some materials for the particular form factor. They also avoid the problem of devs being faced with optimization issues in their hardware ecosystem.
Basically, they'd be emulating what Sony's planning to do with the new PS5 models, but in practice successfully execute a dual-SKU approach Microsoft has failed to actually accomplish with the Series X and S (when we're talking market performance, and developer satisfaction).
It's just simple physics, x86 and RDNA3 on 5nm operating at the likely power budget of a switch (under 10w) at best matches a stock PS4 with a better CPU. Nvidia does have better performance per watt but even using Ada on a 4nm SoC you will at most be 60/70% over that. Then you have a number of other issues, even the best LPDDRx does not offer much faster then 130GB/S bandwidth. Then we have the CPU which is Cortex A78c based at best. That has similar IPC to Zen 2 (albeit no dual threading as ARM does not support it) and no way is that going to clock near current gen consoles.
DLSS 3 is nice but you don't get any real performance boost over FSR, just much improved image quality. Frame generation would not really be used as it is only good for boosting 50-60 fps up to over 100.
The Switch can maybe match the Series S GPU when docked, but it will be weaker in CPU and memory bandwidth. Hoping for anything better is simply going against the laws of physics. 4nm isn't that much better over 7nm.
FWIW when I was speaking of a Switch 2 being roughly comparable to a Series S, I meant when docked. Undocked, there's no chance. I agree that the biggest challenge would probably be the CPU, but the CPUs in the current Xbox & PlayStation consoles aren't really that impressive. So, there's no telling how much or how little the ARM cores in what Nvidia package Nintendo goes with, would have to be in order to match up with what the Series S CPU can perform.
Actually, the other potential issue would probably be storage I/O, though if the Switch 2 is working with lower internal resolutions getting scaled up through DLSS, textures and other graphics assets don't need to be as big, so you naturally save bandwidth that way. The hypothetical design I was talking of, would be something aiming for a lot more than 10 watts total power consumption anyway, probably closer to 40 - 50 watts.
But reading
James Sawyer Ford
's post, I start to think maybe a better implementation of the concept would be to just aim for something with the same CPU, GPU, RAM etc. specs just one as a handheld and the other as a small home console. Conservative clocks in the handheld, higher clocks in the console due to additional TDP and more efficient cooling. As some of what you illustrated, the handheld itself could offer roughly PS4-level performance (weaker on TF, matching or besting it on pixel and texture fillrates, mix of 2 GB LPDDR5 @ 36 GB/s for CPU, 8 GB GDDR6 as 2x 4 GB GDDR6W modules @ 9 Gbps (144 GB/s; GDDR6W is 64-bit per chip)).
Meanwhile, the console variant could increase the clocks on the memory; getting "Series S performance", at least on the GPU, is mainly in terms of pixel and texture fillrates. TF doesn't matter as much. The clocks on the CPU and GPU could be increased. The question is how much could they be increased by and stay at a ~ 50 watt max TDP, or something between 40 and 50 watts. There's no way to really guess that right now because many things about it spec-wise, we don't know a lot of yet. For example, I've heard the possible GPU it could be based on can run up to 2.65 GHz. It won't do that in an actual Switch 2, we know that. But theoretically, if it were to, then they'd only need 20 ROPs to match Series S's pixel fillrate (actually, they'd slightly exceed it).
