SEGA of America's plan for Dreamcast would use a 3DFX Voodoo2

Well, a lot of folks ran only one card, be it S3 Virge, ATI Rage or Matrox Mystique (or some of the later models, video card market was very turbulent in late 90s with cards releasing every year)...those initial 3D cards pretty much all sucked compared to 3dfx Voodoo, but they had 3D accelerators on board along with standard 2D circuits for that time.

Street Fighter 6 runs at 540p and upscales to 1080p with dlss docked.

"While our initial pixel counts of Street Fighter 6 during the Switch 2 Nintendo Direct reveal pointed to a native 1080p based on limited footage, the more recent Nintendo Treehouse feed shows a scale from a native 960x540 to a 1080p target using an AI upscaling method - with evidence suggesting Nvidia's DLSS based on this footage. Looking back, this 540p upscale appears to be in place in the original Direct footage as well. It's a remarkably clean scale in most shots, with the first few frames following a camera cut being one way to catch the base resolution from its docked mode play. We'll be sure to report back once we have more information."

https://www.eurogamer.net/digitalfoundry-2025-hands-on-with-switch-2-the-digital-foundry-experience

This is pretty impressive, for multiple reasons: 

1. Digital Foundry was convinced the output was 1080p, originally. That means the DLSS implementation is working at achieving a near-native output so well that even Digital Foundry misses it, without a more meticulous observation. 

2. The Switch 2 version is actually a lot cleaner than the PS4 version in terms of final output, despite it running at an internal 540p.

3. This allows Switch 2 to maximize efficiency in handheld mode for longer battery life while retaining graphical effects without losing much perceived image quality or performance. 

3dfx was an outlier in that regard and it was an issue just for a couple of cards, it wasn't the norm.

S3 (Virge, Savage), Matrox (Millenium, Parhelia) , ATI (Rage, Radeon), Rendition (Verite), 3DLabs(Permedia, GLINT), Intel (i740), STMicroelectrics (Kyro), Trident (Blade, 3D Image) and more could all do 3D and 2D just fine.
Even 3DFX with the Voodoo Rush and Voodoo Banshee could run SOLO in a PC as it included 2D acceleration.

Otherwise, 3dfx did require an extra graphics card for 2D acceleration (And sometimes an extra card for OpenGL/DirectX) whilst the card handled 3D graphics duties, which even then was best left for 3DFX Glide. ...OpenGL and Direct X could run, but later on, specifically with the Voodoo 4 and 5, TnL was sadly missing, so it wasn't Direct X 7 compliant which would later hamper the ability to run more modern games.

The restriction with the Voodoo 1, 2 and 3 is that you couldn't render 3D in a window due to the lack of 2D acceleration.

I actually still have my 3dfx Voodoo 2's 12MB in SLI mode, paired up with an Athlon XP 2000+, 512MB of SD Ram and a Radeon 9500 Pro.
Voodoo's are only used for Glide, everything else on the Radeon.

And it's a shame they didn't... Because several things would have happened.
3DFX would have retained the revenue they needed to stay in the market... And the Dreamcast would have been significantly more potent from a technology perspective, with easier porting of 3DFX Glide games which looked incredible at the time.
It would still be behind the Original Xbox and Gamecube as they could do pixel shaders and the like...

And Sega may have gotten some extra brand awareness, especially amongst the PC userbase.

Sega was also planning on using a derivative of nVidia's NV1 for it's Dreamcast console at one point as it supported Quadratics, which would have benefited developers who were used to Saturn.
At this point in time, nVidia was almost bankrupt with less than a months worth of cash reserves, Sega actually probably saved them with a $5 million dollar cash injection.

nVidia's first graphics processor, the NV1 relied on Quadratics and was an abstract failure, it actually shared a lot of similarities with the Sega Saturn on the graphics front... But once Microsoft threw it's support for Polygons with Direct X, nVidia's first graphics processors fate was sealed until they released the Riva/TNT cards.

Using those Tensor cores comes at a cost of power.
There is no "extra efficiency" or battery life.

The GPU core is still pegged at 100%, with extra processing required to bolster resolution/upscaling.

What it's enabling is higher visual fidelity.

I'm not sure it's so much tradition as just a commitment to high performance and wanting their games to feel very smooth.

If they targeted 60 even on much more constrained platforms like the Wii I can't see them dropping to 40 now that they have way more powerful hardware. 

That said, if there is indeed a Switch 2 exclusive Prime 5, I wouldn't expect a 120fps mode, or at least I hope not as that would seriously limit what they could do not just in terms of graphics but also gameplay.

Nearly all evidence we've seen is that DLSS doesn't fully utilize tensor cores 100% of the time. Instead you see a spiking utilization pattern where most of the time utilization is low and then, when needed, it spikes up to double digits (high single digits on a 4090, but obviously the Switch 2 isn't comparable there), which is consistent with what we see when we run comparable DL models in other work loads. 

The GPU might be fully-utilized in this case, as resources are re-allocated to improve other parts of the graphics pipeline when saved from running at the higher resolution, or the developer could just under-clock the GPU for better battery life, as they do in lighter-weight games or as we see on PC handhelds when selecting power limits. 

If 540p can be scaled to 1080p that well, that's a very promising sign for DLSS giving Switch 2 a real leg up in terms of performance.

If you only need to render at say 540p to get decent image quality then that really opens up the range of games the console can handle without resorting to the kind of blurriness we saw on many "impossible ports" to Switch 1.

So your evidence is a reddit thread...

But even gleaming the reddit thread we can already glean some glaring issues as you failed to grasp some intrinsic technical aspects of DLSS as you lacked appropriate context in your reply.
DLSS is an algorithm with a fixed amount of resources required to run.

1) Like you alluded to... The user is showcasing a Geforce RTX 4090 with 330 tensor ops via 512 Tensor cores. - This is Turing with 2x the Tensor throughput of Ampere.
2) Switch 2 uses 48 Tensor cores with likely a max throughput of 6 TOPS.

If we assume that 1% of the 4090's is your regular loading on the tensor cores... Then that means DLSS would require about 3.3TOPS.

So Switch 2's Tensor cores are at 50% utilization, the peak exceeds the Switch 2's Tensor throughput entirely.
But this is a like-for-like algorithm, which will not happen with the Switch 2 as it likely uses an algorithm specifically optimized for the hardware limitations and characteristics. (I.E. Being Ampere and not Turing.)

That's also failing to ignore that the Switch 2 is in a walled garden and develoeprs are free to use 100% of the systems resources, regular rasterization and ray tracing will be using 100% of those resources with Tensor operations as an added extra on top.

And the battery life reflects this as battery life on Switch 2 is extremely poor, worst than the original Switch 1 launch model.

The evidence is found in the reddit thread, it's not "the reddit thread." The user conducted an experiment and I shared their results with you, but other users have also validated that tensor core utilization varies over time when running DLSS workloads. It is also something those of us who build and run CNN (and ViT) models on a day-to-day basis see, and makes sense from a theory perspective given the architecture of a CNN (or ViT.) You're not going to be multiplying the same ranked matrices all the time*, nor will your workload always be core-bottlenecked, often the bottleneck is the memory bandwidth. The evidence I shared is the fact that we see a literal order of magnitude difference between average usage and peak usage. Any CNN (or ViT) will have this same usage pattern, because they all use the same tools. Maybe for Switch 2, using a hypothetical bespoke model, it is 3% average vs. 30% peak utilization (instead of the .3% vs. 4% of an RTX 4090), but either way average usage << peak usage. 

THAT was the point I am making, and the one important to the topic of considering the relative power consumption of the tensor cores to the rasterized workloads they are reducing. A workload that spikes up to 100% only one-tenth of the time isn't going to consume as much power as one that is pegged at 100% all of the time. 

Developers are indeed free to use 100% of the system resources, they are also free to limit power-consumption in handheld mode and have done so with the original Switch. That's why battery life varied by title. There were different handheld clock modes that developers used for different titles based on how demanding the title was on the systems resources. What DLSS provides them is the option to reduce clocks more often (if their goal is longer battery life) by reducing the rasterized workload without a power-equivalent increase to the tensor-core workload (even if the tensor utilization eats into it.) In other words, they are more efficiently achieving a similar output. 

I don't even know why you're arguing with this. People do this all the time on gaming handhelds like the Steam Deck for many games. They'll cap their power-limit to 7W and use FSR to make up the difference, maximizing the battery life, and not having that worse of a qualitative experience. When they are on a charger or dock, they change their settings to rasterize at a higher internal resolution, as battery life is no longer a consideration. 

*Matrix multiplication algorithms scale either cubicly with rank for high ranked matrices or super-quadratically, sub-cubicly with rank for low ranked matrices. Then there are factorization layers that can reduce rank based on the matrix sparcity. Different layers in the network are going to have different ranks and sparsities and therefore take up different resources.