cireza
Member
So this confirms it will be shit on Switch 2 then ?
Digital Foundry: Nintendo Switch 2 DLSS Image Quality Analysis: "Tiny" DLSS/Full-Fat DLSS Confirmed
- Thread starter adamsapple
- Start date
Zathalus
Member
Depends on the implementation, version, and internal resolution. But generally, DLSS 3 is superior to not running it and is worth the slight frame time cost vs just boosting the resolution.
winjer
Member
That is because of a few issues, caused by the Switch 2 being underpowered.
One is that the Transformer model is significantly heavier than the CNN model. And the Switch2 can barely run the CNN model.
The other problem is that temporal upscalers are highly dependent on base resolution and frame rate, to achieve good image quality.
Corporal.Hicks
Member
The TAA in RDR2 is terrible. It blurs even static images to the point where they look like they have been upscaled, and the blur is even stronger in motion, accompanied by noticeable artefacts. That's why I chose that game to test DLSS when I bought my current graphics card — I wanted to see if the blurriness could be fixed without downscaling. However, the DLSS implementation that came with the game was very poor, and my initial impression of DLSS was negative. The image was as soft as TAA native, but it was overprocessed on top of that, especially during motion (very strong motion-adaptive sharpening that supposed to counterbalance that motion blur). Luckily DLSS3 (and especially DLSS4 now) finally fixed image quality in this game. The image is razor-sharp, both when static and in motion. DLSS now renders more detail, which is particularly noticeable when looking at distant objects. There are also no noticeable motion artefacts around the grass and trees, so the image is clean and sharp, a very nice combination and massive win on all fronts to the inagame TAA. However, there are some minor imperfections to be aware of. For example, the quality of some effects in RDR2, such as volumetric clouds and SSR reflections on puddles of water, are tied to internal resolution. DLAA (native) can fix all of that, but if you lower internal resolution and use DLSS, then the image will show some pixelation if you look closely at these effects. That pixelation doesnt however lead to motion artefacts, therefore it isnt something that's noticeable unless you look for it. I would I prefer playting this game at 4K DLSS-Performance than TAA native, but on my current PC I can use DLSS Ultra Quality, or even DLAA in this game.
Guess which one is which.
4k version
cireza
Member
That's what my eyes tell me when I see the early results. Issue is that the console is going to be riddled with ports using the technology. I guess some things never change.
You're never going to get current gen ports of AAA titles running at 1080p on the Switch 2 with acceptable settings. So it's really a question as to whether the ports should exist at all, and if people are happy to buy and play them, I don't see the issue. Cyberpunk on the Switch 2 in many cases is able to deliver better image quality than the Series S version due to the vast superiority of DLSS over FSR/TAA.
Senua
Member
Pisser is closer to DLSS3 than 4Just imagine if I would have said the same about the PSSR
ecosse_011172
Gold Member
Still astounds me why people insist that Switch 2 is "underpowered" despite it being hugely efficient and not far off as powerful as a mass market hybrid could be.
According to many, It should have the horsepower of a 300W machine will being a 15W machine.
According to many, It should have the horsepower of a 300W machine will being a 15W machine.
It's not my point. I find ironical such overpraise about the DLSS and the extreme criticism for the PSSR limits meanwhile the incredulity about someone who complains to the DLSS below 1080p because see similar artifacts in a different branded upscaler.
Last edited:
I remember a certain someone telling people that the upscaler on switch 2 is going to be identical to $2000 PCs somehow because "its DLSS".
They are all TAAU based in the end, people shit on me because I doesn't knew it but doubt they can be many differences between them.
Last edited:
winjer
Member
But it's not efficient. It's using Samsung's 8nm node.
One that even in 2020 was well behind TSMC's N7.
The Switch 2 SoC should have been made with something more recent. Such as TSMC's N5 or maybe even N3.
That would have allowed for lower power usage, better temperatures and the chip could have a more powerful GPU.
winjer
Member
In that logic, FSR2.0 and FSR4 are all TAAU based.
But in reality, there is a gigantic difference between the two.
Zathalus
Member
PSSR has severe issues with denoising and image stability when the internal resolution is too low, and this has been called out repeatedly. It's likely one of the reasons it will be upgraded with the work being done on INT8 FSR4. DLSS 4 is simply not comparable, it is just by far the superior solution at this time. As for the Switch 2 "lite" DLSS model, it has been criticized by both DF and threads here. Just look at the Fast Fusion coverage, it got so many complaints the developer patched the game to disable DLSS.
The switch 2 isn't using DLSS 4s new transformer model and so largerly has similar or even worse artifacts.
I was agreeing with you. It was a throwback to this thread when people thought this small handheld would give the same upscaling results as $2000 PCs:
The announcement from last night, the move to change from old DLSS Convolutional Neural Networks (CNN) to transformer model is applicable for all RTX hardware
See ENHANCED: DLSS Super Resolution.
Even Switch 2 with rumoured old ass Ampere will have the best upscaling tech of console world
Next gen, if a console manufacturer goes...
See ENHANCED: DLSS Super Resolution.
Even Switch 2 with rumoured old ass Ampere will have the best upscaling tech of console world
Next gen, if a console manufacturer goes...
As I predicted people are now defending those poor upscaling artifacts too.
Denoising are not exactly natively part of an upscaler though. And when resolution is too low IQ is more unstable on DLSS also to be fair. Something which DLSS fans continue to deny irrationally.
Last edited:
rodrigolfp
Haptic Gamepads 4 Life
One thing will change. Games will at least not be that blurry and aliased mess that SW1 games on SW1 hardware are.
ecosse_011172
Gold Member
Nintendo Switch 2: the Digital Foundry hardware review
The Digital Foundry team delivers its verdict on Nintendo Switch 2. It's an impressive package overall, marred by a disappointing, sub-par screen.
www.eurogamer.net
"Switch 2 looks like an efficiency king then, defying expectations. "
I'll listen to experts, thanks.
winjer
Member
Nintendo Switch 2: the Digital Foundry hardware review
The Digital Foundry team delivers its verdict on Nintendo Switch 2. It's an impressive package overall, marred by a disappointing, sub-par screen.
"Switch 2 looks like an efficiency king then, defying expectations. "
I'll listen to experts, thanks.
LOL, DF are not experts in tech.
Zathalus
Member
The original comment by
Bojji
mentioned DLSS 4. But the DLSS 3 CNN model the Switch 2 uses is still mostly superior to PSSR, while the lite model is obviously not. This will likely change next year, and the new PSSR should be superior to DLSS 3.Yes, as resolution decreases then DLSS starts struggling as well. Although DLSS does offer a superior image to PSSR at any rendering resolution. Next year things will hopefully be different and the gap should shrink significantly.
ecosse_011172
Gold Member
Sure, Mark Cerny mentions his admiration for them in tech presentations but Freddie Fanboy on the internet is better.LOL, DF are not experts in tech.
winjer
Member
You have to be very ignorant in tech, to think DF are experts in tech.
They do little more than capture images and compare frame rate and graphics settings.
They have little knowledge about chips, instructions, process nodes, etc.
Compare them to something like chipsandcheese, realworldtech, highyield, etc, and the difference in knowledge and information is staggering.
Besides, you didn't even consider that the node the switch2 is using, was never a contender for efficiency, not even when it was released, 5 years ago.
This an a basic fact that anyone with the most basic of tech knowledge knows.
Gaiff
SBI’s Resident Gaslighter
In fairness, "tech expert" is broad and vague at the same time. DF could be considered experts in their very specific niche, which is IQ analysis for games and the hundreds, if not thousands of hours of work they have put in peeping at pixels have made them pretty good at their job.
The other guys you cited operate in different fields that certainly require more scientific knowledge than DF and have broader applications.
Regardless, I think deferring to "experts" when it comes to something as subjective as image quality is foolish. Just saying you will listen to DF because they are "experts" is a poor argument. What they do lacks rigor and they most certainly aren't scientists, so their knowledge is not even close to being infallible.
ecosse_011172
Gold Member
Yeah, they could have went with 5nm, higher costs, possibly difficulties in manufacturing at the required volumes and squeezed a bit more power out.
Would have been more expensive and still not a PS5.
winjer
Member
But that is the thing, their expertise is only in analyzing frame rate and comparing settings. If your definition of tech is broad, then they may fit in.
But they have no technical expertise in analyzing chip design, process nodes, energy efficiency, programing, etc.
The question we were talking about is whether the Switch 2 is efficient. And in that regard, DF's opinion is inconsequential, because they have no expertise in that area.
winjer
Member
Depends on what they wanted to do. Keep the same size chip, then the price would increase, but so would performance and capability of the system.
But making a chip with the same transistor count, and it would have a similar cost, but much lower power usage. Which would mean a more efficient SoC.
Nintendo could even choose to just use the same power envelope, with the same transistor count, but have higher clocks for CPU and GPU.
And N5 is already a mature node, so there would be no problems with yields.
Last edited:
ecosse_011172
Gold Member
Sure about that?
The thing is, do people really think that the engineers at Nintendo and Nvidia are morons who have less technical knowledge than rants on forums?
They clearly went for the sweet spot in cost, performance and stability.
Gaiff
SBI’s Resident Gaslighter
Within this context, yeah. DF aren't electrical or computer engineers who know about silicon or chip fabrication. Relying on their expertise in this field as an appeal to authority is weak.
winjer
Member
Nvidia prices it's chips as high as they can. And if they can charge top dollars for a chip made in a node from 5 years ago, they will.
Nintendo doesn't care about graphics, and technical prowess and they haven't for almost 2 decades.
They choose to make a profit on the hardware from the start, so the N8 node was good enough.
But it does have implications. The chip is large for it's performance package and power efficiency is low.
The cost of a chip with 15 billion transistors, in Samsung 8nm, would cost around 80$.
In TSMC, also with 15 billion transistors, it would cost around 100$.
But then the smaller, more efficient chip would require a simpler and cheaper power delivery system, smaller batteries and smaller and cheaper cooling.
So not a huge difference, especially considering the efficiency gains.
In portable mode it's performing similarly to the Steam Deck with around half the power consumption. Ampere on 8nm was only slightly behind RDNA 2 on TSMC N7, and Ada on N5 is only around 50% better in performance per watt.
winjer
Member
50% improvement in power efficiency is huge, especially nowadays, with Moore's Law slowing down.
rodrigolfp
Haptic Gamepads 4 Life
10~12W is half of 15W?
It's huge if you use it to increase battery life, since it should equate to roughly half the power consumption at the same clocks. But Nintendo can still shrink the T239 to N5 on a subsequent revision to get those benefits. On the other hand I don't think an extra 50% of performance would have massively changed the product positioning, and then you can't use the efficiency gains for better battery life down the line.
15W is the power dedicated to the SoC alone, while the unit as a whole can consume 20W+.
winjer
Member
The N5 node is 2.8 times denser than Samsung's N8.
Having 2.8 more units would have made a gigantic difference in performance.
And it would still use less power. Though the cost of the chip would be significantly higher, though still lower than something like the PS5 Pro.
rodrigolfp
Haptic Gamepads 4 Life
25W for the Switch 2 unit. 40~45W for the Deck unit.
Last edited:
The density difference probably corresponds to the wafer cost difference, so we're talking about spending more money for a bigger GPU, which you can always do. What you're describing seems more like the (assumed) PS5 Portable strategy of selling a break even product at the price of a home console.
It's under 10W for the Switch 2 in portable mode, given the 2 hour+ battery life on the ~20Wh battery. While the below article reports a maximum power draw of 27W for the original Steam Deck and 23W for the OLED revision.
What’s In A Watt
About two years ago, Valve debuted the “Steam Deck”, a handheld computer that looked like a Nintendo Switch but was capable of running PC games. Then, 6 months ago, the company followed this up with an updated model called the “Steam Deck OLED”. I’m a fan!
www.willett.io
winjer
Member
Yes, having the same sized chip in N5 would be more expensive.
But the Switch 2 is already struggling to run like CP2077, at low settings. A game that was released 5 years ago.
Even with DLSS3 CNN, isn't enough to help out.
justiceiro
Marlboro: Other M
I'm more curious about their implementation of raytracing/path tracing.
rodrigolfp
Haptic Gamepads 4 Life
.
Last edited:
The same size chip on N5 could be 50% faster at the same battery life, which gives you a portable Rog Ally effectively. It would be nice but you're still well below the Series S. Realistically you need a bigger chip, with a wider bus, which will consume more power. So you will also need a bigger unit to fit a larger battery. And at that point you're pretty much talking about a different product, akin to what Sony seems to be planning.
winjer
Member
If if we go just by clocks increase, and get a 50% boost to performance, that is a lot.
For example, those drops in docked mode into 504p, would become 800p. That is something that DLSS can deal with a lot better.
It's because a lot of people are uninformed and ignorant on these topics and just get their opinions and takes from other uninformed, ignorant buffoons. It's a mob mentality.
The truth is that the Switch 2 is some extremely impressive hardware for the price, Nintendo was smart this time around getting the best perf for reasonable cost, and having Tensor cores will make this system age far more gracefully than the Switch 1 (which fell behind very quickly).
There still is nothing on the market that beats the Switch 2 in gaming efficiency. I love my Steam Deck OLED but it's collected dust since the S2 came out.
rodrigolfp
Haptic Gamepads 4 Life
7.8~12W the entire system portable. Can be half of the Deck, yes, but the discussion is about the APU only.
Last edited:
Bojji
Gold Member
Just imagine if I would have said the same about the PSSR
It's not comparable to DLSS3 or DLSS4 or FSR4. It's just as simple as that.
I tested vast majority of Pro games, when I bought it I had one month of PS Plus so games were on the service in full or in a form of Trials. And unlike you I could compare end results with FSR3, DLSS, TAAU or TSR found in PC versions.
If PSSR was as good as you or some other folks claim, then your lord and savior Cerny woudn't say this:
"The neural network (and training recipe) in FSR 4's upscaler are the first results of the Amethyst collaboration," Cerny told us. "And results are excellent, it's a more advanced approach that can exceed the crispness of PSSR. I'm very proud of the work of the joint team!"
Link.
PSSR is close to DLSS3 only in few games, The Last of Us 2 looks excellent with it. Apparently KCD2 and Yotei also have good implementations. Rest of games have more or less issues.
That's not true at all. Show examples.
It can't be 12W, because there is only a 20Wh battery (well listed at 19.3 Wh), and 2 hours is the absolute lowest battery life. So with 10W as the upper limit, I was assuming the rest of the system consumed at least 2W, giving 8W on average to the SoC, compared with 15W on the Steam Deck.
rodrigolfp
Haptic Gamepads 4 Life
Felessan
Member
PSSR is comparable to DLSS3 as it works on same basis. Not as refined though as it's relatively new and require time for tweaking
FSR4 is a transformer (cnn-transformer hybrid) - of course it's more advanced. It's a level of DLSS4 and Pro is waiting for backporting to int8 that is known to be on the way.
PSSR as a model is DLSS2-3 (CNN).