On the contrary, I'm pretty sure he said it was very recent info. So maybe not to the day current but it at least seemed like he meant within the past month or so.
-
We have an update on ResetEra's ownership. More info here!
On the contrary, I'm pretty sure he said it was very recent info. So maybe not to the day current but it at least seemed like he meant within the past month or so.
We're in the middle of march so I anticipate if this device is coming out in 2021 something more concrete emerges by end of April. By June we should basically know everything relevant.
What you people think about this
You think Nintendo will use DLSS to improve image quality of their virtual console games? Update the snes and nes apps?
You think Nintendo will use DLSS to improve image quality of their virtual console games? Update the snes and nes apps?
Slightly off topic to the current discussion, but there was a discussion a while ago about DLSS performance, and specifically whether the performance required to run DLSS depends only on the output resolution, or if the input resolution has any effect. I had a bit of time today to test this, so I thought I’d share my findings.
To start, I’m running an RTX 3070 (with a Ryzen 3700X CPU and 32GB of RAM), and I used Control as the test game. I kept the settings to max (including ray tracing) and changed only the resolution settings (including DLSS). I used OCAT to log frame times, and as Control doesn’t have a benchmark, I just ran through an identical 60 second route for each run, not interacting with any enemies or physics objects to try to minimise variance from one run to the next. Finally, I took the median frame time for each run, as I want to look at a typical frame, and I’m making the assumption that the time taken to run DLSS should be pretty stable from one frame to the next.
I ran quite a few tests, but unfortunately when I got down to 720p it became CPU limited, so I couldn’t really get anything useful from those runs. The useful runs were as follows (again, in each case I’m taking the median frame time):
1080p native - 14.5735ms
DLSS - 1080p -> 4k - 16.3515ms
1440p native - 21.988ms
DLSS - 1440p -> 4k - 23.4835ms
So, this would suggest that the time taken for DLSS to scale a 1440p source to 4K is about 1.4955ms, whereas the time taken to scale a 1080p source to 4K is about 1.778ms. That would mean that it takes about 19% more time to scale from 1080p to 4K than from 1440p, which indicates that the source resolution does actually have an impact on the performance required for DLSS. I would have liked to see if there was a bigger impact when scaling from 720p to 4K, but as mentioned above I was CPU limited at that point.
I should emphasise that this is a pretty crude way of calculating this, but the numbers do line up pretty closely with what Nvidia have advertised. They stated 1.5ms at 4K for the RTX 2080Ti and the RTX 3070 is in a similar performance bracket. I may be over- or underestimating the times, but they do seem to be in the right ballpark at least.
To start, I’m running an RTX 3070 (with a Ryzen 3700X CPU and 32GB of RAM), and I used Control as the test game. I kept the settings to max (including ray tracing) and changed only the resolution settings (including DLSS). I used OCAT to log frame times, and as Control doesn’t have a benchmark, I just ran through an identical 60 second route for each run, not interacting with any enemies or physics objects to try to minimise variance from one run to the next. Finally, I took the median frame time for each run, as I want to look at a typical frame, and I’m making the assumption that the time taken to run DLSS should be pretty stable from one frame to the next.
I ran quite a few tests, but unfortunately when I got down to 720p it became CPU limited, so I couldn’t really get anything useful from those runs. The useful runs were as follows (again, in each case I’m taking the median frame time):
1080p native - 14.5735ms
DLSS - 1080p -> 4k - 16.3515ms
1440p native - 21.988ms
DLSS - 1440p -> 4k - 23.4835ms
So, this would suggest that the time taken for DLSS to scale a 1440p source to 4K is about 1.4955ms, whereas the time taken to scale a 1080p source to 4K is about 1.778ms. That would mean that it takes about 19% more time to scale from 1080p to 4K than from 1440p, which indicates that the source resolution does actually have an impact on the performance required for DLSS. I would have liked to see if there was a bigger impact when scaling from 720p to 4K, but as mentioned above I was CPU limited at that point.
I should emphasise that this is a pretty crude way of calculating this, but the numbers do line up pretty closely with what Nvidia have advertised. They stated 1.5ms at 4K for the RTX 2080Ti and the RTX 3070 is in a similar performance bracket. I may be over- or underestimating the times, but they do seem to be in the right ballpark at least.
A lot of people have been pretty reflexively negative on Nintendo hardware for a while now, but Aula has been revealing just how deep that goes.
DLSS has nothing to offer to the style of rendering that old 2D systems use. Even for 3D ones, it wouldn't surprise me if implementing it was problematic (though those should benefit a lot from an increase in raw GPU power).
I don't think they would really benefit from DLSS. NSO games are already upscaled at the max resolution that the Switch can deliver, right?
some of his sources include people's profiles, and we don't need internet folks blowing up their page views due to thirst
The hardware should be sufficient to render N64 games in native 4K, and integer upscale S/NES. DLSS doesn't seem necessary.
You could probably get a nice clean looking image out of something like Starfox with DLSS, but yeah, I am betting that DLSS applied to pixel art would yield weird results. Something specifically trained to rework pixel art into a cartoon or hand-drawn style might work, but it could easily yield monstrosities like the upscaled assets in the various Final Fantasy mobile ports.
It's not something that is feasible or workable in any way. Unless you're talking about simple upscaling and even then that's likely to add latency.
Thank you, this is interesting. I am curious about the source of the cost dependence on the input. On one hand, a 1440p input would have more samples than 1080p in the neighborhood of each 4K output pixel that the neural network needs to weight and accumulate; on the other, it intuitively makes sense that 1440p input would converge to 4K faster than 1080p input, since the 1440p sample is a more correct representation of the scene. From your test, it seems like the convergence aspect is more relevant to the final computational cost than the sample density.
Somewhat separately, I also am curious about if DLSS is handling every output pixel with a single forward propagation through one large neural network or if each output pixel is handled separately with a smaller neural network in parallel.
That's not happening. It's highly unlikely they'd be able to do that while still allowing you to just pull the system out of the dock whenever.
Even Super FX is probably being done entirely in software and probably wouldn't benefit. When it comes to systems like NES, SNES, GB, and GBA, generally what you see is what you get, and an emulator is not going to improve that without a lot per game hackery. GB can (and should) leverage the Super Game Boy integration, but that's about it.
For emulating newer hardware that had something that at least kind of resembles modern 3D acceleration. Integrating DLSS might be practical, but 1.) I don't think it can necessarily be assumed that the necessary info will be readily available, and 2.) you could probably push the native resolution of games for any of the relevant systems pretty high on XB1/PS4-tier hardware without any upscaling trickery.
Heck, that could be the answer as to why production could be starting mid-year for something not releasing until next year. Whereas GBASP and DSL could get away with launching only in Japan at first, a worldwide fast turnaround would require having more stockpiled and ready to go.
That's not how DLSS works. They could always add other filtering techniques intended for use with sprites if they really wanted. It's possible with RetroArch on hacked Switches without issue.
Like making a USB dongle that would turn PS4 into PS4 Pro.
I want help eliminating Aula as a legit option. Lets assume for a second, i know against what most of you think, that Bloomberg is the same machine as Aula (Mariko + new oled screen + new hdmi output chipset), and that DLSS and new SOC isn't happening for this unit (maybe a future one?..but let's leave that to the side for this thought experiment). The Aula firmware and the bloomberg reporting are lining up in general, minus one big thing.
That one big piece needed to be answered for Bloombergs reporting to add up, is 4k:
1. How to get to 4k when docked?
The assumption everyone has around this is that the Mariko will see some significant improvements with removing (reducing) the underclock but not enough to get to 4k, but theres no actual evidence of any change in clock speed right? Its just a pure assumption based on the fact that the X1+ cannot do 4k today and that we know its significantly underclocked, so thats the only place "new power" could come from in the absense of a new SOC. Nothing in Aula's firmware discussion was there talk of new performance profiles that i could find. Nintendo might leave off the new profiles until closer as to not let the cat out of the bag, but it seems like that would be the main thing developers would need, but thats the big hole in Aula being the next release. Even with a new profile and unlocked chipset, it doesn't seem possible to reach 4k rendering even known software reconstructions (checkboard, etc); and Bloomberg is reporting that developers are being told be be 4k ready. Which leads into the next question:
2. What does 4k ready mean?
Nintendo could still make that claim if they can output 4k in a theoretical sense (even just video or in via cloud streaming)...this seems unlikely cause everyone would feel this is cheap (didnt stop the 8k logo being on the PS5 box though); and doesnt require developers to do anything. So with the Bloomberg report, it means there's gonna be some sort of gaming angle...but it seems highly unlikely that removing the underclock will get any AAA games running at 4k...even with a checkboard type of non-dlss reconstruction. So triple AAA gaming in reconstructed 4k isnt possible with Aula. The other option is a post-output upscaler, but even that doesnt seem to fit here either, cause developers wouldn't need to be 4k ready if it was just upscaling 1080p video. Could simple indie games or remade classics tetris or a very simple 2-d platformer work at 4k? Again, this would be 4k in name only. But that leads into the final question:
3. What other bottlenecks are there for 4k besides the SOC?
Even if they could push out a non-3d 4k game, is the memory bandwidth fast enough to handle these bigger assets/loads? The Aula firmware has the same 4GB of LPDDR4X ram, but shrunk down to a 10nm process, which could mean it gets some gains at least in power efficiency (this says a 10% less power on a similar chip); but 4k assets take up more memory...and 4gb isn't a ton. Another factor would be game cart's, we already have developers cheaping out on physical cart sizes and requiring downloads for the full game. 4k assets would balloon the size needed on the carts additional and theres no mention of increasing the built in 32gb of storage in the Aula firmware. What other factors would limit 4k rendering other than SOC and memory?
Everything else about the Bloomberg/Aula line up and seems completely feasible for Nintendo to actually produce/release this year. Certainly is more an XL than a pro, and would be disappointing to many expecting something more significant, but technically makes sense with the exception of the 4k, which seems like they could say in only the most basic of sense. So outside of other peoples unconfirmed info, I was hoping you could help me knock this theory down on why this isn't feasible or even likely. The only holes I can find are the 4k parts, which we dont even know what it means...why wouldn't Aula be the same unit as Bloomberg other than it doesnt fit the conflicting DLSS narrative?
That one big piece needed to be answered for Bloombergs reporting to add up, is 4k:
1. How to get to 4k when docked?
The assumption everyone has around this is that the Mariko will see some significant improvements with removing (reducing) the underclock but not enough to get to 4k, but theres no actual evidence of any change in clock speed right? Its just a pure assumption based on the fact that the X1+ cannot do 4k today and that we know its significantly underclocked, so thats the only place "new power" could come from in the absense of a new SOC. Nothing in Aula's firmware discussion was there talk of new performance profiles that i could find. Nintendo might leave off the new profiles until closer as to not let the cat out of the bag, but it seems like that would be the main thing developers would need, but thats the big hole in Aula being the next release. Even with a new profile and unlocked chipset, it doesn't seem possible to reach 4k rendering even known software reconstructions (checkboard, etc); and Bloomberg is reporting that developers are being told be be 4k ready. Which leads into the next question:
2. What does 4k ready mean?
Nintendo could still make that claim if they can output 4k in a theoretical sense (even just video or in via cloud streaming)...this seems unlikely cause everyone would feel this is cheap (didnt stop the 8k logo being on the PS5 box though); and doesnt require developers to do anything. So with the Bloomberg report, it means there's gonna be some sort of gaming angle...but it seems highly unlikely that removing the underclock will get any AAA games running at 4k...even with a checkboard type of non-dlss reconstruction. So triple AAA gaming in reconstructed 4k isnt possible with Aula. The other option is a post-output upscaler, but even that doesnt seem to fit here either, cause developers wouldn't need to be 4k ready if it was just upscaling 1080p video. Could simple indie games or remade classics tetris or a very simple 2-d platformer work at 4k? Again, this would be 4k in name only. But that leads into the final question:
3. What other bottlenecks are there for 4k besides the SOC?
Even if they could push out a non-3d 4k game, is the memory bandwidth fast enough to handle these bigger assets/loads? The Aula firmware has the same 4GB of LPDDR4X ram, but shrunk down to a 10nm process, which could mean it gets some gains at least in power efficiency (this says a 10% less power on a similar chip); but 4k assets take up more memory...and 4gb isn't a ton. Another factor would be game cart's, we already have developers cheaping out on physical cart sizes and requiring downloads for the full game. 4k assets would balloon the size needed on the carts additional and theres no mention of increasing the built in 32gb of storage in the Aula firmware. What other factors would limit 4k rendering other than SOC and memory?
Everything else about the Bloomberg/Aula line up and seems completely feasible for Nintendo to actually produce/release this year. Certainly is more an XL than a pro, and would be disappointing to many expecting something more significant, but technically makes sense with the exception of the 4k, which seems like they could say in only the most basic of sense. So outside of other peoples unconfirmed info, I was hoping you could help me knock this theory down on why this isn't feasible or even likely. The only holes I can find are the 4k parts, which we dont even know what it means...why wouldn't Aula be the same unit as Bloomberg other than it doesnt fit the conflicting DLSS narrative?
Last edited:
I see. I ask because I continuously see it mentioned in videos and such. Then it dawned on me that simply having a new dock doesn't really make sense for Nintendo. A dock would be considered as an accessory and wouldn't necessarily relate to hardware sales.
DLSS isn't really built for that. For older sprite based games you're going to need a highly trained reconstruction AI to attempt to clean them up and/or make them look hand drawn. Similar to what is being done with textures on PS1-PS2 era games currently. I know there have been attempts in recent years, but I'm not sure where that is at.
Could be something the devs/pubs look into down the road, but to really remaster old sprite based content in the way you're thinking it would probably take AI+significant human touchups.
It would require a new System anyway for multiple reasons.
To make a dock that can increase performance to be "Pro"-level, they'd need to have a dedicated GPU in the dock, that would require
- A new CPU fast enough to utilize that GPU, therefore a brand new GPU inside the SoC itself from the get-go.
- Updating the USB Protocol on the Switch to either Thunderbolt 3/4 or USB 4.0 in order to have the dock act like a PCIE E-GPU enclosure.
- The current Switches use a weird USB protocol that isn't quite USB 3.1/USB-C Standard Compliant in regards to power draw/intake/data transfer
- Designing the dock in a way to where the GPU in the dock won't bake the portable tablet system.
- Them finding a way to get the latency of E-GPU solutions down well enough to counter most of the downsides of conventional laptop E-GPU solutions.
Although I do hope they update the USB port to USB 3.2 gen 2x2 (Terrible name still) so we could get external cold-storage support potentially.
There is no way Nintendo will announce it before Holiday season and than not launching it this year,
Nintendo has like 50% of hole years sales during Oct-Dec period, and announcing it around that time would affect greatly on sales during that time.
Nintendo will not announcing much earlier (3m+) and effect sales during most important part of year because potential leaks,
I mean all the time there are some things get leaked but most of general market and buyers are not pay attention to leaks.
Actually Nintendo usually announcing they revisions around 1-2 months before launch, latest example is that Lite was announced in July and released in September, so announcement in January with release date in March has plenty sense and that would be more than enough time,
I would said thats quite possible scenario.
Nate said he is sure they will announce it before FY ends, but he is not sure when will be released.
Last edited:
I mean if Nintendo wants they could easily push IQ of those apps with current hardware, expect not having 4K off course because Switch doesn't support 4K.
As far as I can tell, Aula is just a way for Nintendo to work on a new dock that can output (and do a basic scaling) to 4k. It's probably just an interim development piece for them to get accessory work done for Switch Pro 2 Advance Super 64. They're probably just using some basic test software with it. Move the block around and test button presses and test latency and such.
I don't feel like getting into this again. We have only a few examples of this happening to begin with so you're guess is as good as mine as most revisions tend to be announced and release in the lead up to the big holiday period , not after it. (not only to catch the big sales season but also consumers know well ahead of time the options they have)
The last discussion had people fixated on '2 months' based on lots of announcements to launch in the middle of the year. Which to me isn't really a fair applicaiton of that rule.
The most important thing that I thought was obvious but probably isn't in hindsight is that for a manufacturer, giving consumers choices and not fooling them into buying a product is key. (see Sega's many hardware releases and the loss in trust with the company)
if I'm going into the holidays and I know I have a desirable SKU that is coming soon after the holidays I'd want clarity for my consumers so if they buy this Christmas they;re wont be anger/regret next year when I drop an announcement literal weeks after Christmas. To me, a company as savvy in PR as Nintendo would understand implcitly and would care about it more than marginal lost sales at Christmas.
And this brings me to my next point. The idea that a majority of consumers will simply stop buying OG Switch to buy a Pro, which will likely be more expensive anyways, is false, as is evidenced by PS4Pro sales as a proportion of PS4 hardware sales. They can chew gum and walk at the same time.
If there is indeed 2 SKUs like what Nate is speculating, they could announce both at the same time, similar to what Sony did, even if one SKU is delayed (but i really hope both arrive on time). So consumers have clarity going into the holidays.
This is the last time I'll discuss this.
thanks.
To get 4k out of the current switch you have to clock it so high I don’t even think it‘s sane to entertain the idea. What like over 4GHz just to get 4k out of 1080p switch games? That’s only for the GPU, now consider the other aspects of the system like memory for example? Once you start to take these into consideration you get a divide by zero error and from a Portable device no less. Even taking more sophisticated software techniques like CBR over plain upscaling, the switch isn’t capable enough for that sort of method I think. PS4 Pro used it more often than not. And to implement such a technique is more work than having TAA implemented in-engine.
But that is for gaming, for video playback the TX1 is enough already for 4k, the one in the OG switch that is and the X1+. But as you brought up below
The video aspect is useless and the gaming aspect is the only one that applies. Having a post upscaler that is not like right there in the SoC of the main unit (the tablet) would introduce some lag. Like for example in the dock would definitely introduce lag
But I believe if its just above 1440p or something they can market the game as 4k ready since a TV upscaled from then already and no TV does the 1440p resolution that I know of. 1080p FHD to 2160p UHD. So even if not exactly 2160p, they can still mark it as 4K.
The size of the device, the memory bandwidth, the laws of physics. You get a more general idea even if I exaggerate the point. For current switch games you need more than double boost to go from 1080p to 2160p. To support it in a feasible way you would need a bigger boost regardless even to RAM. As for cart sizes, I don’t think that matters since games will still target the base switch, so those would be via patches not on the cart really.
Base switch would not be doing 4k.
I am always ready for Aula regardless, but I also don’t believe it blindly if what Mochizuki said it does 4k because I raise an eyebrow as to what method they are even implementing for such a device.
Either way if it has other QOL improvements I think people will welcome it easily and *ahem* find it to be neat
But what also doesn’t align is insiders and some devs commenting that it isn’t the same thing. Like Devkits going out where... why if it is the same profiles I’m docked and handheld to not create too many disparities.
That’s what some of us are also assuming regarding that. The Realtek chip doesn’t upscale content but it is a DP1.4 2 HDMI2.1 SoC which would fit with the current dock that uses an older SoC from 2012 for its job. This new one would work better with a new switch for several years even. Even for Switch 2 maybe. Until 2 Pro I think. DP2.0 to HDMI2.x
Anyway Aula is tiring lol
Last edited:
OP
OP
Welcome to the ERA > Random Famous Website Hint > ERA mode
Yeah, right now we are likely (hopefully) reaching a zenith on rumors of the Switch Revision, where the last two big options are having their last arguments to see which one would be most likely. It's just both sides make some degree of sense to the proponents of the side to where they can't be shut down without a hard confirmation from a reliable source.
Until someone comes out and says "There is only one model and it's X (X being a Pro or Mariko-OLED)" or "There are 2 models" this thread will likely be stuck in this spiral loop of confusion.
Personally, I'm sticking with a 6SM variant as my guess, there just are too many loops to jump through to make a 4SM Ampere system work as a "4k Ready" Pro system.
- 4SMs means the system wouldn't be able to lock 1080p at native res often, meaning they'd need to use Ultra Performance DLSS to hit 4k
- 16 Tensor Cores in a 4SM system would likely not be enough to do 720p to 4k upscaling, much less at framerates above 30fps unless if NVIDIA pretty much redoes the DLSS pipeline
- 4SMs of CUDA cores would heavily hold back the far faster A78 CPU cores in the SoC based on what NVIDIA's ARM CPU licenses are.
- So they are in a back and forth of issues with a 4SM config, it's not strong enough to do native 1080p reliably enough to use Performance Mode DLSS to 4k, but it wouldn't have enough Tensor cores to use Ultra Performance DLSS to make 720p into 4k.
Fuels a desire for real news. I think there's a fairly well speculated next system from Nintendo (Orin Based A78(C?)/Ampere(+?)), and it's just a matter of waiting to see how they market it and if it replaces the existing Switch models or not.
Screen type, control changes, new form factors, and whatever are really window dressing along side that.
I wish I was capable of ignoring it all until some real news showed up, but I just don't function that way.
OP
OP
only when we let Aula go will we be able to move on
you'd need a different upscaling model from DLSS. DLSS is specifically trained to be a better TAA
Aula makes a lot of sense to me lately because I don’t think they can price the pro model at 300. It sounds more like a 400 model going off reports and insiders. That’s why there is a need for a cheaper approach priced at 300 which basically just upscales to 4k but doesn’t really improve games in any other way. They’ll probably ride out the Lite as it is for a few more years as it’s more geared toward handheld gaming but having 3 models at 200, 300 and 400 seems like a good approach.
Can’t do that when the device sounds more and more Aula like, for every article that comes out :p Can’t put my faith 100% in insiders.
I think one can suspect two revisions - XL and Pro. Remember, I've only ever heard of Pro and have assumed that is the model Bloomberg refers to. I'd assume the case with Matt is similar. Doesn't mean Bloomberg isn't Aula/XL. Just means we may be referring to a different product than the report. The two can coexist.
Unless it's something you actually heard of I really see no basis to think two things are coming. Surely we'd be hearing rumblings from at least the supply chain which talks to Bloomberg.
Really the only reason the whole two models idea came up was DF's video speculating on the Bloomberg article. But that was apparently just speculation.
It seems a lot likelier that there's only one product planned for this FY. Mochi saying it will be released this year could just be referring to the most recent plan he'd heard of, which may not be the current plan. It's a lot easier to resolve all of the reports we've heard if there's one model planned.
Are there estimates of how much the SOC costs? Either Mariko now or the original back in 2017? I find some old not very helpful reports like this which say things like "the touch screen, Tegra SoC, and dock combined total $167". Since there's no reason to think parts like Joy-Con/battery/dock/whatever will have notable increases, it seems like a new SOC would have to be ridiculously more expensive for a hundred dollar increase to be attributed primarily to it.
Honestly $400 seems low for a brand new custom ampere based mobile processor with dlss that is wildly power efficient, seems that would be be at least premium phone level pricing unless Nintendo takes a loss per unit. it's also why i am having a hard time believing it too... just don't see a cost effective path to what is being proposed as the non-Aula option. If an existing chip existed that worked then i would have a much easier time accepting it.
That's called confirmation bias
The device Bloomberg has been decribing since last year doesnt fit Aula.
Nintendo releasing a revision to the Mariko Switch with an OLED screen this holiday then turning around and releasing whatever the "Pro" is a quarter or so later would be some grimy business, imo.
I really don't understand the renewed interest in AULA, or why the Bloomberg articles are being interpreted to be talking about something different than what every other source out there is talking about. The "enhanced graphics" and mention of developers having to deal with them alone, to me, throw the whole AULA thing out the window.
I really don't understand the renewed interest in AULA, or why the Bloomberg articles are being interpreted to be talking about something different than what every other source out there is talking about. The "enhanced graphics" and mention of developers having to deal with them alone, to me, throw the whole AULA thing out the window.
What doesn't line up? both rumored to use oled screen at same resolution and both have an updated 4k output. the only issue seems to be how to get to 4k... but we aren't sure what that even means... could be marketing crap like the 8k logo on the ps5 box.
Once again, Aula was never rumored to use an OLED screen at 720p, you're confusing rumor, datamined info and speculation. Aula was datamined and it was determined that it was likely using a different type of display. Nothing said OLED, that was speculation from the hacker. Nothing at all indicated 720p screen, the hacker was wondering if the screen was 4k even.