This is my understanding of it too, it seems like they just have an updated displayport->HDMI passthrough in the dock because it has to pass through HDMI2.1 4K signals now instead of 1080p, it's not there to do any scaling or image processing.
-
We have an update on ResetEra's ownership. More info here!
This is my understanding of it too, it seems like they just have an updated displayport->HDMI passthrough in the dock because it has to pass through HDMI2.1 4K signals now instead of 1080p, it's not there to do any scaling or image processing.
Where is that information from? Matt just confirmed it isn't Alua, and the datamine profile of Alua didn't say it was an OLED screen, that was speculation based on the way the display manages power. It is also not unusual for Nintendo to test hardware and never use it in the configuration that it was tested in.
Thanks Nate.
If you’re not a very techy person it’s all becoming a bit messy to follow now.
For someone like you, you’ll have a much better filter than I will.
People in the Bloomberg thread but guess it isn’t really based on anything.
OP
OP
Yeah, you're right. Smartphone displays are more brittle because smartphone displays are more thinner, so that smartphones are not bricks. I've edited my original post.
I also recommend checking out the OP since I try to collect all the possible relevant news and rumours in the OP.
In that case, that Aula profile doesn't make sense as a revision: why pay for such a converter but not provide any specs that make use of it? I think the explanation that it could be a stress test profile makes more sense. Aula just does not seem to be configured as a well-thought out market product.
Aula as a small revision or QoL revision could make sense. Up some clocks, 4k output, maybe some small othet improvements like more storage. Sure. However, Aula as a pro model is dumb. Needs to be thrown in the bin. If it wasnt implied TX1 production was ending I would 100% support that. (I am not saying TX1 production is ending before people freak out)
Only to reiterate what the TC said, do check the OP lol. Some questions and concerns can have an answer provided in a more concise and organized manner in the OP and should be the place to look at before asking, at least to make sure it hasn’t been covered yet. It has a good chance of brining you up to speed. Also the threadmarks!!
OP
OP
I think it's because HDMI 2.1's currently the latest HDMI spec, which is currently only supported by the PlayStation 5, the Xbox Series X|S, Ampere GPUs, and RDNA 2 GPUs, if I recall correctly. And HDMI 2.1 has native support for VRR and dynamic HDR (I imagine to a certain degree since DisplayPort 1.4's max bandwidth is lower than HDMI 2.1's max bandwidth).
As others have pointed out, those had less changes than earlier suggested in the conversation. I just want to add one more thing: cost-wise, it was a different era.
Base PS4 was made on TSMC 28 nm. The Pro version was made on TSMC 16 nm. Now I'll link to a chart of chip design costs from here.
Nowadays, one would need to put a lot more thought when it comes to what node to use.
Of course, the way to try to suppress these design costs is to re-use as many IPs compatible with the target node as you can (this article's from 2014, so it has an older chart compared to what I linked above)
Wait, I'm relatively new here. Who's Matt?
I wish I did. I still use my ALBW XL model from 2013. I should've at least picked up a refurbished new 3DS while I had the chance.
OP
OP
He's a game developer with reliable inside information.
I do feel it should mentioned, that I believe Natedrake originally said back in 2016 that the Switch would be based on Pascal technology and it ended up using Maxwell? Since he's the most reliable guy who's saying the pro will use DLSS tech I think that's something worth remembering.
I don't know if he's using different sources now when it comes to hardware stuff but I'd say it's always best to be skeptical for stuff like this. I know I don't want to let myself get disappointed if Aulas ends up being real or something because a powerful Switch Pro sounds like a dream to me.
Disregard all that if I'm mistaken and the Switch does using pascal or something based around pascal haha.
I don't know if he's using different sources now when it comes to hardware stuff but I'd say it's always best to be skeptical for stuff like this. I know I don't want to let myself get disappointed if Aulas ends up being real or something because a powerful Switch Pro sounds like a dream to me.
Disregard all that if I'm mistaken and the Switch does using pascal or something based around pascal haha.
To me it's either a testbed like you said or it just means all future Switch docks produced for either model will use the new model with HDMI 2.1 passthrough, rather than manufacturing both docks?
OP
OP
Pascal is practically the same GPU architecture as Maxwell, except fabricated at a lower process node, which is what we got in the current Nintendo Switch model, which is a die shrunk version of the original Nintendo Switch, and the Nintendo Switch Lite, since it's using the same SoC as the current Nintendo Switch model.
Last edited:
Which is why I delineate the different eras.
NES-SNES were solid machines for their time, skimped a bit here in some areas and there and mainly used off the shelf components which scaled down in cost rapidly to be quite cheap.
N64-GCN was Nintendo's ambitious 3D era, these chips were closer to bleeding edge, if not full on bleeding edge but still affordable. They were also more custom/exotic designs than the 2D machines.
Wii-DS-Wii U-3DS was Nintendo going for low end, cheap, recycled tech.
Switch is maybe a bit more like the NES-SNES era, but it's more limited hardware wise because it has to be able to run on battery power rather than just a wall outlet.
I think Lite and V2 would technically qualify as the actual Pascal since they are on the smaller node.
I see, so if I'm getting it right Nate's source was right in the past then?
Makes me feel a little better about the pro using DLSS then, even if I'm still skeptical haha.
OP
OP
I would say reliable since it's easy to mistake Pascal with Maxwell due to Pascal being the same architecture as Maxwell, except fabricated at a lower process node.
Even if the Switch Pro basically just allows say the system to run full Mariko clocks (and perhaps full 394 GFLOP undocked) + has a larger OLED screen + a 4K upscaler of some kind that improves the visuals of a game decently .... I would certainly want that model over the model Nintendo is selling now.
Now I hope it's more than that, but even if that's all it is, that would still be a more attractive SKU than the current Switch.
It’s a little complex, as an Ampere SM has quite a bit more going on inside than the Maxwell SM in Switch.
For the Maxwell SMs on TX1 in Switch, each SM can run either:
128 FP32 operations OR 256 FP16 operations (or some INT operations, I’m not quite sure what the INT pipeline is like on Maxwell)
per clock.
For Ampere, it’s a bit more complicated, partly because the config is different between A100, the desktop cards, and Orin, and we don’t know exactly what configuration Nintendo would use. For the desktop cards, it’s:
64 FP32 ops
AND
64 FP32 ops OR 64 INT32 ops
AND
128 FP16 ops OR 512 FP16 tensor ops (or a variety of lower precision ops)
For Orin, as far as I can tell, each SM would run per clock:
64 FP32 ops
AND
64 FP32 ops OR 64 INT32 ops
AND
256 FP16 ops OR 1024 FP16 tensor ops (or a variety of lower precision ops)
So if we’re to assume that Nintendo’s SoC will also use the Orin configuration, then a single SM could run both 128 FP32 and 256 FP16 ops per clock, rather than choosing between them on TX1. So, if we were to look at the 4 SM @1.1GHz config posted previously, it could in theory hit both 1.1 Tflops of FP32 and 2.2 Tflops of FP16 simultaneously, whereas on TX1 a mixed-precision workload would have had to split the GPU to get around 200 Gflops FP32 and 400 Gflops FP16.
Of course the problem is the “in theory” part. If DLSS is running, then it’s going make pretty heavy use of the tensor cores, so the FP16 performance available to shaders would drop accordingly, which is also true for any other applications of the tensor cores. There’s also going to be some amount of integer operations which would cut into the FP32 performance. Finally there could be many other bottlenecks which could limit performance aside from floating point capabilities. In PC games we don’t see a jump in performance in line with the raw Tflops numbers between Turing and Ampere, because the raw floating point performance wasn’t the only bottleneck, and when running on Ampere the game might be limited by texture units, ROPs, bandwidth, local data storage, etc., etc. In fact on PC I’d say floating point performance is probably almost never the bottleneck with Ampere so far.
On a console like Switch, I would expect developers to work around those bottlenecks as best they can if they‘re putting a reasonable amount of work into optimising a game for the device, so they may choose approaches which require more floating point performance but less bandwidth (let’s say), whereas on another system that has lots of bandwidth they may choose a different approach. I’d argue that the Ampere SM gives a lot of scope for optimisation in a console setting, as there’s a lot of floating point performance to work with, and quite a bit of flexibility on how to use it, but there are still many other potential bottlenecks that developers would have to work around.
Not my problem. They have to keep up with the tech experts on resetera.pixel artists right now
Bloomberg: Nintendo plans Switch model with 7-inch 720P OLED Display & 4K when docked ; Mass production as early as June [Dev Kits going out] News
The iPhone XR screen was over 300 dpi, which was clearly enough for most people. I have a iPhone 12 mini and my partner still has her coral coloured iPhone XR. I notice better clarity with regards to fonts/words when reading news articles, and photos look much better when you pinch-and-zoom...
www.resetera.com
Hehe, this is cute.
It's not really an upscaler. It's more just a media box focused SoC.
Wii U and 3DS were very much not "cheap, recycled" tech.
What a little shit. Lol.Bloomberg: Nintendo plans Switch model with 7-inch 720P OLED Display & 4K when docked ; Mass production as early as June [Dev Kits going out] News
The iPhone XR screen was over 300 dpi, which was clearly enough for most people. I have a iPhone 12 mini and my partner still has her coral coloured iPhone XR. I notice better clarity with regards to fonts/words when reading news articles, and photos look much better when you pinch-and-zoom...
Hehe, this is cute.
I’m going to - based on nothing - interpret “neat” as having some cool features but not being some mind blowing change. Fine by me if I can put my launch model safely away.
he has implied that it's akin to the n3DS, but it's worth for both primarily handheld and primarily docked players to upgrade
The New 3DS was a great upgrade. Sure, I’d have loved some more game’s to utilize the extra power, but there was so much improved that it was worth the upgrade (for me). I’d certainly appreciate a better WiFi chip, but I can’t imagine an update doing much to improve the OS like the New model did.
....oh no! Faceplates are cute!!
3DS is a pretty poor chip with the 3D screen gimmick, Wii U was kind of a nightmare of all designs, by having to stick to the Wii architecture they made themselves a monstrosity of a system that was recycled and poor performance, but not even cheap.
Still I put the Wii/DS/Wii U/3DS as one contiguous era mainly under Iwata's watch.
This is very interesting, thanks! The modularity is quite intriguing, and the scenarios you sketch make a lot of sense. It might also be boon for native handheld rendering, considering we have a 720p screen, so DLSS applicability is limited there. In that case, you could use the Tensor cores to accelerate your FP16 computations to help with rendering. There could help the Ampere chip punch above its FLOP-for-FLOP weight vs. The Maxwell architecture. And if devs can optimise around the tensor cores, it could punch above its weight even more. Pretty exciting idea (assuming it's being used in the next system, of course).
(This goes into the weeds, I don't blame anyone for dismissing it) But as you say, it can be used in many different ways. Even in docked, if 30% of the operations are FP16, you would use the Tensor cores for only 15% of the time needed for the ALU's FP32 computations. If you were to do DLSS at 30 FPS, for example, you could have 33 ms spent on FP32 in the ALU, 5 ms on FP16 computations in the tensor cores, and 28 ms on DLSS (considering that the uArch allows DLSS and native rendering to overlap). This would then correspond to a 25 fps performance on a comparable (isoclock, iso-SM) Maxwell architecture (ignoring other potential architectural improvements), and furthermore applies DLSS to the images, potentially boosting a theoretical 1080p image to 4K. So we have better fps, 4k DLSS'ed vs. 1080p, and all of that at the same theoretical FP32 FLOPS number. Of course, this is probably not realistic, as other bottlenecks and timing issues might interfere, but I think these general computations outline the theoretical architectural improvements correctly, right? The uArch might have some bookkeeping overhead that takes away from this, of course.
DLSS should be doable within the 28 ms time slot I outlined above I think, so an Ampere system could simultaneously boost the fps of a game and boost it to 4K via DLSS (for a 30 fps update, 60 fps is probably a bit tougher due to the shorter buffer time), all at the same FP32 FLOPS count. Now add in a boost in theoretical FP32 FLOPS (highe SM count and clock speed), and you have a very potent GPU architecture.
Now, for some ultimately useless and contextless math: Assuming this scenario, and assuming bandwidth limitations, the Ampere architecture would have an advantage of compared to Maxwell architecture given by (1 + x_16 / (2*x_32) ),assuming that the share of computations in FP32 (denoted x_32) is larger than half of the FP16 computations: x_32 >= 0.5*x_16. For a 30% FP16 share, Ampere would have a 1 + 0.3/1.4 ~= 1.21x speed advantage (it can do 1.21x as much work) at similar clock per SM. Of course, this is assuming no other bottlenecks.
OP
OP
I think he's referring to Matt's comment that you posted.
I don't think he said anything about n3DS, just responded to people saying that it is worth the upgrade for both handheld and dock focused players.
Bloomberg: Nintendo plans Switch model with 7-inch 720P OLED Display & 4K when docked ; Mass production as early as June [Dev Kits going out] News
Switch barely handles 1080p. Native 4K ain’t happening. No one is talking about native 4K.
www.resetera.com
might not mean much since it's only in relation to to the n3DS. but it was still a pretty big upgrade, and the Pro seems to be bigger than that
He said the Switch Pro was "neat", and someone responded that the New 3DS was neat, and they were hoping for more of "nice".
To that he responded with "It's neat".
Bloomberg: Nintendo plans Switch model with 7-inch 720P OLED Display & 4K when docked ; Mass production as early as June [Dev Kits going out] News
Switch barely handles 1080p. Native 4K ain’t happening. No one is talking about native 4K.
www.resetera.com
Bloomberg: Nintendo plans Switch model with 7-inch 720P OLED Display & 4K when docked ; Mass production as early as June [Dev Kits going out] News
Switch barely handles 1080p. Native 4K ain’t happening. No one is talking about native 4K.
might not mean much since it's only in relation to to the n3DS. but it was still a pretty big upgrade, and the Pro seems to be bigger than that
omgitshappening.gifHe said the Switch Pro was "neat", and someone responded that the New 3DS was neat, and they were hoping for more of "nice".
To that he responded with "It's neat".
Bloomberg: Nintendo plans Switch model with 7-inch 720P OLED Display & 4K when docked ; Mass production as early as June [Dev Kits going out] News
Switch barely handles 1080p. Native 4K ain’t happening. No one is talking about native 4K.
New 3DS almost tripled the CPU and doubled the RAM. I really doubt Switch will get similar amounts of bump.
And it definitely was not "neat", it was awesome. But GPU constraints and third party absolutely hating the system made it kinda useless.
2.5 Ghz CPU - 8GB RAM Switch would have been absolutely blast but unrealistic as hell.
And it definitely was not "neat", it was awesome. But GPU constraints and third party absolutely hating the system made it kinda useless.
2.5 Ghz CPU - 8GB RAM Switch would have been absolutely blast but unrealistic as hell.
English isn’t my native language. Is neat more positive than nice?
If it was a 8 core A78, even at the same 1Ghz it's be way over a 3x jump. More like 6x.
Also 8GB isn't unrealistic in any fashion, anything higher I would agree, but just 8GB is not a difficult thing to procure these days on the market.
It is used differently depending on the region, but in this context it refers to an approval of the thing in question, in a more positive manner. It’s acceptable, great, good, etc. It is synonymous with nice in this sense.
no. Matt is being nondescript. his comment technically means nothing
Thank you both. And is Matt a trustworthy insider? I had never heard about him until now.
Neat isn’t a word used for something dealing in overjoy, but I don’t see him saying “neat” was hinting at anything at all anyways.
OP
OP
I agree that the frequency of almost all the CPU cores being 2.5 GHz is unrealistic. But I think 8 GB of RAM is definitely plausible.
Unrealistic as in, wouldn't that increase overall cost of the console a lot? 4 - 8 GBs of RAM difference was at least $200 of price change in laptops back then.
Lmao. Matt said as a handheld only guy he is there day one and it will be worth it for docked players too. That neat comment is him saying "I said what I said"
Nate already told us all this
Nate already told us all this