• Hey everyone, staff have documented a list of banned content and subject matter that we feel are not consistent with site values, and don't make sense to host discussion of on Famiboards. This list (and the relevant reasoning per item) is viewable here.

StarTopic Future Nintendo Hardware & Technology Speculation & Discussion |ST| (Read the staff posts before commenting!)

Reader mode
So I’m increasingly convinced that either it’s only a small delay, or nothing’s truly changed and early (?) 2024 was always a likely release time. Enough of the 11+ devs planning releases might have encountered development issues (like basically everybody else in the industry), and if enough couldn’t hit their target dates, Nintendo might have made a decision to shift out 6-8 months.

Zelda releasing ahead of it upsets me, but it also indicates that they’re confident in what comes next, and that’s exciting. Mario almost certainly won’t just be Odyssey 2 at this stage - I vaguely recall somewhere (Kit/Krysta?) mentioning a bit of competitiveness internally between Mario and Zelda teams. I wonder what’s in store, and if it’ll take full advantage of the new hardware, even if it’s cross-gen.

As for Zelda, I can only hope that 1. they’re hitting an effective 1080p docked on base Switch; And 2. that it gets a major patch upon new console launch. I’d hate to feel like I could only enjoy the game in handheld mode, and that’s happened quite a bit for me these days.
 
karmitt said:
So I’m increasingly convinced that either it’s only a small delay, or nothing’s truly changed and early (?) 2024 was always a likely release time. Enough of the 11+ devs planning releases might have encountered development issues (like basically everybody else in the industry), and if enough couldn’t hit their target dates, Nintendo might have made a decision to shift out 6-8 months.

Zelda releasing ahead of it upsets me, but it also indicates that they’re confident in what comes next, and that’s exciting. Mario almost certainly won’t just be Odyssey 2 at this stage - I vaguely recall somewhere (Kit/Krysta?) mentioning a bit of competitiveness internally between Mario and Zelda teams. I wonder what’s in store, and if it’ll take full advantage of the new hardware, even if it’s cross-gen.

As for Zelda, I can only hope that 1. they’re hitting an effective 1080p docked on base Switch; And 2. that it gets a major patch upon new console launch. I’d hate to feel like I could only enjoy the game in handheld mode, and that’s happened quite a bit for me these days.
Click to expand...

Or most likely thing that happened was those 11 devs never had a Switch Dev Kit in the first place.
 
oldpuck said:
I've lost the tread a bit myself. The end goal is to estimate Drake's performance in DLSS 2, yes?
Click to expand...
It just establishes that the performance numbers in the various white papers are indeed consistent with the numbers in the source code snippets. The long and short of it is that if DLSS happens to use a tensor operation that accumulates to FP32, which we can't confirm, then it could potentially run twice as fast per clock per SM on Drake compared to the RTX 3000 series. I'll lay everything out, although be warned, it's pretty dry.

Hidden content is only available for registered users. Sharing it outside of Famiboards is subject to moderation.
 
davec00ke said:
Or most likely thing that happened was those 11 devs never had a Switch Dev Kit in the first place.
Click to expand...

Not sure how that’s the “most likely” outcome.

You either believe 1. A Bloomberg journalist totally fabricated a story of 11+ independent claims from sources about a 4K Switch; 2. The sources were all totally misinformed about the system and timelines they were targeting; 3. They collectively lied to him. Seems sensible.

Unless of course you’re implying something else I’m not understanding?
 
Last edited:
karmitt said:
So I’m increasingly convinced that either it’s only a small delay, or nothing’s truly changed and early (?) 2024 was always a likely release time. Enough of the 11+ devs planning releases might have encountered development issues (like basically everybody else in the industry), and if enough couldn’t hit their target dates, Nintendo might have made a decision to shift out 6-8 months.

Zelda releasing ahead of it upsets me, but it also indicates that they’re confident in what comes next, and that’s exciting. Mario almost certainly won’t just be Odyssey 2 at this stage - I vaguely recall somewhere (Kit/Krysta?) mentioning a bit of competitiveness internally between Mario and Zelda teams. I wonder what’s in store, and if it’ll take full advantage of the new hardware, even if it’s cross-gen.

As for Zelda, I can only hope that 1. they’re hitting an effective 1080p docked on base Switch; And 2. that it gets a major patch upon new console launch. I’d hate to feel like I could only enjoy the game in handheld mode, and that’s happened quite a bit for me these days.
Click to expand...

I’m thinking more of a reveal around early/middle of 2024 and release it 3-4 months after. I’m not so sure they want to lose potential holiday sales by announcing new hardware during autumn or near the end of the year.

I still think that Nintendo will ride out the Switch for as long as possible despite the SoC being seemingly ready already (?)
 
Anatole said:
It just establishes that the performance numbers in the various white papers are indeed consistent with the numbers in the source code snippets. The long and short of it is that if DLSS happens to use a tensor operation that accumulates to FP32, which we can't confirm, then it could potentially run twice as fast per clock per SM on Drake compared to the RTX 3000 series. I'll lay everything out, although be warned, it's pretty dry.

* Hidden text: cannot be quoted. *
Click to expand...
If it really really is twice as fast, it begs the the question why this only makes sense to do on Drake and not desktop cards.
 
Anatole said:
It just establishes that the performance numbers in the various white papers are indeed consistent with the numbers in the source code snippets. The long and short of it is that if DLSS happens to use a tensor operation that accumulates to FP32, which we can't confirm, then it could potentially run twice as fast per clock per SM on Drake compared to the RTX 3000 series. I'll lay everything out, although be warned, it's pretty dry.

* Hidden text: cannot be quoted. *
Click to expand...
Thank you. All of this tracks with other weirdness I had seen while trying to game out DLSS performance, but my assumption was that the accumulate instructions were only really useful during model training, not execution.

The fact that the tensor cores have been modified at all on Drake would make me assume it's for DLSS reasons . The only two reasons I could imagine Drake's tensor cores not being desktop Ampere would be "It's for DLSS performance" or "It's a holdover from Orin". Since Drake's tensor cores are different from Orins, that leans hard into it being for DLSS performance.

I've done estimates of the DLSS performance on Drake, and if you assume bog standard Ampere, 4K Quality mode is like 10ms a frame, which is basically unusable. Ultra Performance would be the only really viable 4k path in that case.
 
Mr Swine said:
I’m thinking more of a reveal around early/middle of 2024 and release it 3-4 months after. I’m not so sure they want to lose potential holiday sales by announcing new hardware during autumn or near the end of the year.

I still think that Nintendo will ride out the Switch for as long as possible despite the SoC being seemingly ready already (?)
Click to expand...

I continue to feel like this ongoing concern about protecting holiday sales is overblown, and at some point they have to value a strong push for the new hardware over wringing every drop from the current device.

End of year 7 sales demographic just cannot possibly be the same audience as new adopters. Holiday deals for OLED and the remainder of the line will be enough to give them an excellent holiday.
 
Hermii said:
If it really really is twice as fast, it begs the the question why this only makes sense to do on Drake and not desktop cards.
Click to expand...
Because Drake is much smaller than any of the desktop cards. If your card is fast enough to run DLSS in 2-3ms, having that time is a nothing win.

But if you’re running Drake and DLSS might that 5-10ms? Having that is a game changer.
 
0
Hermii said:
If it really really is twice as fast, it begs the the question why this only makes sense to do on Drake and not desktop cards.
Click to expand...
I'm not sure. Nvidia clearly could have implemented it in the RTX 3000 cards if they wanted to, since the professional GPU series does support it. I suppose the RTX 3000 cards have lots of SMs and higher clocks, so maybe it's a less useful feature there, like oldpuck suggested in the post above this one.
oldpuck said:
Thank you. All of this tracks with other weirdness I had seen while trying to game out DLSS performance, but my assumption was that the accumulate instructions were only really useful during model training, not execution.

The fact that the tensor cores have been modified at all on Drake would make me assume it's for DLSS reasons . The only two reasons I could imagine Drake's tensor cores not being desktop Ampere would be "It's for DLSS performance" or "It's a holdover from Orin". Since Drake's tensor cores are different from Orins, that leans hard into it being for DLSS performance.

I've done estimates of the DLSS performance on Drake, and if you assume bog standard Ampere, 4K Quality mode is like 10ms a frame, which is basically unusable. Ultra Performance would be the only really viable 4k path in that case.
Click to expand...
For each layer of a convolutional neural network, you can implicitly map convolution to matrix multiplication, so under the hood during inference, the ML framework is essentially evaluating many matrix-multiply accumulate operations on the tensor cores. I agree with you that I only would have expected Nvidia to use mixed precision during training of DLSS, not inference. I figured that DLSS as deployed to consumers was either pure half-precision FP16 or quantized to INT8.

Hidden content is only available for registered users. Sharing it outside of Famiboards is subject to moderation.
 
Quailman said:
Nate just dropped his 2023 prediction video for Nintendo

Click to expand...

So I've did short summaries and transcripts for the hardware portions.

9:24: NateDrake and MVG believe that Nintendo releasing a Tears of the Kingdom limited edition OLED model this year is a given.
NateDrake: Now I do want to talk about hardware, but not how some people may be thinking. I'm not talking about new hardware this year.
MVG: You're talking about a limited edition Zelda Tears of the Kingdom Nintendo Switch, which come on, there has to be.
NateDrake: That's a given.
MVG: Yeah, there's gotta be one. OLED model, there has to be one.
NateDrake: Yeah, that's a given. I saw some of the leaked package photos dating back to January. Looks legitimate. Likely to be announced on a February Direct. Make it official, give people something excited [exciting]. Likely would release towards the middle of April, because that's what we see Nintendo do with other limited edition models that come up about a month ahead of the actual game. And it's 100% expecting something like that.
Click to expand...
10:15: NateDrake thinks that Nintendo's going to do a price cut for the Nintendo Switch family of hardware based on Bloomberg's article. MVG thinks Nintendo won't do so, but thinks NateDrake has many good points on why Nintendo could do a price cut for the Nintendo Switch family of hardware.
NateDrake: The other thing I want to bring up with hardware is we've seen the reports out of Japan with Bloomberg and Nikkei suggesting that Nintendo is looking to ship 20 million+ units in the coming fiscal year. So that would be April 1st, 2023 to the conclusion of March 2024. And for a system this late into the generation to have a forecast potentially of that magnitude is something that we have not seen in many, many years. And I'm going to operate that the report is true. We won't know Nintendo's actual forecasts until May. But we're going to operate that it's true. In my opinion, the only way Nintendo is achieving those type of hardware numbers is if we see a price drop across the board of all Switch systems.
MVG: Ooh. You think there's going to be a price drop?
NateDrake: I think there has to be. I don't see how Nintendo can sell 20 million+ units if there's no price drop. I know we have Tears of the Kingdom. We have the Mario movie.
MVG: I was going to ask you don't think Tears of the Kingdom, plus whatever coincides with the Mario movie, plus the continued strength of Mario Kart is going to be, and Pokémon of course, is going to be enough for them [Nintendo] to consider a price drop? A price drop does make sense if they want to hit that amount, but a part of me wonders whether they feel like they're just going to go for it, because you know Zelda is obviously a very strong title that's going to attract a lot of people. But do you think there's just no possible way that they couldn't hit that number without a price drop?
NateDrake: If we look just at this past year, sales were down in most regions, be it, you know 15%, 20%. I know stock was an issue for a part of the year. But when I look at Tears of the Kingdom as you mention, this is a sequel to Breath of the Wild, a title that sold 28 million copies on the Switch. How many people are there waiting on the sidelines to buy a Switch to play Tears of the Kingdom who currently don't own a Switch?
MVG: Right.
NateDrake: Now I could see a case of a Switch launch owner perhaps saying "I'll buy an OLED model, or even the Zelda OLED model, to play Tears of the Kingdom." And that's where I think a price drop would come into play. If you reduce that price by $50, it now becomes more attractive where you'd say "You know what? I had my launch Switch now for 6 years. I'd like to play Tears of the Kingdom in the best way possible in handheld mode. $350 was too rich for my taste. $299 with a trade-in, and now I'm getting it down, let's just say, to $225."
MVG: Yeah.
NateDrake: Say you get $75 trade-ins. I don't know what the trade-in rates go for. You know, that's appealing now. You get the Lite to come down to $150. It's a lot more...
MVG: Yeah.
NateDrake: ...appetising. The price drop would also pair well with the Mario movie. You're going to have a new wave of kids who are going to see the movie who may not have the Switch. And if you can come in with the Lite at a price of $150, I'll even go to $175 as a ceiling price, parents will look at that and say "You know what? For that price, I'm willing to buy little Timmy and Jane this Switch, so they can play New Super Mario Bros U Deluxe or Mario 3D World and Bowser's Fury. You have to make it appealing, and with the looming possibility of a recession, and eggs being $5 to $7, and essentials going up in price, a price drop is the smart move. And in the era where we are seeing Sony and Microsoft raise the price of their hardware in other regions, Nintendo to do the opposite, it may be a gutsy move, I may be completely off case with this prediction, but it's the only way I see Nintendo achieving such hardware sales in this upcoming fiscal year, again assuming the report is accurate.
MVG: I think you say a lot of things that make sense. I personally think that Nintendo's not going to drop the price, they're going to stick to their guns. And they're going, I don't wanna say roll the dice, because that sounds like it's a 50/50, you know, chance, but I think they probably feel like they have enough here. And look, we don't know what the rest of the year looks like as well. We don't really know anything beyond Tears of the Kingdom at this point.
NateDrake: Uh-huh.
MVG: So for me, I feel like with Zelda Tears of the Kingdom, they [Nintendo] feel like with everything else they have going on this year, plus the Mario movie, that they're gonna get there without any type of price cut. But you know, we'll see. I don't really know the history of when Nintendo would price cut consoles in the past. And look, we can't really look at the Wii U as, you know, supporting evidence, because the Wii U was its own thing. (NateDrake laughs) But I think that Nintendo is...they have 2 SKUs, right? And there's obviously a big enough price difference between the two where you can just pick the one you want. And I feel like it's not going to change. But hey look, you're very right in that it [Nintendo] could easily just drop the price when the Mario movie comes out. They [Nintendo] could easily do that. They could easily take even $25 off the price, you know.
NateDrake: Yeah
MVG: And maybe have some type of bundle deal as well. There's definitely a lot of merit to what you say there.
NateDrake: That's also one way they can do it [a bundle deal]. They could disguise the price drop by introducing a new standard bundle where, let's just say come April, the start of the new fiscal year, the Mario movie is out, all Switches come with New Super Mario Bros U Deluxe, and it'll say "A $60 value!" And that disguises the price drop by keeping the sticker price the same, but you get a free game.
MVG: Yes.
NateDrake: They're [Nintendo are] profiting more on that bundle. Or they're going to get extra sales on that bundle. You think you're getting a deal.
MVG: Oh yeah.
NateDrake: And Nintendo sitting there saying "Hey, that game, we're making money."
MVG: They've done that before as well. You know, they did it with the 2DS with the Mario bundle. The Mario Kart bundle I mean. I hate to say this, but even the Wii Mini, they have that Mario Kart bundle where they bundled the Wii Mini and a copy of Mario Kart. That's a genius move. A genius move, you know? So I mean they could easily look at that again and offer some type of bundle.
NateDrake: A lot of moveability. They have a lot of legacy titles with Mario in the title where, you know, even Mario Odyssey, a game that came way back in 2017. Yes, it's still selling, but you know, now you're going to get new interest in the title. Bundle it.
MVG: Right.
NateDrake: You know, two for one type of thing, and you profit from it. So I'll be interested as to see how Nintendo does approaches their hardware plans in the calendar year as well as the fiscal year, again assuming the 20 million figure is accurate.
Click to expand...
 
Gerald said:
How powerful is ARM A78C compared to X86 Zen2 in PlayStation 5 and Xbox Series X|S?
Click to expand...
Apples to oranges.

Zen 2 has better single core performance on paper. In practice, it’s much better. Total inverse with multi core, where A78C’s modest advantage is much larger in practice.
 
Dakhil said:
So I've did short summaries and transcripts for the hardware portions.

9:24: NateDrake and MVG believe that Nintendo releasing a Tears of the Kingdom limited edition OLED model this year is a given.

10:15: NateDrake thinks that Nintendo's going to do a price cut for the Nintendo Switch family of hardware based on Bloomberg's article. MVG thinks Nintendo won't do so, but thinks NateDrake has many good points on why Nintendo could do a price cut for the Nintendo Switch family of hardware.
Click to expand...

The TOTK LE is a given given the leaks and Nintendo's pattern/history with their major titles, they even mention it. I'm not even sure Nate/MVG treat that as a prediction more as stating the obvious.

On the subject of price cuts explaining the bloomberg/Nikkei reports of increased production for next FY, I'm not sure. Switch sales have declined for 2 years in a row. That's also the projected trajectory of every console once they peak. Granted Switch is still selling at a very high rate. Assuming they beat expections, it's still going to be 19+ million on year 7. But Year 7 is not going to be 19+ again. it's going to decline 15-20%. A price cut would dull the decline this year, but are we really expecting a 21 m Switch sales for Year 7 ? They think that's unprecedented and I agree.

It could ultimately be a mix of things, price cuts, more major game releases this FY , a new SKU that isn't necessarily Switch 2, and Nintendo ordering extra parts for Switch 2 and telling the leaky supplier it's for Switch. This is just me throwing every possibility out there.

But I find Switch sustaining hardware sales, let alone increasing on year 7, to be very hard to believe. I'd like to be proven wrong, but as MVG /Nate asked, how many people in year 7 are sitting on the sidelines waiting for TOTK to release to buy a Switch. Their only answer is to get people like me with an OG Switch whose been waiting for a reason to upgrade to double dip. But I don't think there's enough of us to double dip of a Switch OLED LE or a OLED with a 50 dollar price cut. They'd need a Pro or a Switch 2.
 
oldpuck said:
Apples to oranges.

Zen 2 has better single core performance on paper. In practice, it’s much better. Total inverse with multi core, where A78C’s modest advantage is much larger in practice.
Click to expand...
Weird. I thought it was the other way round. per GHz, I thought single core performance they were nearly identical.. while AMD usually has a leg up on multi thread performance
 
Gerald said:
So Switch 2 will be have good CPU
Click to expand...
It's going to be loads better than last gen Jaguars and TX1 A57s. But it won't match zen 2. Mostly because zen 2 has high CPU clocks (desktop speeds after all). I'm just hoping the CPU gap narrows this gen at least. Last gen has PS4 CPUs about 3.5x faster than switch. I would be happy just getting at least 1.5Ghz CPU clocks on 8 cores (7 for gaming). But the higher the better.

If Switch 2 ends up releasing next year, it better have lpddr5x RAM >_>
 
Last edited:
ShadowFox08 said:
It's going to be loads better than last gen Jaguars and TX1 A57s. But it won't match zen 2. Mostly because zen 2 has high CPU clocks (desktop speeds after all). I'm just hoping the CPU gap narrows this gen at least. Last gen has PS4 CPUs about 3.5x faster than switch. I would be happy just getting at least 1.5Ghz CPU clocks on 8 cores (7 for gaming). But the higher the better.

If Switch 2 ends up releasing next year, it better have lpddr5x RAM >_>
Click to expand...
16RAM will be great, imo it will be 2GHz
 
I absolutely love all of the deep diving into source code and whitepaper documentation.
All of this has been great to speculate about what it can bring to DLSS on the new Switch, but I am also hopeful that this will give extra added performance when needed in bringing impossible ports over from PS5/Series X.

Nvidia's hardware in desktop form really doesn't make full usage of all that goes into the architecture and relies much on brute strength to win market share, but the Switch being a dedicated gaming device would have more reason to need mixed precision functions and actually use this in customized software for the platform.
 
NineTailSage said:
I absolutely love all of the deep diving into source code and whitepaper documentation.
All of this has been great to speculate about what it can bring to DLSS on the new Switch, but I am also hopeful that this will give extra added performance when needed in bringing impossible ports over from PS5/Series X.

Nvidia's hardware in desktop form really doesn't make full usage of all that goes into the architecture and relies much on brute strength to win market share, but the Switch being a dedicated gaming device would have more reason to need mixed precision functions and actually use this in customized software for the platform.
Click to expand...
the only downside will be if the hardware is announced and it doesn't align with speculation. I try to keep things at arms length for this reason.

Also the reason why i prefer an announcement even if the hardware isn't as powerful as i want than years of waiting and speculation that could mean nothing int the end. With a hardware announcement, i can see this thread really zeroing in on its capabilities and scoping out what games will play and look like.
 
ILikeFeet said:
if it's on TSMC 4N the maybe
Click to expand...
Are...or rather would they able to switch to TSMC earlier on? Perhaps that was what caused a delay? (Samsung fabs not fabbing enough chips?)

I'm still confused about the whole "commitment to Samsung fabs/being able to change to TSMC" scenario.
 
ShadowFox08 said:
Weird. I thought it was the other way round. per GHz, I thought single core performance they were nearly identical.. while AMD usually has a leg up on multi thread performance
Click to expand...
Single core performance is roughly the same, but Zen 2 pulls ahead as the clock goes higher - sustained server benchmarks tend to show advantages for Zen at high clockspeeds. The fact that Drake will almost definitely be clocked way below the other consoles exacerbates that gap.

AMD CPUs are great at multithreading. AMD can shove huge numbers of cores onto a socket, and each of those cores supports two threads. ARM CPUs only support one thread per core.

In the Zen 2 case, threads on the same core essentially can share data for free, but on different cores they have to copy data. That's the ARM advantage - the whole cluster can share data for free. In this exact situation, where it's 8 Zen 2 cores versus a single cluster of 8 ARM CPUs, that's advantage ARM on multithreading.

Desktops and Servers often have lots and lots of threads devoted to wildly different tasks on the OS side, touching different pieces of data. Video games love to have lots of threads all try to access a single, global piece of data about the game world. That's why Drake is able to make more of the multithreading advantage than you might think on paper.
 
mjayer said:
It's probably nothing but... a retailer has listed Pikmin 4 for release on May 26th...

Now, if true, why would Nintendo drop this puppy so close to TOTK... unless...

Switch 2 Launch
Click to expand...
Having two very different but still undeniably big games releasing so close to each other does feel like something you'd expect at the start of a console's lifespan, rather than towards the end. At this point, I feel that Directs have an advantage over other forms of press releases in that people actively anticipate them as a brand rather than individual presentations, so maybe Nintendo feels that they can just reveal their next system in a Direct and work from there.

Congratulations, you've managed to get me thinking that the Switch's successor could be announced next week.
 
Dekuman said:
the only downside will be if the hardware is announced and it doesn't align with speculation. I try to keep things at arms length for this reason.

Also the reason why i prefer an announcement even if the hardware isn't as powerful as i want than years of waiting and speculation that could mean nothing int the end. With a hardware announcement, i can see this thread really zeroing in on its capabilities and scoping out what games will play and look like.
Click to expand...
A hardware announcement woudnt change anything imo, it's not like Nintendo would talk about spec's anyway.

I would expect it to be Drake even in 2025 imo.
 
0
SiG said:
Are...or rather would they able to switch to TSMC earlier on? Perhaps that was what caused a delay? (Samsung fabs not fabbing enough chips?)

I'm still confused about the whole "commitment to Samsung fabs/being able to change to TSMC" scenario.
Click to expand...
I don't think there is any switch. If it's on Samsung 5nm, it will stay there. Nintendo isn't trying to win benchmarks so they may not encounter the same issues with comparisons that Qualcomm ended up with. That said, I've always believed that Drake chips would come out of their allocation of TSMC wafers
 
0
Boxle said:
I'm going in and expecting the classic.

Aonuma appears, explaining the game isn't quite ready yet, and a delay to Holiday 2023. We apologize for the inconveniences, but don't worry, we have another 50 second trailer for you.

Then Switch 2 speculation for Holiday 2023 will run rampant again, as we live in the endless now.

But no doubt a trailer rivaling Botw's Jan 2017 one would be really nice.
Click to expand...
I would hate that.
 
After the T239 Drake specification, I think that most or even all third party AAA games from PS5 and XSX|XSS will be ported to switch 2
 
Gerald said:
After the T239 Drake specification, I think that most or even all third party AAA games from PS5 and XSX|XSS will be ported to switch 2
Click to expand...
They could literally use an exact copy of the ps5 soc/ ssd and that still woudnt happen.

Edit: and there will still be games that won't be ported for technical reasons imo, especially cpu heavy games and/or games relying heavy on fast storage.
 
Hermii said:
They could literally use an exact copy of the ps5 soc/ ssd and that still woudnt happen.

Edit: and there will still be games that won't be ported for technical reasons imo, especially cpu heavy games and/or games relying heavy on fast storage.
Click to expand...
I think that the CPU will finally not be a problem, when it comes to memory, I think they will give SSD
 
Mobius said:
Wouldn't they need to add some kind of loading screen if they like want to port such game to PC?
Adding a requirement for RTX IO/whatever would be wild.
Click to expand...
I believe pc can compensate to a large degree by having higher memory requirements.

Much of the reason the consoles went with fast storage solutions to begin with, is the prohibitive cost of having a generational inchrease in ram.
 
0
Skittzo said:
Why does the fact that it uses a fanmade cover art say the date is wrong?

I mean, it's probably wrong of course but I don't see this meaning so.
Click to expand...
When I bought Sonic Frontiers on Amazon they were using a fanmade cover as well. After the game was already out. And I definitely received the real thing. Sometimes employees just google the name of the game and grab whatever looks like cover art.

So I agree with you, I'm not saying I believe the Pikmin 4 listing is accurate, I'm just saying if it isn't then it's not because of that cover art.
 
Something I'm surprised I don't recall seeing brought up before, when it comes to speculation on an OLED Lite or similar revision launching instead of an upgrade: the Lite and OLED both showed up in firmware at least 17-18 months before launch (15 months before announcement). Unless Nintendo suddenly started really caring about hiding revisions, which I doubt, and did so successfully, there's no way anything else is launching besides the eventual upgrade.
 
LiC said:
Something I'm surprised I don't recall seeing brought up before, when it comes to speculation on an OLED Lite or similar revision launching instead of an upgrade: the Lite and OLED both showed up in firmware at least 17-18 months before launch (15 months before announcement). Unless Nintendo suddenly started really caring about hiding revisions, which I doubt, and did so successfully, there's no way anything else is launching besides the eventual upgrade.
Click to expand...
I brought it up a couple of times actually. But completely agree.
 
So

1) I FINISHED THE DLSS CALCULATOR
(and it's not looking that good...)

Here it is : https://www.calconic.com/calculator...or-3000/63dec5fcd47bb900296c3ff9?layouts=true

2) the technical stuff
I wanted to give an detailed explanation about the whitepapers and the source code but @Anatole did it far better than I ever could. So go read his posts first if you want to understand what comes after. I just want to add a few points :

1) Why is full rate FP32 accumulation disabled on desktop Ampere, when it could have easily been enabled ? The answer : artificial segmentation.
Nvidia don't want AI researchers to use the gaming cards. @oldpuck made the assumption that FP32 accumulation is only useful for training; and it's higly possible this assumption is 100% true. Looking at the Internet, model training is the only use case I see mentioned. It would make sense for Nvidia, who wants to sell the AI researchers the more expensive stuff, to cripple the gaming cards in that specific task, making it much less useful for AI researchers but with no incidence on gamers since FP32 accumulation is only for training, not inference.
But that leads us to point
2) The full rate accumulation in Drake will probably be useless. It's there because Nvidia doesn't see the point in intentionally crippling Drake as it's not like it's gonna be used for AI research, but that doesn't mean that full rate will bring any actual improvement to DLSS performance.
3) (minor point) DLSS probably uses FP16 and not INT8 because in each marketing material where they talk about DLSS, they talk about TFLOPS and not TOPS. The schematics they use to illustrate the AI operations for DLSS are also the same as for FP16.

Now, regarding the calculator :
I used 2 videos from Digital Foundry : the one about an eventual Switch pro, and the one reviewing DLSS2.3 vs FSR2.0. I ended up using almost exclusively that second video, because the frame time costs were very different probably because of the different versions, and also in that first video they say some flat out wrong stuff, like DLSS cost is only related to output resolution (it's not)(Edit : actually it may be true which would indicate the 2nd video is the bogus one. Oops).
To approximate the ms cost, I based the calculator on the RTX 3080 data. The RTX 3090 data doesn't change much, and the other data point is the 2060 which doesn't benefit from sparsity, and as we don't know what would be the improvement this is a pretty bad base to estimate Drake performance. I did however use the 2060 data , combined with the 3080 data as the Ampere reference, to estimate what would be that sparsity improvement: I calculated it a 1.25x, which tracks with what @Anatole said about the performance boost likely being far from the theoretical max of 2x because of the real-time aspect of DLSS (If I understood correctly).
I do think this calculator would really benefit from more data, as what's in the DF videos is really limited and leaves a huge margin of error. I you have any idea where I can find more data, I'd love to hear it.

Edit : Just discovered new and significantly better sources of data. Will update when I can.
 
Last edited:
Paul_Subsonic said:
So

1) I FINISHED THE DLSS CALCULATOR
(and it's not looking that good...)

Here it is : https://www.calconic.com/calculator...or-3000/63dec5fcd47bb900296c3ff9?layouts=true

2) the technical stuff
I wanted to give an detailed explanation about the whitepapers and the source code but @Anatole did it far better than I ever could. So go read his posts first if you want to understand what comes after. I just want to add a few points :

1) Why is full rate FP32 accumulation disabled on desktop Ampere, when it could have easily been enabled ? The answer : artificial segmentation.
Nvidia don't want AI researchers to use the gaming cards. @oldpuck made the assumption that FP32 accumulation is only useful for training; and it's higly possible this assumption is 100% true. Looking at the Internet, model training is the only use case I see mentioned. It would make sense for Nvidia, who wants to sell the AI researchers the more expensive stuff, to cripple the gaming cards in that specific task, making it much less useful for AI researchers but with no incidence on gamers since FP32 accumulation is only for training, not inference.
But that leads us to point
2) The full rate accumulation in Drake will probably be useless. It's there because Nvidia doesn't see the point in intentionally crippling Drake as it's not like it's gonna be used for AI research, but that doesn't mean that full rate will bring any actual improvement to DLSS performance.
3) (minor point) DLSS probably uses FP16 and not INT8 because in each marketing material where they talk about DLSS, they talk about TFLOPS and not TOPS. The schematics they use to illustrate the AI operations for DLSS are also the same as for FP16.

Now, regarding the calculator :
I used 2 videos from Digital Foundry : the one about an eventual Switch pro, and the one reviewing DLSS2.3 vs FSR2.0. I ended up using almost exclusively that second video, because the frame time costs were very different probably because of the different versions, and also in that first video they say some flat out wrong stuff, like DLSS cost is only related to output resolution (it's not).
To approximate the ms cost, I based the calculator on the RTX 3080 data. The RTX 3090 data doesn't change much, and the other data point is the 2060 which doesn't benefit from sparsity, and as we don't know what would be the improvement this is a pretty bad base to estimate Drake performance. I did however use the 2060 data , combined with the 3080 data as the Ampere reference, to estimate what would be that sparsity improvement: I calculated it a 1.25x, which tracks with what @Anatole said about the performance boost likely being far from the theoretical max of 2x because of the real-time aspect of DLSS (If I understood correctly).
I do think this calculator would really benefit from more data, as what's in the DF videos is really limited and leaves a huge margin of error. I you have any idea where I can find more data, I'd love to hear it.
Click to expand...
That's really nice work!
 
Reader mode
You must log in or register to reply here.
Please read this staff post before posting.

Furthermore, according to this follow-up post, all off-topic chat will be moderated.
Last edited:


Back
Top Bottom