Eclissi
Moblin
... and General Direct was in March (2025)!
-
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.
-
Do you have audio editing experience and want to help out with the Famiboards Discussion Club Podcast? If so, we're looking for help and would love to have you on the team! Just let us know in the Podcast Thread if you are interested!
StarTopic Future Nintendo Hardware & Technology Speculation & Discussion |ST| (Read the staff posts before commenting!)
- Thread starter Dakhil
- Start date
lattjeful
Like Like
- Pronouns
- he/him
Not sure why you're so confused. Everybody knows the alphabet goes A, B, C, E, D...
Eclissi
Moblin
Not in my alphabet
oldpuck
Bob-omb
- Pronouns
- he/they
I had some anxiety about seeing your reply because when I woke up this morning, I thought I'd been kind of a dick. Sorry about that.
I actually misunderstood what you meant about FP4 dual issue. I thought you were referring to the transformer engine in the tensor cores - which can convert some types to fp4 for performance improvements during model training - not dual issue in the shader cores. That was my mistake.
I think it feels like Nintendo could have gotten more, because we're not looking at "what chip could Nvidia have delivered by 2025" and instead looking at "all the technology that's available if you started in 2025". But Orin came out a few months before Ada, it came out after the Cortex-A715, it came out a year after LPDDR5X, two years after Apple silicon hit 5nm.
Yet it is still an Ampere, A78, LPDDR5, 8nm chip. And it costs $2,000! This is not a price constrained device! That's because integrating each of these separate technologies into a single, stable chip, along with the custom hardware built for it, takes time and money. A state-of-the-art SOC will always lag behind the state-of-the-art-individual-components. There is a reason that the GPUs in Sony and Microsoft's consoles are secretly still RDNA1 under the hood despite releasing the same time as RDNA2.
12 months of additional development time would not have changed the price considerations on the node, when you're talking about a device that Nintendo is planning to keep for 7 years anyway. 12 months of additional development time would not have changed the CPU choice, since that's maxed out by backwards compatibility. 12 months of additional development time might have baked in a minor upgrade to the memory, but it wouldn't have changed bandwidth targets, which would be based on the node (and it's power consumption) and the needs of the GPU. The GPU size is dictated by the node, and Nintendo's performance target.
The GPU architecture is a halfway design between Ampere and Ada. Perhaps it could have gone "full" Ada, but beyond what we've already got, that's only three updates. A minor, researcher targeted, update to the tensor cores. A minor update to the RT design. Frame gen, which is already not viable.
These things, to me, do not constitute a gap between "long lived device" and "woefully out of date"
CalmMind
Unconventional wisdom
I have seen shook Nintendo. We all have. Nothing about what's now happening behind the scenes or in plain sight is giving off fear and uncertainty. The only panic I see is from forum dwellers and people who are chronically online.
Lancelot
Like Like
Don't worry Oldpuck, it's okay. This is hardware speculation in the end of the day, people are gonna preach what they believe in and stick to it. You know how much I appreciate your posts and i'm sure the others do as well!
So, essentially... It feels like Nintendo could've went for more because of its already delayed release date, and nearly three full years is quite a bit on computer timescale for people. The technology might go all the way back to 2022 technically, but getting it into an SoC (also power constrained) in the first place still takes time, right? I must admit, when do you see it from this perspective... Maybe Nvidia really couldn't offer anything better other than an Ampere+ device, and that's reassuring at least! I assume the posters reading eased their concerns as well, if they had any.
Now, on a related note I suppose... As far as I know, dual issue shader cores are a full Ada feature, but there's really no chance for them to be backported in an hypothetical scenario? The compute left on the table for a closed system is pretty significant, and Nintendo clearly need every single trick under its sleeve. Imo, it's the only reason full Ada might have been worth it, although i'm equally curious about the tidbits T239 already has (other than the potential node).
Shoulder
Like Like
I think it'll depend really.
Programmers might be forced to get more creative with their code to run more efficiently, and/or find new ways to reduce the computational requirements. Some limitations though might also force the innovation of new types of processors with more fixed-function hardware built in that'll process information at a much faster, and more...efficient rate while sipping power (which we already see with Tensor, and Neural cores for example).
And expanding a little, we're already seeing the beginnings of a shift with the resurgence of Analog Computing, which technically never left, but is now starting to be reintroduced for certain instructions. Compared to a regular digital computer that'll crunch some computations using 100watts of power, an Analog computer could do it in roughly the same amount of time, but at a fraction of the wattage. This doesn't apply to every scenario mind you, but it shows analog chips will have a future as Moore's Law slows down.
EDIT: Here's an example of what Analog Computers can do, and its history. Take it away, Derek!
Last edited:
oldpuck
Bob-omb
- Pronouns
- he/they
Thanks! I really appreciate your input as well.
Basically. Nvidia technically is saying they'll have a Blackwell SOC around the same time that Blackwell comes out - but that is the Atlan/Thor projects which has been very troubled, and I would take what they're saying with a grain of salt.
Maybe Nvidia could have offered a really accelerated SOC development process to Nintendo, and gotten them a Blackwell based SOC in late 2025. But that wouldn't be a Nintendo console in 2025, they'd still have to finalize the hardware around the chip. So a late 2025 SOC probably would have made a 2026-2027 piece of hardware.
Sony and Microsoft are really good at branding the technology. Like, Microsoft made a big deal about having "real" RDNA 2 inside the Series S and X. But under the hood, they don't actually. They have RDNA1 with backported RDNA 2 features, because they needed to spend time integrating their chip while RDNA 2 was still baking. Note that they don't have Zen 3, which released at the same time. Integrating 2 beta technologies, plus the custom Microsoft hardware (the decompression block, the custom audio block, the ML block) was probably a bridge too far.
Very similar to what Nintendo is getting with Ampere + backported Ada. Just Nintendo is never going to talk about the underlying tech like that, which lets the narrative get defined by Sony+MS.
I think Nvidia could offer something better, but I don't think they could have done it at the same cost. Potentially if they'd gone fully last minute, and shelled out the cash to really rush the project, they might have been able to offer a 3nm based chip. But it would have been really expensive, even before adding the "rush job" costs and consequences.
At the same cost, same power budget, I think the improvements to Drake would be limited to a few of the more minor Ada niceties.
I'm honestly not sure. I don't believe it can be backported, though I don't think it's useful if it was. But I could be misunderstanding the tech. @Anatole is probably the person here
My limited understanding is that Hopper/Ada/Blackwell provide some lower precision datatypes, and a transformer engine - a way to use high precision data-types for part of AI training and the faster, lower precision datatype later, when the loss of precision has little impact. This is primarily a training optimization
I've seen FP4 referred to as an acceleration for MoE models during inference, as in when the model is executing, not being trained. This is even further outside my wheelhouse, but these models are extremely large. This is about taking models so large their execution time is measured in seconds, or even minutes, and making them faster. Stuff like ChatGPT, large language models. I don't know that these models are ready for real time graphics operations, and if they are, I doubt they would ever run on something as small as Drake.
omniryu
Bob-omb
- Pronouns
- He/him
Lol but I am not getting upset. In fact, i find this place very informative and a bit humorous.
ziggyrivers
thunder
- Pronouns
- He/him/his
It’s gonna be a boring time if video games take longer and longer to develop. Focus should go from realism to R&D in simplifying development times
KarlRemarx
Cappy
I love how Nintendo not attending Gamescom has now been spun into an all-out "boycott" because Nintendo are "frightened", "panicking", and "self-sabotaging", and that Gamescom is no longer just a consumer trade show but also apparently the only possible communication method Nintendo has with 3rd parties, all of a sudden. Again - all of this on the basis of a single report that delayed Switch 2 from one side of Christmas to the other, which cites absolutely fucking nothing about Nintendo arbitrarily trying to hold back their own successor hardware.
It couldn't be, I don't know, that Nintendo don't particularly want to spend resources attending Gamescom and having to produce showfloor demos of the last handful of Switch 1 games, when at that point they will be going into the last half-year of development before Switch 2.
Aufhebung RPG
Boo
I can't understand it as well, because it seems to me that some of the familiar faces under that thread are exaggerating the significance of gamescom for Nintendo, and please note that the only thing that means anything to Nintendo is their own press notifications with ND, everything else is optional.And the strange logic that canceling attendance at gamescom equates to a delayed release of a new console is even more implausible, and I even wonder why so many Nintendo veterans don't get the whole thing that Nintendo is getting shorter and shorter on the announcement period for new console?Similarly, I'd go so far as to say that the claims of new console delays are coming from third-party media outlets and various Leakers who can't prove their credibility, who have taken it upon themselves to assume that Nintendo will be releasing new consoles in the fourth quarter of 2024, and then have taken it upon themselves to state that Nintendo's new consoles have been "delayed" to the first quarter of 2025.
The 2024 launch of the new console is entirely a result of various third party media and rumors based on the timing of console replacements over the years and thus speculation as to when Nintendo will choose to replace them, and of all the official information,.only Furukawa has said that support for the switch will be maintained until March 2025, so the answer is obvious, there are far too many people who are self-deceived.
The 2024 launch of the new console is entirely a result of various third party media and rumors based on the timing of console replacements over the years and thus speculation as to when Nintendo will choose to replace them, and of all the official information,.only Furukawa has said that support for the switch will be maintained until March 2025, so the answer is obvious, there are far too many people who are self-deceived.
Last edited:
lattjeful
Like Like
- Pronouns
- he/him
Isn't that kinda the primary benefit of real time raytraced lighting? It just so happens that it also looks really pretty.
And is crazy expensive, computationally.
Giancarlo
Nintendo connoiseur
theres a inovation barely focused on it consoles,that is VR/AR, the Nintendo Switch sucessor or the sucessor of sucessor of the Nintendo Switch(Switch 3) could focus on VR/AR as it main concepth/gimmick
Goodtwin
Like Like
Nobody needs SNG revealed more than the full time YouTubers. These channels cant help but to create daily videos that focus on Nintendo's next gen hardware, even when we are absent any new details. There just isn't much enthusiasm for the withered Switch platform at this point, and the completely lack of compelling high profile software on the release schedule makes it all the worse. Nothing wrong with a couple remakes, but that doesn't raise the hype meter in a meaningful way.
I was watching a clip from DF discussing a portable Series S handheld from Microsoft, and Rich pointed out that the Series S pulls around 80 watts. The Series S APU is on TSMC 7nm, and even with a couple of node shrinks, it would likely still be around 20 watts, best case scenario. They pointed out the Rog Ally, which already uses an advanced 4nm APU (Ryzen Z1), and that unit has a TDP or 15w with an option for the a 30w Turbo mode. Even when pulling 15-30 watts, this APU is not on par with the Series S, still falling well short.
I was watching a clip from DF discussing a portable Series S handheld from Microsoft, and Rich pointed out that the Series S pulls around 80 watts. The Series S APU is on TSMC 7nm, and even with a couple of node shrinks, it would likely still be around 20 watts, best case scenario. They pointed out the Rog Ally, which already uses an advanced 4nm APU (Ryzen Z1), and that unit has a TDP or 15w with an option for the a 30w Turbo mode. Even when pulling 15-30 watts, this APU is not on par with the Series S, still falling well short.
while realism has a higher ceiling, I think we got to remember the reason we always see realistic art styles when pushing technology is because our measuring stick is real life. how close we get to that is what was used to define progress. and all those tools feed other styles
like Hi-fi Rush, which I wouldn't be surprised if people used that as an example of "simple art styles lead to simple development". but that game used every feature that realistic games did
simply put, all this push for realism is coming from simplifying development times. GTA6 probably wouldn't be much faster or cheaper to make if it looked like Hi-fi Rush
Aufhebung RPG
Boo
Is it biased for DF to use native rendering performance to measure switch2, especially since switch2 will have many extensions from nvidia the expressiveness is close to xss I don't think it's a hard thing to do.Nobody needs SNG revealed more than the full time YouTubers. These channels cant help but to create daily videos that focus on Nintendo's next gen hardware, even when we are absent any new details. There just isn't much enthusiasm for the withered Switch platform at this point, and the completely lack of compelling high profile software on the release schedule makes it all the worse. Nothing wrong with a couple remakes, but that doesn't raise the hype meter in a meaningful way.
I was watching a clip from DF discussing a portable Series S handheld from Microsoft, and Rich pointed out that the Series S pulls around 80 watts. The Series S APU is on TSMC 7nm, and even with a couple of node shrinks, it would likely still be around 20 watts, best case scenario. They pointed out the Rog Ally, which already uses an advanced 4nm APU (Ryzen Z1), and that unit has a TDP or 15w with an option for the a 30w Turbo mode. Even when pulling 15-30 watts, this APU is not on par with the Series S, still falling well short.
tetsugo
Rattata
Nintendo is one of the few publishers in the world that has the full-range of games, from smaller scale games like Kirby's Dream Buffet and F-Zero 99, to full AAA games like Zelda and 3D Mario. I believe this helped A LOT Nintendo on this gen filling the gaps in the schedules, unlike Sony for instance, that don't has any games to release this year.
I believe people forgets that for Nintendo be able to do some marketing in an event like this, is needed prior preparation like billboards, standing material, play booths, and all of this not just costs money, but time and outsourced material which could lead to leaks. To attend in an event like this, the marketing material should be done for games announced, idk, 2-3 months before? Even if we have a June direct, they wouldn't have time to prepare the material to Gamescon for games announced in June, and the cost to get just evergreen games to this event wouldn't make up for it.
Imagine how crazy would it be if these YouTubers played and reviewed actual games?
navierStokes
Tektite
I have to say, Pikmin 5 better needs that real-life fruit in there, because it's getting closer and closer. I can almost smell the pineapple and mango.
Shoulder
Like Like
I think we're approaching a threshold of the limits of how long someone wants to develop a game. Once you get past that 4-5 year mark, I think similarly how as graphics get more real, it’s diminishing returns at that point. I think game development follows a similar pattern.
At what point does a game developer, after working on the same fucking game for 8-10 years just goes, “Why the fuck am I doing this?” On the flip side, those perfectionist types that are willing to work on it as long as it takes, and the game lives on forever as a symbol of how to finish a game (I.e. Zelda TOTK), or it is revered for how good it is, there maybe something to this.
It used to be the limits of technology were limiting developers' vision on what they wanted, and they had to adapt and be creative. Nowadays, they are not bound by those same limitations anymore, so I feel they try to cram as much as they can.
Replace the word “scientists” with ”developers,”and you get a similar result:
Aufhebung RPG
Boo
The Legend of Zelda production team is precisely the result of playing with gameplay ideas in a limited amount of time rather than cramming in as much stuff as possible, and this development model you're thinking of doesn't work for Nintendo because for the Legend of Zelda development team, they have no more time than a lot of top-tier 3A games, maybe even less, so they have to focus on what they're good at.I think we're approaching a threshold of the limits of how long someone wants to develop a game. Once you get past that 4-5 year mark, I think similarly how as graphics get more real, it’s diminishing returns at that point. I think game development follows a similar pattern.
At what point does a game developer, after working on the same fucking game for 8-10 years just goes, “Why the fuck am I doing this?” On the flip side, those perfectionist types that are willing to work on it as long as it takes, and the game lives on forever as a symbol of how to finish a game (I.e. Zelda TOTK), or it is revered for how good it is, there maybe something to this.
It used to be the limits of technology were limiting developers' vision on what they wanted, and they had to adapt and be creative. Nowadays, they are not bound by those same limitations anymore, so I feel they try to cram as much as they can.
Replace the word “scientists” with ”developers,”and you get a similar result:
Shoulder
Like Like
I should clarify I wasn’t taking a dig at Nintendo specifically. It’s mostly all the other developers who want to cram as much in an open world not realizing they may not be compatible with each other.
I was using TOTK as an example of what developers SHOULD be doing given the kinds of development times we're seeing. There are games with longer development that don’t have the polish, nor the consistency of what Tears of the Kingdom managed. Like you said, it’s about working within their constraints, and focusing on what they’re good at.
Quite honestly, I think games like Ass Creed for example aren’t really good at anything anymore besides having a beautiful looking world.
Aufhebung RPG
Boo
Yes, I quite agree with this, the development of major North American big games after the 7th gen consoles has become a dilemma of going for more and more so-called big, yet more and more mediocre, whether it's Bethesda or Ubisoft or Sony, the development cycle will always be limited, and the pursuit of everything will only lead to more and more boring and uninspired games.
Anatole
Octorok
- Pronouns
- He/Him/His
FP8 is tricky and still fairly new; FP4 is even trickier. You need to allocate bits for 3 different things in floating point precision: the sign, the exponent, and the mantissa, which is the actual number you care about. For example, in base 10, 0.5578 * 10^-2, the sign would be +1, the mantissa is 0.5578, and the exponent is -2. In base 2, you can always assume that the leading digit is 1, so for example, in 1.011011 * 2^-4 (base 2), you only need to store the mantissa (.011011).
FP8: 8-bit floats
Depending on how you allocate the bits, you can emphasize precision or dynamic range. For example, if you allocate 4 bits for the exponent and 3 bits for the mantissa, the largest normal number you can represent up to the sign is 1110.111 (base 2), which is on the order of 2^8 (because of exponent bias, and allocating a few numbers for things like NaN and infinity), while the smallest normal value is 0001.000 (base 2), on the order of 2^-6 (any values with an exponent of 0000 are considered subnormal).
If you instead use 2 bits for the mantissa and 5 for the exponent, the largest and smallest normal numbers respectively become 11110.11 (base 2), on the order of 2^15, and 00001.00 (base 2), on the order of 2^-14. So increasing the number of bits for the exponent enhances the dynamic range, at the expense of precision. Lots more info here: arXiv
You are right about what the transformer engine does; generally you need more dynamic range during training and more precision during inference. It's also rescaling the loss function so that the gradients fall within a range that can be represented by the precision. Nvidia's blog post for FP8 explains how calculating the maximum value for each tensor just in time for scaling would be slow, so they instead save the max value history and use that history for scaling.
FP4: 4-bit floats
As for FP4, I am extremely skeptical about it. All the issues of FP8 are pushed even further. I don't know if the details for Blackwell are public, but allocating 1 bit for the exponent essentially gives you an unevenly distributed version of INT4, so I assume they are using 2 bits for the exponent and 1 bit for the mantissa. The list of numbers you can represent is small and finite (with positive and negative versions of each of these, and an exponent bias of 1):
- 11.1 [1 + 1 * 2^(-1)] * 2^(3-1) = 1.5 * 2^2 = 6 (may also be reserved for NaN or infinity)
- 11.0 [1 + 0 * 2^(-1)] * 2^(3-1) = 1.0 * 2^2 = 4 (largest normal number)
- 10.1 [1 + 1 * 2^(-1)] * 2^(2-1) = 1.5 * 2^1 = 3
- 10.0 [1 + 0 * 2^(-1)] * 2^(2-1) = 1.0 * 2^1 = 2
- 01.1 [1 + 1 * 2^(-1)] * 2^(1-1) = 1.5 * 2^0 = 1.5
- 01.0 [1 + 0 * 2^(-1)] * 2^(1-1) = 1.0 * 2^0 = 1 (smallest normal number)
- 00.1 [1 + 1 * 2^(-1)] * 2^(0-1) = 1.5 * 2^-1 = 0.75 (subnormal)
- 00.0 (reserved for 0)
Exponent bias is 0, at least, the way exponent bias is typically defined.
• 1.11 [1 + 1 * 2^(-1) + 1 * 2^(-2)] * 2^(1) = 1.75 * 2 = 3.5 (may also be reserved for NaN or infinity)
• 1.10 [1 + 1 * 2^(-1) + 0 * 2^(-2)] * 2^(1) = 1.50 * 2 = 3
• 1.01 [1 + 0 * 2^(-1) + 1 * 2^(-2)] * 2^(1) = 1.25 * 2 = 2.5
• 1.00 [1 + 0 * 2^(-1) + 0 * 2^(-2)] * 2^(1) = 2 (smallest normal number)
• 0.11 [1 + 1 * 2^(-1) + 1 * 2^(-2)] * 2^(0) = 1.75 (subnormal)
• 0.10 [1 + 1 * 2^(-1) + 0 * 2^(-2)] * 2^(0) = 1.50 (subnormal)
• 0.01 [1 + 0 * 2^(-1) + 1 * 2^(-2)] * 2^(0) = 1.25 (subnormal)
• 00.0 (reserved for 0)
• 1.11 [1 + 1 * 2^(-1) + 1 * 2^(-2)] * 2^(1) = 1.75 * 2 = 3.5 (may also be reserved for NaN or infinity)
• 1.10 [1 + 1 * 2^(-1) + 0 * 2^(-2)] * 2^(1) = 1.50 * 2 = 3
• 1.01 [1 + 0 * 2^(-1) + 1 * 2^(-2)] * 2^(1) = 1.25 * 2 = 2.5
• 1.00 [1 + 0 * 2^(-1) + 0 * 2^(-2)] * 2^(1) = 2 (smallest normal number)
• 0.11 [1 + 1 * 2^(-1) + 1 * 2^(-2)] * 2^(0) = 1.75 (subnormal)
• 0.10 [1 + 1 * 2^(-1) + 0 * 2^(-2)] * 2^(0) = 1.50 (subnormal)
• 0.01 [1 + 0 * 2^(-1) + 1 * 2^(-2)] * 2^(0) = 1.25 (subnormal)
• 00.0 (reserved for 0)
DLSS and lower precisions (FP8/FP4/INT8/INT4)
If DLSS were hypothetically implemented with FP4 in the most optimal case, it would run at 4 times the speed of FP16. However! As with FP8, this ideal case will not work if the dynamic range of the tensor you are trying to represent is outside the inherent range of the precision you are working in. With FP4, I bet this criterion would fail more often than not. There's a reason that INT8/INT4 quantization usually happens after training, or with "quantization aware" training in FP16; INT8/INT4 simply do not have the dynamic range to represent the gradients needed to train a typical network.
For many ML image applications, the final output doesn't have any dynamic range or precision. For example, in classification ("tell me if this image is a cat or a dog"), the final few layers aren't usually even convolutional. In segmentation ("generate two binary maps, one labeling all dog pixels and one labeling all cat pixels"), the situation is similar. In this situation, it's possible that FP4 might work, but I don't imagine the results would be much better than INT4. FP8 is a far superior option.
However, for reconstruction like DLSS, the final output is not a label or a binary map; it's an image, so human perception matters. You need to be able to propagate the dynamic range and precision of your input images to your output image. If any of your internal layers loses precision or dynamic range, then your output layers are hallucinating (making up from scratch) all of the fine detail, which isn't something that Nvidia wants. Ideally, the actual pixel samples would be represented in the final, reconstructed image. It's for this reason that DLSS runs at FP16 for HDR inputs and INT8 for tonemapped inputs; if you want an output that looks good, you need the precision and dynamic range that FP16 nets you.
tl;dr
I don't think FP4 will ever be feasible for DLSS. Honestly, I doubt it has any real utility other than pumping up the FLOPS numbers in Nvidia's marketing. There is a chance that FP8 could be workable for DLSS with careful rescaling at each layer, and I'm curious to see if Nvidia tries it. But dynamic range and precision are both important for reconstruction, so you ultimately need to choose a floating point precision that captures both.
King_Dodongo
Bob-omb
Realism isn't everything.
Games like Okami still look prettier than many others that strive to detail human pores.
Games like Okami still look prettier than many others that strive to detail human pores.
Shoulder
Like Like
Hell, Hollow Knight is only a mere Indie title made by a team of less than a half dozen people, and yet looks, and plays better than 80% of the shit that’s come out since.
ItWasMeantToBe19
Manakete
This is a really detailed and good post.
Still would like the option D:
(In case breakthroughs do happen or FP4/8 is used for other stuff like maybe VRAM decompression)
Wockio
Kikwi Enthusiast
- Pronouns
- He
Hollow Knight looks like a flash game
NineTailSage
Bob-omb
I do find it funny how most of the big publishers and developers left the AA industry behind for the big budget titles.
Which left lots of room for indie developers to step right up and make their games fit into this space.
During the 360/PS3/Wii era indie games were relegated to very small gimmick like cell phone games and now we have things like Hades or Hollow Knight, which can stand up there with the best of them...
oldpuck
Bob-omb
- Pronouns
- he/they
I would say that's not true, but also, is it a bad thing?
Aufhebung RPG
Boo
It was Japanese games that were pushed out of the mainstream in the 2000s by the massively industrialized production model of big games in North America, but in retrospect one can see that this process quickly morphed into throwing good money after bad, and with Nintendo's return to full force in the 8th generation and the general mediocrity of big games and the slowing down of the development of visual effects, the mainstream market began to place a renewed emphasis on the strengths of Japanese games, namely gameplay, whileNorth America also entered an era of highly segmented markets along with the general rise of indie games (mostly thanks to steam).
oldpuck
Bob-omb
- Pronouns
- he/they
This is fantastic, thank you! Nvidia seems to be pushing the FP4 support specifically for generative LLMs, so perhaps there is some super narrow use case there? Regardless, they're only talking about it in relation to Blackwell, and right now all their public Blackwell info is about the datacenter chips.
It'll be interesting to see if Nvidia continues to segment out some of these features for enterprise customers, where the margins are still high, or let them flow down into consumer products.
omniryu
Bob-omb
- Pronouns
- He/him
I wonder if anime helped out. I mean we had a couple of big boom in anime. The overall trend in Japanese culture I think rose.
But I need to see there pores especially in Madden. Where I pay for the same crap every year? Who will take my money for pores?!
Shoulder
Like Like
Right? Even if it was one, it'd still be not only the best looking, but also the best flash game ever made. And at the same time, deserving of a full-fledged release on a console.
Truth be told, I got out of flash games in the early 2000s during when Newgrounds, and ebaumsworld were still the shit. So I pay zero attention to the current state of flash games.
Aufhebung RPG
Boo
The last data I saw was that Japanese anime in the streaming era is still mainly for the local market, but the US is the second largest market, but if you look at it from my perspective Japanese anime is not enough to support its gaming status because Japan is the founding country of the rules of the modern gaming industry, and Japanese anime is very local in nature, with a fair amount of anime that is only popular in East Asia (China Japan South Korea) such as anime likeRobot animation, daily type of animation, etc. in North America no one asked, but the Japanese game even the most downtrodden time is also the world's second.
redmutineer75
Bob-omb
- Pronouns
- He/Him
The main application that FP4 is used for is in training super sparse data sets. No need for those extra bits when most of your data is all zeroes.
ItWasMeantToBe19
Manakete
Is any of this true, lol.
Like, gaming is completely dominated by Fortnite and a bunch of really shitty titles (Roblox, EA FC, Madden, 2K, Call of Duty, mobile shovelware, porn garbage like Nikke, Genshin, etc).
None of the dominating games that control most of the revenue of the game industry are from Japan, idk.
And all of these games are wildly expensive to make.
So the games that are the most popular games of all time largely do not have very good graphics, gameplay, or story but do probably have budgets of at least $300m each so idk.
Last edited:
Aufhebung RPG
Boo
However, Nintendo does have a close relationship with Japanese animation, and the founder of its art style is Kotabe Yoichi, who collaborated with Hayao Miyazaki on many animated films and is Miyazaki's predecessor, so whether it's The Legend of Zelda or Mario or you can say that the entire Nintendo art style is basically a combination of Toei Animation and Ghibli + Disney.
Aufhebung RPG
Boo
I see what you mean, what I mean is that if you only look at traditional buyout games, the fact that Japan returned to the mainstream after 2015 is an easily observed fact, and 2015's MGSV Splatoon Bloodborne basically kicked off the full return of Japanese gaming, but as you said, in the pc and console side of the service based gaming segment Japan is stillLackluster, as Japanese online gaming was already beaten by South Korea & China in the 2000s lol, and Japan's service-based gaming revenue relies heavily on the many anime games on smartphones, which I'm afraid I've never even heard of in North America, but have quite a bit of notoriety in China.
By the way, Genshin, which has become popular in recent years, is actually a Chinese game grafted onto the Japanese animation style, in which both China and South Korea have largely embraced the style of Japanese animation and produced related games, which is a special phenomenon in East Asia, except that Genshin is really too hot globally.
OP
OP
Samsung Develops Industry’s Fastest 10.7Gbps LPDDR5X DRAM, Optimized for AI Applications
Industry-leading features come with 25% higher performance, 30% more capacity and 25% higher power efficiency The new LPDDR5X is the optimal solution for future on-device applications and is expected to expand adoption into PCs, accelerators, servers and automobiles
Aufhebung RPG
Boo
I have a classic case of a North American game producer discriminating against Japanese games, it was probably the producer of Fez (sorry forgot his name) in 2011 who mocked Japanese games as "it's suck", and Jonathan Blow expressed a similar viewpoint euphemistically, and I think the target of his attack included Nintendo rather than just extremely backward makers like falcom .This whole thing speaks volumes about how unpopular Japanese games have become with the mainstream market during the ps3/x360 era.
oldpuck
Bob-omb
- Pronouns
- he/they
I do kinda get it. Flash is vector art, hand keyed animation, and often flat colors because that's the core toolset and because it compressed well.
I dunno what Team Cherry used to generate the assets, and I don't know if they're rendered as vectors or just high resolution images with great anti-aliasing. But it's flat line art, gradient fills, and extremely clean lines. I think it is undeniably gorgeous, but it is a style that would have worked well in flash games.
But another thing that looks like that? X-Men '97. The hand drawn animation of the original show had pencil lines which are aggressively cleaned up by being rebuilt in Toon Boom Studio, and it has flat color fills which were done because they were cheap from underpaid Korean animators in the 90s, and are cheap (and unnaturally clean) when done digitally.
Vector art is a certain generation's pixel art, but that doesn't mean it's invalid, anymore than pixel art is. And just like "pixel art" doesn't mean "looks like Mario", just because you have vector art as your basis, doesn't mean you "look like Homestar Runner"
Lancelot
Like Like
Artstyles no longer mean a game will be more or less demanding (especially with UE5 being the defacto), they're just that... Artstyles. You can totally like one more than the other, but the underlying technology is likely just as complex if not more, you shouldn't advocate for stylized graphics for the sake of "shorter development" or such pro-dev narratives, it hasn't worked that way since PBR (and especially UE5) became a thing. I know many people have a preference for 2D/pixel/vector art stuff and how it looks, but it plays on a completely different league and isn't relevant to the reality of 3D rendering (and by extent modern game development) as a result... If you really like that stuff, just say you do but don't try to fit it into a "cheaper/faster/better" narrative, they're not comparable and doing so is throwing all context through the window.
Last edited:
BreakAtmo
Moblin
- Pronouns
- He/Him
LPDDR6 and Blackwell's release schedules aren't out either, but that didn't stop you there
Does this mean that Switch 1 BC for Switch 3 is out of the question? Or would they be able to get around this issue somehow?
ItWasMeantToBe19
Manakete
The Switch 3 will probably not release until like 2033 or 2034 and be based around neural rendering so it’s hard to make any predictions about BC, lol.
Darknut85
#TeamRemake
Either they could make a custom chip with 32 bit support, build in a 32 bit supporting chip soley for BC or create an translation layer.
Darknut85
#TeamRemake
Hidden content is only available for registered users. Sharing it outside of Famiboards is subject to moderation.
Last edited:
