• 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| (New Staff Post, Please read)

But what makes a dev determine how a game should be CPU or GPU intensive?
You’ve gotten some good answers, already. It’s determined by the workload the game executes.

But a dev’s goal is to use 100% of both, no more, no less. Failing to use all the resources available is performance left on the table. Overusing resources means bad performance.

Being able to scale down and up makes porting easier and games future proof. But this is really hard. Inevitably, the nature of a game will mean that it consumes one resource more than another. That’s what makes a game “CPU limited” or “GPU limited”.

When you’re CPU limited, you might have lots of spare GPU power but the GPU can’t do anything because it’s waiting on the CPU to finish. When that happens, you can’t lower resolution or visual settings to increase performance.

The reverse can also be true. Games generally don’t scale on CPU as well as GPU - a Bokoblin has the same AI no matter what resolution the game is running. But that’s why some games (like Spider-man) have settings for crowds and traffic density.
 
0
A doubt that came to me a while ago. We are talking about a system limited by power draw, that is, the theoretical peak performance of both the CPU and GPU would be far beyond what we will see being used in practice. Would it be feasible for Nintendo to offer more clock profiles combining CPU and GPU variations?
Example: A certain game would perhaps benefit much more from having the CPU running at 2GHz than the GPU at 1.1GHz. Another game might run even better if it sacrificed the CPU at 1GHz for a GPU boost to 1.3GHz. And some games would perhaps have their best balance in profiles between these 2 extremes.
Is there anything stopping Nintendo from doing this? Or is it just not worth the work?
 
I can understand that entirely but the same development house that made TW3 also made Cyberpunk 2077 and the cost doesn't reflect when comparing the two games... For Nintendo's second tier games even 40-50 million dollars would probably be an increase in development budget as I don't think most of those games cost close to that to make.
$40/50 milion budget isnt a bit expensive for the smaller scale games Nintendo make?$50 milions for a WarioWare WOW
 
So for Alan Wake 1, according to the communication director, Nintendo wanted that game and Epic also pushed for it to be on the platform. He did mention that it was a learning experience on how to approach the hardware. I can imagine lol, Control was a cloud version so I dont think they would have needed a dev kit for that. AW did get better with patches. So I can see them doing a port, specially with Epic and Nvidia wanting to push their tech.


That would be a huge win if AW2 can get a port to Switch, but I also hope the game is good enough to justify a double dip lol
 
I’ve thought something similar. I don’t think it’s impossible. Switch already has a special high CPU/low GPU mode for asset loading, so there is precedent
I wonder what an actual game on such a mode would be. Maybe a Paradox-style grand strategy game where graphics are just a 2D map but you have a bunch of interconnected systems between countries working in tandem?
 
AW2 getting ported would be insane. Wonder if they do Control as well.

Also re: GTA on Nintendo, clearly the solution is have to Mario beat up hookers and shoot cops in the next 3D Mario
 
people debating about the amount of ram again
So, right now, all the things that are been debating here again and again are:
  • If it will have a good 16 GB or RAM that will make Ray trace easier or 12 GB that will need to make magic to true use it often
  • If it will be on 4NM node that makes sense or a big fat power hungry 8NM node because Nintendo
  • If the controller will have color buttons like the good hardware's sequels or Wii U like buttons
  • If we will be playing in the first good half of the year or in the buzzy and super distant second half
  • If it will have a Ps4 power to be the weaker 2024's release or a health power (3.4+ Tflops) that make sense for that date
  • Same power consummation than Switch or power hungry like Deck
  • A hardware with good DLSS and Ray trace that will be used or functions only to show (too weak to use)
I missing something?
 
#Team16gb
I am now 11 pages behind. I really need to get back into the habit of checking on this thread more. That might take a bit with some looming irl stuff, but I'll manage. 😤
There will be pages you wish you could skip but you'll be rewarded if you continue to push through.
 
DLSS isn't really a power multiplier, it's better thought of as a very fancy upscaler.

Yeah - DLSS doesn't really improve performance, it improves image quality for a relatively small cost compared to the jump you see. Lowering resolution improves performance (surprise!).
 
0
Micron Collaborates with Qualcomm to Accelerate Generative AI at the Edge for Flagship Smartphones

Delivering world's highest mobile performance of 9.6 Gbps, Micron LPDDR5X now sampling for Snapdragon 8 Gen 3​

BOISE, Idaho, Oct. 24, 2023 (GLOBE NEWSWIRE) -- Micron Technology, Inc. (Nasdaq: MU), announced today that it is now shipping production samples of its low-power double data rate 5X (LPDDR5X) memory — the industry's only 1β (1-beta) mobile-optimized memory — for use with Qualcomm Technologies, Inc.'s latest flagship mobile platform, Snapdragon® 8 Gen 3. Running at the world's fastest speed grade of 9.6 gigabits per second (Gbps), Micron LPDDR5X provides the mobile ecosystem with the fast performance needed to unlock generative artificial intelligence (AI) at the edge. Enabled by its innovative, industry-leading 1β process node technology, Micron LPDDR5X also delivers advanced power-saving capabilities for mobile users.

"Generative AI is poised to unleash unprecedented productivity, ease of use, and personalization for smartphone users by delivering the power of large language models to flagship mobile phones," said Mark Montierth, corporate vice president and general manager of Micron's Mobile Business Unit. "Micron's 1β LPDDR5X combined with Qualcomm Technologies' AI-optimized Snapdragon 8 Gen 3 Mobile Platform empowers smartphone manufacturers with the next-generation performance and power efficiency essential to enabling revolutionary AI technology at the edge."

As the industry's fastest mobile memory offered in speed grades up to 9.6 Gbps, Micron's LPDDR5X provides over 12% higher peak bandwidth1 compared to the previous generation — critical for enabling AI at the edge. The Snapdragon 8 Gen 3 allows powerful generative AI models to run locally on flagship smartphones, unlocking a new generation of AI-based applications and capabilities. Enabling on-device AI additionally improves network efficiency and reduces the energy requirements and expense of more costly cloud-based solutions, which require back-and-forth data transfer to and from remote servers.

"To date, powerful generative AI has mostly been executed in the cloud, but our new Snapdragon 8 Gen 3 brings revolutionary generative AI use cases to users’ fingertips by enabling large language models and large vision models to run on the device," said Ziad Asghar, senior vice president of product management at Qualcomm Technologies, Inc. "Our collaboration with Micron to pair the industry’s fastest mobile memory, its 1β LPDDR5X, with our latest Snapdragon mobile platform opens up a new world of on-device, ultra-personalized AI experiences for smartphone users."

Built on Micron's industry-leading 1β process node and delivering the industry's most advanced power-saving capabilities such as enhanced dynamic voltage and frequency scaling core techniques, LPDDR5X offers a nearly 30% power improvement2 and the flexibility to deliver workload-customized power and performance. These power savings are especially crucial for energy-intensive, AI-fueled applications, enabling users to reap the benefits of generative AI with prolonged battery life.

Offered in capacities up to 16 gigabytes and providing the industry’s highest performance and lowest power consumption, Micron's LPDDR5X delivers unprecedented support for on-device AI, accelerating generative AI’s capabilities at the edge.
1 Compared to 8.533 Gbps for previous generation LPDDR5X

2 Measured against competitors' 1-alpha-based LPDDR5X
 
But to step up the GPU by 3W, gaining little more than half a TF of GPU performance, and gaining NOTHING else

Wait, wouldn't the RT cores and tensor cores also get the higher clocks? My understanding is that the whole GPU gets faster at once, so you would also get better RT acceleration and faster DLSS upscaling (although that may not matter if DLSS Concurrency is used).
 
0
I suppose AW2 if ever ported to Nintendo would mean they'd rather backport the Nvidia rendering paths to Drake rather than the modified PS5/Xbox Series ones... They might be superior lightning-wise and better suited to Nvidia hardware, would be very interesting to see imo.
Yes, that's what I mean. I suppose the core of the work will be adapted from the same work done on the Series S version due to similar performance of both machines. But on Switch 2, as you said, they can enable Nvidia features and rendering paths to achieve a very custom product. I really wonder how a Switch 2 version would look and run. Hopefully they'll port it to the machine.
Wait, what? This isn't LPDDR5T? Did Micron manufactured a officially validated overclock of LPDDR5X? Because I don't remember 9600 MT/s being JEDEC specs at all. Really crazy stuff.
 
It does seem we've hit the logical end-point for feasible graphics in video games with the 9th generation as a whole. With a good enough team that optimises the hell out of their game, we can have photo-realistic graphics that look insane in motion... but is unreasonable to expect, make and replicate for every game going forward. The crunch and brutal work hours are just not worth it. Having PS4 graphics is a seriously good thing to have, and Nintendo makes games on lower hardware that puts modern games to shame in a lot of ways. That Galaxy comparison is insanely on point.

That "mid-generation" jump point that @Machrider made also has a lot of weight. We're not increasing in capabilities as fast as we did 40 years ago, we've slowed down a ton. Most of the big advancements recently are to do with optimisation (DLSS, FSR, Frame Generation) and streamlining specific features (Ray-tracing with lighting). Graphical features aren't jumping as much, but that's not a bad thing. If Nintendo uses the Switch 2 to experiment with what other companies could've done with 8th (or arguably 9th) generation hardware, I'll be more than happy. Tears of the Kingdom was made on hardware with a similar power-level to the 2005 Xbox 360. Imagine what they can do with hardware far more powerful?

That thought alone makes me happy with the generational jump with the Switch. I want Nintendo to show how to make a damn fine, experimental and fun game.

For the next generation, I would pretty much be ok with visuals staying at the TLOU2/HFW level, but with a switch to total path-tracing in games. We'd get a significant visual jump but development would actually become easier.
 
If it's running 32fps on top pcs it's going to be like 8fps on switch 2 lol. Unless they do some black magic or something
Nothing would stop Remedy from running the PS5 version of Alan Wake 2 at 360p/30fps with Ray Reconstruction in portable mode and upscaling it to 1080p. Hell, in docked they'd probably then go 480p/30fps with DLSS to 1440p/30fps and use a Spatial Upscaler afterwards like what Returnal does on PS5 (900p TAAU->1440p + Spatial -> 4K)




This is CyberPunk 2077 with RTX overdrive running on an RTX 3050 with the rest of the settings on high. It looks blurrier because of YouTube compression. This is also before Ray Reconstruction came out. What the developers of Alan Wake could do is reduce the majority of settings from the PS5 version to their lowest in portable mode so they can use Ray Reconstruction for non-path traced RT.

I think if RTX overdrive ran on a 3050 with DLSS Ultra Performance and most settings on high at around 38-40fps , surely in docked mode Remedy could put the settings on a low-medium mix with RT or even path tracing and upscale from 480p/30fps to 1440p.



I found a video that does use RAY RECONSTRUCTION on a 3050 (8GB) with Path Tracing and it indeed runs at a solid 30fps, even during a fight scene! And even better, this is running at 540p with EVERYTHING ON ULTRA! (8:28 timestamp)




Surely this means that the theoretical Switch 2 could run RT at the very least or Path Tracing in docked mode with lower settings while upscaled from 360p->1080p or 480p->1440p right?


Edit: My apologies, I mean running a downgraded PS5 version and then adding Ray Reconstruction for the Next Gen Switch version
 
Last edited:
Nothing would stop Remedy from running the PS5 version of Alan Wake 2 at 360p/30fps with Ray Reconstruction in portable mode and upscaling it to 1080p. Hell, in docked they'd probably then go 480p/30fps with DLSS to 1440p/30fps and use a Spatial Upscaler afterwards like what Returnal does on PS5 (900p TAAU->1440p + Spatial -> 4K)




This is CyberPunk 2077 with RTX overdrive running on an RTX 3050 with the rest of the settings on high. It looks blurrier because of YouTube compression. This is also before Ray Reconstruction came out. What the developers of Alan Wake could do is reduce the majority of settings from the PS5 version to their lowest in portable mode so they can use Ray Reconstruction for non-path traced RT.

I think if RTX overdrive ran on a 3050 with DLSS Ultra Performance and most settings on high at around 38-40fps , surely in docked mode Remedy could put the settings on a low-medium mix with RT or even path tracing and upscale from 480p/30fps to 1440p.



I found a video that does use RAY RECONSTRUCTION on a 3050 (8GB) with Path Tracing and it indeed runs at a solid 30fps, even during a fight scene! And even better, this is running at 540p with EVERYTHING ON ULTRA! (8:28 timestamp)




Surely this means that the theoretical Switch 2 could run RT at the very least or Path Tracing in docked mode with lower settings while upscaled from 360p->1080p or 480p->1440p right?

PS5 can’t run Ray Reconstruction though

Edit: im unsure if you are saying the PS5 version as on the Ps5, or a version that is equivalent to what the PS5 is doing running on T239
 
As someone who plays in handheld almost exclusively, I hope handheld mode can get raytracing regularly, even if in a diminished form compared to docked mode.
 
No, more like 10-10.5GB :p
10.5 GB is the absolute minimum if it's 12 GB hardware.

811 MB (current OS)
+ 250 MB (extreme handwaving about "extra features")
+ 475 MB optional video capture buffer (5x larger than current)
= 1536 MB/1.5 GB

Not to mention that games needing more memory can disable the video capture buffer. And it's possible Nintendo will allow the game to specify settings or choose presets for video length/quality in order to control the size of the buffer, since the game has to donate memory to it; it's not actually part of the OS allocations.
 
Maybe give devs the option to disable some features their games straight up don’t need for more memory.

Like in an RPG that makes no use of Voice Chat for instance, turning it off.
 
Surely this means that the theoretical Switch 2 could run RT at the very least or Path Tracing in docked mode with lower settings while upscaled from 360p->1080p or 480p->1440p right?
It doesn't mean that, unfortunately. Which isn't to say that it's impossible just that this video does say that.

This is running an AMD Ryzen 7 5800X3D, a core that benchmarks at 2.5x the speed of the likely Switch 2 CPU. It's got 16GB of system memory + 8GB of VRAM. There is no way of knowing from this alone how much these things affect performance themselves.

The 3050 is a 9 TFLOP Ampere card. Switch 2 is probably in the realm of 3 TFLOPS. So, ballpark, to achieve the same thing, you'd need to start at 1/3 of 540p in terms of pixel count, so yes, 360p, which, in theory you could throw Ultra performance mode at and get up to 1080p.

But DLSS is going to take 3 times longer as well, because it's got 1/3 the FLOPS. Ray Reconstruction will also be tripled. My ballpark math says that will take ~19ms just for the DLSS pass. So the question is - with the 60% slower CPU on the left hand, and the 19ms DLSS pass on the right hand, is there enough time left for even a 360p render with these settings? And, with that low an internal resolution, will the RT effects hold up.

There is a version of Cyberpunk that runs on Switch 2 - if for no other reason than the phenomenally bad PS4 version exist. There is probably a version that uses Ray Tracing as well. Is that version the best looking/running version possible on the hardware, or is there a more cut back version with a less dramatic upscale that plays better? It's hard to describe how much YouTube destroys visual detail, so comparison is hard, but I'm not sure I'd prefer the ultra-low-res-but-path-traced version over 1440p-but-regular-RT version, or a 1080p-no-RT-but-60fps version.
 
Guys if it has low ram we could just download more
Fklt5ld.jpg
 
We need to stop gaslighting ourselves to demand less from Nintendo. They can and should release a beastly hybrid console and all signs point that way.

Mobile tech can do much better. No doubt Nintendo developed games can look good even if Switch 2 is just two switches ducked taped together, but that's not an excuse for Nintendo to release such a device
I think it's totally reasonable to expect PS4-level visuals but better with DLSS and RT, and better CPU capabilities as the floor for this device.

Those are pretty healthy expectations to have. Anything extra is a massive bonus. I don't think this is letting Nintendo off easy.
 
It doesn't mean that, unfortunately. Which isn't to say that it's impossible just that this video does say that.

This is running an AMD Ryzen 7 5800X3D, a core that benchmarks at 2.5x the speed of the likely Switch 2 CPU. It's got 16GB of system memory + 8GB of VRAM. There is no way of knowing from this alone how much these things affect performance themselves.

The 3050 is a 9 TFLOP Ampere card. Switch 2 is probably in the realm of 3 TFLOPS. So, ballpark, to achieve the same thing, you'd need to start at 1/3 of 540p in terms of pixel count, so yes, 360p, which, in theory you could throw Ultra performance mode at and get up to 1080p.

But DLSS is going to take 3 times longer as well, because it's got 1/3 the FLOPS. Ray Reconstruction will also be tripled. My ballpark math says that will take ~19ms just for the DLSS pass. So the question is - with the 60% slower CPU on the left hand, and the 19ms DLSS pass on the right hand, is there enough time left for even a 360p render with these settings? And, with that low an internal resolution, will the RT effects hold up.

There is a version of Cyberpunk that runs on Switch 2 - if for no other reason than the phenomenally bad PS4 version exist. There is probably a version that uses Ray Tracing as well. Is that version the best looking/running version possible on the hardware, or is there a more cut back version with a less dramatic upscale that plays better? It's hard to describe how much YouTube destroys visual detail, so comparison is hard, but I'm not sure I'd prefer the ultra-low-res-but-path-traced version over 1440p-but-regular-RT version, or a 1080p-no-RT-but-60fps version.
I mean being fair the videos in my post were all running on Ultra settings, so there is still some hope VRAM wise.
 
0
That is more or less just a half assed solution to a problem that the fde will solve on Drake.
There may be some complications if the overall GPU clock affects the FDE speed, but the FDE doesn't remove the CPU portion of loading, just a pretty big chunk of it. The game still has to parse the data, run procedural generation and initialization, do scene management, etc.
I have to imagine the video world be encoded as it was being made, right? I mean, there's no way the RAM is holding uncompressed HD video.
Pretty sure it's some sort of rolling buffer that holds the NVENC output. Not familar with the exact technical details of how that would work, but there definitely isn't space for anything significantly less compressed.
 
There may be some complications if the overall GPU clock affects the FDE speed, but the FDE doesn't remove the CPU portion of loading, just a pretty big chunk of it. The game still has to parse the data, run procedural generation and initialization, do scene management, etc.

So you think the loading mode with the upclocked CPU might make a return
Pretty sure it's some sort of rolling buffer that holds the NVENC output. Not familar with the exact technical details of how that would work, but there definitely isn't space for anything significantly less compressed.

That's why I was thinking/hoping that AV1 encoding hardware could be used. It could cut the required space in half, or maybe allow for more video to be held if that's what they wanted.

It would be pretty cool if devs could allow for a variable amount of space to be used for holding video. Like maybe if a game genuinely didn't need anything more than 6GB, for example.
 
So you think the loading mode with the upclocked CPU might make a return?
Imo no. As Pokemaniac said, the FDE will remove a pretty big chunk of the cpu requirements for loading, and an octacore A78 is very capable on its own.

Edit: Downclocked gpu for loading migh make a return, purely for battery life reasons. But not sure about how much power that would actually save, if its worth it.
 
Last edited:
For the next generation, I would pretty much be ok with visuals staying at the TLOU2/HFW level, but with a switch to total path-tracing in games. We'd get a significant visual jump but development would actually become easier.
Doesn't even have to be TLoU2 levels tbh. Games like Hi-Fi Rush and most Nintendo games look better than LoU2 to me because of the intense stylisation of the artstyle.
If we're talking photorealism though, I do agree. If we stopped here in terms of texture quality, I would be absolutely fine.
 
Nintendo making a “move” means reducing their cut or funding the port.

Take Two will argue that a high quality port to a small device will be expensive, will sell less than the PC version, and drice hardware sales for Nintendo, so they deserve a lot of money.

Nintendo will argue they’re going to sell 100 million of these things and if even 10% of people buy the game there, that’s half a billion dollars, and Take Two doesn’t need a penny.

Both companies know that if they give more than usual to the other, it will be used as leverage against them by other companies. Both can afford to walk away.

So yes, these conversations will happen. Whether or not there is a number in the middle that makes both happy is another question.
Would other companies really be able to use the results of this conversation as leverage against Nintendo/Take Two? Wouldn't this hypothetical contract be entirely confidential and not available to other parties? Not just the details but the existence of it even.

With that said, it's true that this conversation might set a precedent between the two companies. In 4 years TT may ask Nintendo for money again when it's time to launch RDR3 and use GTAVI as a precedent to expect that. But again, if GTAVI sells well, Nintendo will say, nope, you should bring RDR3 without our help. If GTAVI doesn't sell well, Nintendo won't care too much to lose RDR3, so there's no issue for Nintendo regardless.

Basically, I think whether this conversation happens or not almost entirely depends on whether Nintendo thinks GTA trilogy selling extremely well on Switch 1 should already be enough pull for TT or if they need to prove their audience yet again to TT on a new platform.
 
0
The numbers are not universally agreed upon and just broadly don't work super well. It's best to just be explicit about which systems you're referring to.
As opposed to using last for last gen and last for the gen before, numbers are definitely better.
 
For the next generation, I would pretty much be ok with visuals staying at the TLOU2/HFW level, but with a switch to total path-tracing in games. We'd get a significant visual jump but development would actually become easier.
are you aware for Switch sucessor to acomplish such/visuals, it would take a massive amount of resources/time and personal, it took years, massive resources, for Guerilha/Naught Dog acomplish such visuals on PS4, this such visuals is not you can do in few years, this is pushing PS4 to it absolutely limit
 
Please read this new, consolidated staff post before posting.

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


Back
Top Bottom