I think the main reason they said it that way, is because core counts are much easier to explain to laymen than the intricacies of cpu design.
The main issue for those ports were the total available performance of the cpu.
Last edited:
-
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
The main issue for those ports were the total available performance of the cpu.
ertaboy356b
Resident Troll
Did Nvidia ever made a custom SOC for a customer? Just curious. They mostly did software stuff.
The first surface book. The ps3?
None of those were probably customized to the same level of Drake though.
OP
OP
I believe the PlayStation 3 used a GPU, not a SoC, from Nvidia.
ReddDreadtheLead
#TeamLate2025WithAPotentialForEarly2026
- Pronouns
- He/Him
Yes, less cache and they have the GDDR memory that has pretty high latency, and CPUs are more sensitive to latency which affects their performance to be a bit worse.
It’s just that the 6 core Ryzen is the closest you’ll get to the console PS5.
The PS5 does use a very heavy OS, and i remember something that happened early in the PS4 generation is that it used 6 cores and later on had a 7th core unlocked but only partly. I suspect that as time goes the 7th core will be made more available but at the moment games that are currently being developed are around the 6.5 core limitation.
Remember that the PS5 is also a VR device.
It having the full 7th core just helps it a bit. In this case that is.
So did the surface book, but you know what I mean.
The original Xbox also had something from Nvidia, can't remember if that was custom or not.
Also funny how Xbox/Playstation only tried Nvidia once and never again.
The advantages of getting everything from one company are just to advantageous to ignore in a console.
Nvidia doesn’t offer a product like that for x86. AMD does.
ertaboy356b
Resident Troll
Nvidia seems to just offer an existing product and tweak it slightly. No wonder everyone steered clear from them.
They don’t have something like an apu/soc for the desktop space. They would have to get the cpu and gpu from two separate companies.
maybe they'll do something crazy and put Grace Hopper into a desktop chassis
Either way, I don’t see Ms or Sony giving up their AMD relationship, unless they get an exceptionally fucking good deal elsewhere.
Concernt
Optimism is non-negotiable
- Pronouns
- She/Her
As the Nintendo Switch family of systems marches towards the future, I fully expect it to expand beyond hybrids and handhelds. I don't see them dropping the Nintendo Switch branding, perhaps ever, I could even see their first AR-VR dedicated devices being Switch devices replacing handheld mode with VR mode.
With all that said, while I think it's utterly unrealistic RIGHT NOW, the possibility of an Nvidia powered Nintendo home console that plays Switch games with absurd graphics isn't impossible in the future. A Nintendo Switch Cube- a GameCube sized console for playing Nintendo games at 8K, 16K, whatever it is.
That is the only scenario where I see Grace Hopper entering the consumer video game console business. I doubt Nintendo wants them working with anyone else, and I doubt Nvidia wants to risk their partnership with Nintendo to do that even if they aren't contractually obliged to refuse. Plus the other big players in the industry are on AMD and moving to NVIDIA would be a huge development hurdle. Nvidia would have to push hard and offer huge savings and support to get anyone on board- which they have no incentive to do when Nintendo is so successful for them.
Concernt
Optimism is non-negotiable
- Pronouns
- She/Her
PS6, powered by Snapdragon 8XC Gen 10
just so you know, Grace Hopper is an absurd chip designed for HPC. I don't think there's a way to scale it to fit in a tablet
ReddDreadtheLead
#TeamLate2025WithAPotentialForEarly2026
- Pronouns
- He/Him
It uses ARM v9 so that makes it irrelevant.
Thraktor
"[✄]. [✄]. [✄]. [✄]." -Microsoft
- Pronouns
- He/Him
It's still fully compatible with ARMv8 (at least for 64 bit code, I believe the Neoverse core they're using has dropped 32 bit compatibility).
The one place where Grace Hopper (or something like it) could come into play is if Nintendo starts a streaming service. Hopper wouldn't work here, as it's missing some of the graphics-specific hardware, but Grace+AD102 could probably virtualise 15-20 Switch games running at Drake-level performance.
Concernt
Optimism is non-negotiable
- Pronouns
- She/Her
Correct, I never claimed they would or could.
if it was exclusive to Drake, it wouldn't have been rated for switch
ReddDreadtheLead
#TeamLate2025WithAPotentialForEarly2026
- Pronouns
- He/Him
Let’s ask AI about the switch 2:
Depends on the positioning of Drake.
guess it depends on the ratings group. ESRB separate exclusive platforms
for example
Xenoblade Chronicles 3D - ESRB
This is an adventure role-playing game in which players lead characters on a mission to save their worlds from destruction. As players explore fantasy environments and interact with characters, they use swords and magical projectiles to defeat various creatures (e.
www.esrb.org
ReddDreadtheLead
#TeamLate2025WithAPotentialForEarly2026
- Pronouns
- He/Him
I meant with respect to the scalability, ARM V9 seems to aim to be more of a scalable architecture that can go both ways, something that was difficult with ARM V8. Scaled up to DC but couldn’t scale down.
ARM v9 aims to be scaleable in both ways. Well, just to do it better than the ARM v8 CPUs. Maybe not perfect, but just perform it in a better/easier way. Though, there shouldn’t be much of a difference between the DC cores and say, a mobile SoC core…
D
Deleted member 887
Guest
And here is that CPU compared to Orin at comparable clocks.
https://browser.geekbench.com/v5/cpu/compare/18647313?baseline=9535279
Orin core perf (only really comparable since cluster sizes are not the same) is 2.6x Jaguar. The CPU in the Pro consoles was clocked at 2.3 GHz. Roughly comparable. Unless I am doing something deeply stupid?
Neither, I was bouncing back between comparing the Series S and the PS4 and swapped numbers.
PS4 is 18CUs @811Mhz. Drake is 12SMs @???Mhz. Raster perf scales linearly in both arches with clock and CU/SM number, and Ampere vs GCN4 is 1.8 more efficient controlling for clock and CU/SM. At the same clocks, Drake is 20% more powerful than PS4 in raster perf.
The short version is "I considered Infinity Cache in my analysis, and that is one of the reasons the 135% number is a floor."
The longer version is I have reason to believe that Series X and PS5 are not memory bandwidth constrained. RDNA2 GPUs have unusually low memory bandwidth, and Infinity Cache is their solution. Ampere GPUs just have a big fat bus.
Series X and PS5... also just have a big fat bus. PS5 has a 256-bit bus, with 448 GB/s peak bandwidth. The RX 6700 has the same clocks and number of CUs, but has 130-bit bus, with a 320 GB/s peak. Additionally, while PS and Series X don't have Infinity Cache, they use a shared L3 cache between the CPU and the GPU. Where a desktop card would need to decompress a texture to RAM, copy to VRAM (over the bus), then read from VRAM to render, the consoles can decompress, skip the copy (because it's already in VRAM), and often skip the read because the CPU has primed the L3 cache for the GPU.
@Look over there has determined that Drake likely is a "well fed" system from a memory bandwidth perspective, that the bandwidth available tracks with similarly sized Ampere GPUs. Looking at PS5/Series X, by comparison to their GPU counterparts, they similarly well fed.
Ampere GPUs have fairly consistent bandwidth/SM ratios. RDNA 2 has consistent in the high end and consistent in the low end, but the two don't match at all. So I decided to use a card in the higher RDNA2 band, and compare it to a comparably specced Ampere card. The RX 6900XT is in that upper band, and the 3080ti has the same number of compute units, and clock/memorybandwidth/VRAM ratios are in line with the rest of the 3000 series and the general ratio between Ampere and RDNA2.
Their performance ratios are very close and very consistent, but RDNA2 is clocked significantly faster. This is not surprising, Ampere SMs are packing more "cores" than RDNA2 CUs, the extra clock speed helps them turn over faster. Internal comparisons show that performance does tend to scale linearly with number of compute cores in both designs, unsurprisingly, as well as clocks, though there are ceiling effects where high clocks don't help much.
TL;DR at the scales we are talking about, we really can trade clock speed for compute units, and vice versa. Ampere's arch is generally better optimized for pure performance, RDNA2 for a cost/perf ratio that really suits a mid-to-high range card, but absolutely falls apart at the bottom - and stops being cost effective at the top. 135% perf improvement, measured either as clock speed or at CU/SM count is a data based floor for Drake's performance relative to the other current gen consoles.
D
Deleted member 887
Guest
Addendum: Series S is a mess. I see what Microsoft was trying to do, but the memory situation over there is bad. While the Series S GPU is going to be superior in raster perf to Drake's, Series S is not a well balanced console on this front. When I say "Drake can't meet the Series S" I'm trying to narrowly speak to the Series S GPU power. The Series S as a console is a whole 'nuther thing.
Series S has a memory bandwidth to CU ratio that is lower than any GPU AMD ships, unlike its bigger brother, which has bandwidth on the high end of what AMD ships. It's also heavily memory constrained, with what seems like a reasonable memory cut down, but with very similar OS overhead eating into that number. On the other hand, it's CPU is basically the same as Series X (again, probably for porting simplicity). The creates situations where Xbox One X and Series S have similar framerates, for opposite reasons. One X is CPU constrained, and needs half frame rate to compute AI/collision logic/etc, meanwhile the Series S CPU can churn through all that in 60fps time, and the GPU could keep up, but memory thrashing slows the whole thing down.
Drake is outperformed by Series S on the GPU and the CPU front, but the extremely limited RT power of Series S and its really unfortunate memory situation mean that it's tricky for games to take full advantage of what Series S can do. But CPU constrained and GPU constrained games on Series S won't have an easier time of it on Drake.
Series S has a memory bandwidth to CU ratio that is lower than any GPU AMD ships, unlike its bigger brother, which has bandwidth on the high end of what AMD ships. It's also heavily memory constrained, with what seems like a reasonable memory cut down, but with very similar OS overhead eating into that number. On the other hand, it's CPU is basically the same as Series X (again, probably for porting simplicity). The creates situations where Xbox One X and Series S have similar framerates, for opposite reasons. One X is CPU constrained, and needs half frame rate to compute AI/collision logic/etc, meanwhile the Series S CPU can churn through all that in 60fps time, and the GPU could keep up, but memory thrashing slows the whole thing down.
Drake is outperformed by Series S on the GPU and the CPU front, but the extremely limited RT power of Series S and its really unfortunate memory situation mean that it's tricky for games to take full advantage of what Series S can do. But CPU constrained and GPU constrained games on Series S won't have an easier time of it on Drake.
D
Deleted member 887
Guest
It's not fatter than a Windows install. It can actually leave the reserved threads unusually idle.
They have a normal, non-real time scheduler in their kernel, and then use reservations and processor affinity to make quasi-real time guarantees for the OS. This leaves the game in it's own sandbox. Devs get a desktop-ish scheduler that gives high throughput by default, and while it's entirely possible for thread A to starve out thread B on the same core, the devs control both thread A and thread B, so it's their problem under their control.
But if you want all your PS5 system services to be instantly responsive in game, you have to overprovision your OS reservation or be immaculate in designing your shell. Especially with Sony wanting to ship new features over the lifetime of the device, immaculate probably wasn't an option.
IIRC (please someone correct me if I don't) uses a strategy where a single core is reserved, and the UI lives there, but low priority OS tasks can get scheduled on the "game" cores
brainchild
Moblin
This is the way:
I'm serious. This is the future of NPCs.
That being said, there are a couple of implementation issues with this when it comes to real-time interactive AI in video games. For one, OpenAI is extremely picky when it comes to commercial licensing, so its utility for commercial use is limited right now. More importantly, it is not that easy to implement models like these in certain development environments that impose restrictions that are in conflict with the means through which OpenAI APIs function.
Also, it would be nice to have improved GPU acceleration to better assist with GPT local tuning of hyperparameters, TTS latency reduction, etc. so I hope we will see more promising advancements from GPU manufacturers in these areas, which would benefit future console hardware development as well.
Introducing ChatGPT
We’ve trained a model called ChatGPT which interacts in a conversational way. The dialogue format makes it possible for ChatGPT to answer followup questions, admit its mistakes, challenge incorrect premises, and reject inappropriate requests.
openai.com
I'm serious. This is the future of NPCs.
That being said, there are a couple of implementation issues with this when it comes to real-time interactive AI in video games. For one, OpenAI is extremely picky when it comes to commercial licensing, so its utility for commercial use is limited right now. More importantly, it is not that easy to implement models like these in certain development environments that impose restrictions that are in conflict with the means through which OpenAI APIs function.
Also, it would be nice to have improved GPU acceleration to better assist with GPT local tuning of hyperparameters, TTS latency reduction, etc. so I hope we will see more promising advancements from GPU manufacturers in these areas, which would benefit future console hardware development as well.
- Pronouns
- He/Him/His
Literally the best/worst thing Nintendo could ever possibly do is install these separate learning AI personalities into Animal Crossing villagers.This is the way:
Introducing ChatGPT
We’ve trained a model called ChatGPT which interacts in a conversational way. The dialogue format makes it possible for ChatGPT to answer followup questions, admit its mistakes, challenge incorrect premises, and reject inappropriate requests.openai.com
I'm serious. This is the future of NPCs.
That being said, there are a couple of implementation issues with this when it comes to real-time interactive AI in video games. For one, OpenAI is extremely picky when it comes to commercial licensing, so its utility for commercial use is limited right now. More importantly, it is not that easy to implement models like these in certain development environments that impose restrictions that are in conflict with the means through which OpenAI APIs function.
Also, it would be nice to have improved GPU acceleration to better assist with GPT local tuning of hyperparameters, TTS latency reduction, etc. so I hope we will see more promising advancements from GPU manufacturers in these areas, which would benefit future console hardware development as well.
Could you imagine how devasted they would be if you just up and disappear for two years? I don't think I could handle that level of responsibility
D
Deleted member 887
Guest
It's pretty interesting. Thinking about most games, free text interaction is rare, but a system where minor NPCs don't have to be written by hand, and even major ones can modify responses to follow the flow of conversation is a tasty concept.This is the way:
Introducing ChatGPT
We’ve trained a model called ChatGPT which interacts in a conversational way. The dialogue format makes it possible for ChatGPT to answer followup questions, admit its mistakes, challenge incorrect premises, and reject inappropriate requests.openai.com
I'm serious. This is the future of NPCs.
What sort of restrictions are you referring to? Just access to the remote OpenAI endpoionts?
- Pronouns
- He/Him
Yeah there is a game out there called AI Dungeon which is a text adventure where an AI responds to your prompts. It works... decently. A lot of the time the flow of the game makes it feel much more like a fever dream than an actual text adventure game but advances in AI text response could really make something like that a lot more palatable.
KeijiKG
Moblin
As someone who criminally dabbles in Stable-Diffusion, I can say that AI is far from infallible.Let’s ask AI about the switch 2:
brainchild
Moblin
Transformer-based Animal Crossing NPCs (fine-tuned to Nintendo's liking) would be incredible!
Yes, and that's the primary focus of my project, which is one of the reasons it has taken so long to make significant progress outside of the development of the game world itself. It is not yet commercially viable, though it is already working phenomenally. Still, there's too much of a delay between when you audibly ask the NPCs questions and when they audibly answer.
Can't really dig into the details of that without breaking NDA, so my statement was intentionally vague, sorry.
- Pronouns
- he / him
As a counterpart to the split comparison of 720p games scaled up to 1080p, I made some images with 1080p originals split with 720p downscales. Of course, unless viewed on a 7" 4K screen it's not going to help settle the matter of how much detail most people can appreciate at that size.
- Pronouns
- He/Him
It'll be about his supposed hatred for Mario Sunshine.
If you want to get pedantic, the OS is always running on all of the cores*, it's mostly just a matter of what work is scheduled where. The OS kernel is always what runs on the bare metal, which presents an abstraction of the system that all of the game and system processes run in. Even a full Game Core™ is going to end up doing some non-zero amount of OS work, albeit probably primarily in response to calls from the game itself. The OS reserved cores are not necessarily for making the OS go (though it's definitely possible some core tasks have been shifted to them), but really more for taking care of any background stuff that needs to be able to happen without directly interfering with the game.
That said, I'm not really familiar with the specifics of what Sony has done on PS5 beyond it probably still being a BSD.
Modern Xboxes actually run like 3 operating systems at once, with two of them being in VMs. It is probable some of the resource partitioning happens at the VM level.
Its about the subway sanwitches
So it's gonna be about pokemon?
when a game needs to use something, that something is loaded into memory. when it's not being used anymore, it's removed from memory. memory leak is when it doesn't get removed and the problem persists for other things that are requested, leaving less and less memory.
it could be a problem for pokemon, but it probably wouldn't explain all the problems for pokemon.
Raccoon
Fox Brigade
- Pronouns
- He/Him
I'm kind of stupid but I have a bachelor's degree in computer science, I know what a memory leak is
kvetcha
hoopy frood
- Pronouns
- he/him
I’m an English major, help
YolkFolk
Tingle
Maybe the false narrative that Pokemon looks how it does because of the Switch hardware.
D
Deleted member 887
Guest
"System" tasks is probably a more reasonable term. But userland-vs-kernel gets semantic, fast.
I'm actually less familiar with the Xbox strategy - what are the other two OSes, and does it use a hypervisor model underneath?
Hosermess
Koopa
They went through all that two weeks ago. There won't be anything Switch Pro related (apart from the repetitive questions at the end).
I wouldn't be surprised if he's talking about the drama surrounding Smash World Tour, Panda Gaming and Nintendo.
ReddDreadtheLead
#TeamLate2025WithAPotentialForEarly2026
- Pronouns
- He/Him
*2.84x
If Normalized to the same frequency.
I’m not sure if that’s an apt comparison, the 28SM 3060 outperforms the Vega 64 with 64 CUs by 20% as per TechPowerUP. While also having less memory bandwidth. And roughy the same TFLOP count.Neither, I was bouncing back between comparing the Series S and the PS4 and swapped numbers.
PS4 is 18CUs @811Mhz. Drake is 12SMs @???Mhz. Raster perf scales linearly in both arches with clock and CU/SM number, and Ampere vs GCN4 is 1.8 more efficient controlling for clock and CU/SM. At the same clocks, Drake is 20% more powerful than PS4 in raster perf.
AMD Radeon RX Vega 64 Specs
AMD Vega 10, 1546 MHz, 4096 Cores, 256 TMUs, 64 ROPs, 8192 MB HBM2, 945 MHz, 2048 bit
NVIDIA GeForce RTX 3060 12 GB Specs
NVIDIA GA106, 1777 MHz, 3584 Cores, 112 TMUs, 48 ROPs, 12288 MB GDDR6, 1875 MHz, 192 bit
If we were to normalize it, CU to SM, 2.28:1 ratio just using that kind of metric.
Add the 20% when at equal TFLOPs, and you get that an SM performs like 2.736 CUs of GCN5.0 who uses rapid packed math.
Or let’s compare the RX Vega 56 to the 3060 mobile.
56CUs vs 30SMs, both around 10TF (latter is a bit higher), the 3060 mobile outperforms that by 11%
So the performance would be 2.07CU:1SM in this case. Without accounting for the TF, 1.866CU:1SM to have roughly similar performance.
AMD Radeon RX Vega 56 Specs
AMD Vega 10, 1471 MHz, 3584 Cores, 224 TMUs, 64 ROPs, 8192 MB HBM2, 800 MHz, 2048 bit
NVIDIA GeForce RTX 3060 Mobile Specs
NVIDIA GA106, 1425 MHz, 3840 Cores, 120 TMUs, 48 ROPs, 6144 MB GDDR6, 1750 MHz, 192 bit
So, if we were to take it by strictly using the CU to SM method,
Using the latter example, to have the same performance, it would be 24.4% more performant/efficient than the PS4. If both were the same 1.843TF, then accounting it the Drake would be 38% more performant.
If using the former example, then 52% more performant/efficient than the PS4 while having less. And if they are the same TF, then 82.4% more performant/efficient than the the PS4… which I think might be unrealistic?
Unless my math doesn’t check out here….
Let’s use the GCN3.0 Radeon R9 Fury, a 56 CU graphics card that has the same shader count as the 3060 the 28SM card, the latter outperforms it by 1.78x… but also has ~1.78x the TF or close enough.
7.168TF vs 12.74TF
AMD Radeon R9 FURY Specs
AMD Fiji, 1000 MHz, 3584 Cores, 224 TMUs, 64 ROPs, 4096 MB HBM, 500 MHz, 4096 bit
So I’m not really sure using CUs and SMs is really a good idea in this case, or Drake would be >20% the PS4.
They do not. AMD APUs don’t do that.
Intel, Apple Silicon, Snapdragons (via SLC acts as L4$ to CPU) and nVidia SOCs do that. Also Mediatek via the SLC (acts as a L4 cache to the CPU)
And AMD does that by design. The L3 is exclusive for the CPU. GPU gets 0 access to that. It goes from the L2 to the VRAM and that’s about it.
None of their APUs do that “GPU access to the CPU L3”. They go straight to the RAM.
That is only if Drake doesn’t exceed 1.043GHz. After that it has worse memory bandwidth allocation per TF than the desktop Ampere.
Basically, why I said before about the ceiling for Drake with respect to TF is 3.2, after that it doesn’t scale in the favor and you get a more imbalanced system like the Tegra X1.
And also why I think that with the GPU clocking to half in portable mode, a bandwidth of 68.2GB/s is “preferred” to balance the scaling from portable to Dock as best as possible. If they used LPDDR5X it can make it easier, being 2x the bandwidth scaling with 2x the FP32 increase.
And yes it does account for the CPU.
I don't think they've talked about this a whole lot since the initial Xbox One reveal, so I might get some details wrong, but basically there's a manager OS that then launches (I think basically Hyper-V) VMs containing a "Game" OS (which is what Xbox games run on) and an "App" OS (which runs the Windows Store apps). I'm not entirely sure where the system UI itself lives, but it's probably either on the host or the App OS.
Last edited:
