• Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.

When next gen hits (Xbox Next, PS6), where will be with node shrinks and feasible performance?

As always, every new console is tempered by two things, cost and heat.
To continue the march forward in gains, we rely on node shrinks in the main to keep heat and costs down.

In November 2027, where would we be with realistic node shrinks that would be cost efficient to use in a console, and how many more Compute Units would that node be able to house keeping it in the same neighbourhood as current console APU.

The thing I see is that node shrinks are slowing down, and the cost is going up, which is a big change from what the consoles have relied on historically.
However, I think there is the ability for the consoles to be more expensive and still not hurt sales. The PS5 going up $50 hasn't hurt it at all, and with inflation I think we could easily put $100 more on the cost of a console.
But again, the issue with just saying have a bigger chip, both companies really want to come in with relatively low power draws to look all green, which I think will be helped a little by lower power draws from the SSD, RAM and the disc drives being removed from the next generation of consoles.
However, it might not be enough.

Where will we be in 2027?
 
Last edited:

Daneel Elijah

Gold Member
Kinda weird that it feels like this gen hasn't even properly begun, but we're almost halfway through.

Each passing gen we're getting fewer and fewer games released.
It's getting less attractive to buy a new console.
But games are longer than they were before in return. I loved MGS2. 8 hour game, with half of it cutscenes. Sonic 2 and super mario ? Even less than that. But after the PS2 golden age it is true that we lost the AA games. This gen is really slow to begin, but we got good cross gen games in return. So if we do get great PS5/ Series X games I will not mind this cross gen period that much. But I have to admit that next gen the PS6 will need good exclusives to convince me to buy one.
The PS3 was awful in the first few years. So I waited for Bloodborne to buy the PS4. Still got 1 or 2 years of no games that interested me. The PS5 is better in that regard. But this was because we got Demon's Souls at launch. I would have stayed with my PS4 otherwise.

For next gen I can see Sony and Microsoft doing huge changes to their hardware. Microsoft will either continue their S/X duo, or focus on the X, or even focus on the S( by going early and promising a pro version just after the PS6 for example). Sony now know how much FOMO the Playstation can generate so we may see Mark Cerny go nuts for what will be probably his last chance to make the console of his dreams, as he is getting older and probably won't have the opportunity to have the trust of Sony and the money/ great prospects that the Playstation brand has right now. At least I want that to happen.
 
I think when it comes to the PS6 and the Xbox Whatever they will probably be in the same ball park when it comes to performance. Just like this gen with no massive differences between the two.

Microsoft might take a different approach when it comes to Xbox Cheapo. I guess they will do a better job and not manufacture as many cheapos if the demand for Xbox Whatever is a lot greater. Xbox Cheapo will be far behind the other two when it comes to performance.

What I do see happening with all systems is a simplification of how they are designed. I definitely see both Sony and Microsoft design them in a way where they are easier to manufacturer.

Edit:

Sorry for the Xbox Cheapo and Xbox Whatever. I have no clue what they will be called at this point. Only PlayStation seems obvious to me.
 
Last edited:
I think consoles come earliest in november 2028. I also think their chips are bit smaller than typical 350mm2 ish. Maybe 2nm from tsmc with that backside power thing could be affordable at 699 dollars.
So if we have a certain amount of transistor density for 7nm, what will 2nm gives us? Like, can we fit say three times the amount of transistors on a 2nm than a 7nm?
 
According to some sources, 7nm to 5nm gives 80% more density. N3 to 5nm is 70% greater density. The best guess for N3 to 2nm is around 25%.
So all up maybe 3.8 the amount of transistors from 7nm to 2nm?
 

PeteBull

Member
First expect midgen upgrades, soonest holidays 2024 or sometime in 2025.
With that u can expect trully next gen consoles soonest 2028 and more likely 2030, which is bit too far away to predict stuff, tons of delays can happen, hell we can even have china's invasion on taiwan/ww3(coz then US and their allied forces will jump in ).
 

Mr.Phoenix

Member
I believe the nextgen consoles show up around 2028 at the earliest. At this time, we will be on the much longer cadence and at the time mature, N2 node which is expected to begin volume production around H2 2025 or H1 2026.

This should allow for around a 4x jump in performance compared to what we have now (in the 9th gen consoles). And that is just from a logic perspective not taking into account any architectural advancements. And then you can also add clock and thermal benefits too. So overall we are ikey going to be getting performance in the 5x+ range.

This should put the next-gen consoles in at around 40-50TF of raw FP32TF range. In addition to probably coming with stuff like AI hardware, better RT hardware and more cache. I also suspect there would be some rather advanced and innovative architecture advancements between now and then, guess we can say we are already seeing these things happening now on the PC side of things. Eg, it's possible that that 40-50TF chip may present as an effective 80 - 100TF chip due to scheduling improvements and changes....etc.
 
Last edited:

Reallink

Member
I believe the nextgen consoles show up around 2028 at the earliest. At this time, we will be on the much longer cadence and at the time mature, N2 node which is expected to begin volume production around H2 2025 or H1 2026.

This should allow for around a 4x jump in performance compared to what we have now (in the 9th gen consoles). And that is just from a logic perspective not taking into account any architectural advancements. And then you can also add clock and thermal benefits too. So overall we are ikey going to be getting performance in the 5x+ range.

This should put the next-gen consoles in at around 40-50TF of raw FP32TF range. In addition to probably coming with stuff like AI hardware, better RT hardware and more cache. I also suspect there would be some rather advanced and innovative architecture advancements between now and then, guess we can say we are already seeing these things happening now on the PC side of things. Eg, it's possible that that 40-50TF chip may present as an effective 80 - 100TF chip due to scheduling improvements and changes....etc.

Despite pre-launch wish-crafting to the contrary, there's very little difference in RDNA (PS4) and RDNA2 (PS5) FLOPS in practice. PS5 is very much in line with where you'd expect an ~2.2x PS4 Pro to perform with a radically faster CPU. The saddest part is the half year old 4090 is already a genuine 40-50TF in the old raw single instruction per clock method (90-100+ in new 2 instruction/clock method) and you're talking about a 2028 best case scenario console possibly being in the same ballpark.
 
Last edited:

Schmendrick

Member
Despite pre-launch wish-crafting to the contrary, there's very little difference in RDNA (PS4) and RDNA2 (PS5) FLOPS in practice. PS5 is very much in line with where you'd expect an ~2.2x PS4 Pro to perform with a radically faster CPU. The saddest part is the half year old 4090 is already a genuine 40-50TF in the old raw single instruction per clock method (90-100+ in new 2 instruction/clock method) and you're talking about a 2028 best case scenario console possibly being in the same ballpark.
PS4 was not RDNA and we're just getting started with next gen developments. So far we've basically seen nothing but old engine architectures using old featuresets. Look at the matrix demo, what that could do with an unfinished lumen and Nanite system...
 
Last edited:

Mr.Phoenix

Member
Despite pre-launch wish-crafting to the contrary, there's very little difference in RDNA (PS4) and RDNA2 (PS5) FLOPS in practice. PS5 is very much in line with where you'd expect an ~2.2x PS4 Pro to perform with a radically faster CPU. The saddest part is the half year old 4090 is already a genuine 40-50TF in the old raw single instruction per clock method (90-100+ in new 2 instruction/clock method) and you're talking about a 2028 best case scenario console possibly being in the same ballpark.
Never said there was a difference in TF... TFs are TFs. And yes, PS5 is very much in line with a 2x+ PS4pro with a faster CPU, and then all the other architectural benefits, like SSD, More RAM, and faster bandwidth. There is no rocket science to it.

Some more context? This may surprise you.

The first 10TF+/12TF AMD GPU were the RX56 and 64 respectively... Which came a whole 3-4 years before the PS5/XSX consoles. Or Nvidia equivalents the GTX 10xx series.... which were all 2017 hardware which meant they were finished and developed around 2016.

But 10TF in 2020 is not the same thing as saying 10TF in 2017 is it? That's where architectural improvements and innovations come into play. And they do exist whether or not you choose to admit it or not. Be it simply having more l2 cache, better IPC, more bandwidth, RT cores, Hardware decompression, AI cores...whatever. There will be a lot of architecture improvements on top of that raw TF number, which as some of us keep saying, is just one part of the equation.

I have said before, if just looking at raw TF, whatever we see in the high-end GPUs 2-4 years before the next-gen starts, is the TF number we will be seeing in the consoles. What that theory doesn't account for, is that in addition to that, there will be a lot of architectural benefits we get too. Eg. In 2019 AMD didn't even have a GPU with RT cores.

You will be remiss, ignorant even, if you use just logic density and clock improvements, which is what I did to get that 40-50TF number and draw the line there. All that stuff, the new instruction set method, AI, infinity cache, mesh shading, innovations in RT...things that are here now and things we do not even know about yet but will be available then or be more mature.

That 40-50TF? That's the projected minimum, which is why I said based on just logic density alone. On N2, you can have 4-5x more transistors in the same space as you do in N7. At the same clock, you take performance from 10TF to over 40TF. But that's not taking into account that an N2 chip will not only draw less power but will be able to clock much higher than an N7 chip. Eg... PS4 GPU on N28 had a clock of ~0.8Ghz, on N7, its clocked at almost 3x faster.

Oh... just one last thing. Your way of looking at power with regard to consoles and PC is totally warped. Yes, I expect next-gen consoles to be around as powerful on the raw TF scale as a 4090. Maybe 15-20% more. Not accounting for other improvements between now and then. What you are missing though, is that those consoles, with all their 4090+ goodness, and new and better CPUs, would at that time, become the new baseline. Games will be made on those machines that run at 30fps/60 quality and performance mode respectively..... see where I am going with that?
 
Last edited:
I believe the nextgen consoles show up around 2028 at the earliest. At this time, we will be on the much longer cadence and at the time mature, N2 node which is expected to begin volume production around H2 2025 or H1 2026.

This should allow for around a 4x jump in performance compared to what we have now (in the 9th gen consoles). And that is just from a logic perspective not taking into account any architectural advancements. And then you can also add clock and thermal benefits too. So overall we are ikey going to be getting performance in the 5x+ range.

This should put the next-gen consoles in at around 40-50TF of raw FP32TF range. In addition to probably coming with stuff like AI hardware, better RT hardware and more cache. I also suspect there would be some rather advanced and innovative architecture advancements between now and then, guess we can say we are already seeing these things happening now on the PC side of things. Eg, it's possible that that 40-50TF chip may present as an effective 80 - 100TF chip due to scheduling improvements and changes....etc.
So my guess is that if there is around a 3.5 increase in transistor density when next gen releases, if we look at the XSX as a base, we could go from say 52CUs to 175 CUs. We would expect a speed increase to around 4ghz.
Basic maths says that's around 90tflops.

If we do a midgen doing 2x the OG, and then 2x again for the new gen then we are looking 48tflops.

MS this gen went to pains of pushing ways to gain efficiencies, such as VRS and Sampler Feedback Streaming. I think next gen for both Sony and MS, DLSS type upscaling has to be a big part of their consoles. Has to be. The only thing about that is that AMD seem hell bent on not going down that path.
 

Mr.Phoenix

Member
So my guess is that if there is around a 3.5 increase in transistor density when next gen releases, if we look at the XSX as a base, we could go from say 52CUs to 175 CUs. We would expect a speed increase to around 4ghz.
Basic maths says that's around 90tflops.

If we do a midgen doing 2x the OG, and then 2x again for the new gen then we are looking 48tflops.

MS this gen went to pains of pushing ways to gain efficiencies, such as VRS and Sampler Feedback Streaming. I think next gen for both Sony and MS, DLSS type upscaling has to be a big part of their consoles. Has to be. The only thing about that is that AMD seem hell bent on not going down that path.
Well yes.. kinda.

This is how I arrived at that 40-50TF Raw TF number.

I used the PS5 as the base, as I feel its a better representation of what the minimum w can get is. So 36CU. 4x density bump puts us at 144CU. If we use the same ~2.2Ghz PS5 clock, then we end up with 40.5TF.

As you can see, I was being extremely conservative.

What is likely is that clocks may be as high as at east 3Ghz... that puts us at the 50TF point. But then you consider some other things, like say, maybe the CUs aren't even quadrupled, maybe what changes is how many instructions they can run simultaneously. And maybe there is more of a shift and focus on RT and AI hardware.

There is just SOOOOOO much we do not or cannot possibly know yet. Outside some raw rudimentary TF calculations.

Oh... and AMD doesn't have to go down that path officially. All they have to do is add AI cores to their GPUs. Which they are already rumored to be doing with their next gen GPUs. At this point, they don't have a choice.
 
Last edited:

Matt_Fox

Member
As always, every new console is tempered by two things, cost and heat.

I'm not sure the heat issue (the unseen equation in Moore's Law) in consoles and high performance graphics cards is solvable even with the current ever increasing system of ridiculous fans and coolant pipes. We're fast approaching a point where if its a hot day, you simply have to resign yourself to keeping your PC or console turned off.

BcSpKBx.jpg


That's one of the reasons - not the only one - but one of the reasons why I see cloud gaming as an inevitability in the 2030s and beyond.
 

NahaNago

Member
Playstation 6, Xbox USS-Enterprise D1 and the rumored Switch 2 will probably hit somewhere around 2026-2027
Too early for ps6 and xbox alphabet. Most likely will be 2028 but I would love for you to be right. The switch 2 should hit sometime around 2024 to 2025.
 

Three

Member
The thing I see is that node shrinks are slowing down, and the cost is going up, which is a big change from what the consoles have relied on historically.
However, I think there is the ability for the consoles to be more expensive and still not hurt sales.

Where will we be in 2027?

With xbox selling and pushing the Series S, a console that is barely above the previous generations One X and with most games being crossgen I think the drive for performance has slowed down already.

I do agree with you that cost will more acceptably go up though. We are seeing people happy to pay $800-900 for a portable to play the same library of games when $600 for a high performance console would have been laughed at years ago. I would say that's happening already too where performance is paid for through a profit margin rather than them taking a loss.

I think devs have really shifted to GaaS/P on any device (including mobile) and performance consoles will just become more of a niche instead of a necessity to run the latest games.
 
Last edited:
Wherever it is it will still be at least one generation behind what's available on PC.
Begs another good question, if PC cards are hitting 100tflops now, what will the top end cards be pumping out when the next consoles come out?
 
Last edited:
With xbox selling and pushing the Series S, a console that is barely above the previous generations One X and with most games being crossgen I think the drive for performance has slowed down already.

I do agree with you that cost will more acceptably go up though. We are seeing people happy to pay $800-900 for a portable to play the same library of games when $600 for a high performance console would have been laughed at years ago. I would say that's happening already too where performance is paid for through a profit margin rather than them taking a loss.

I think devs have really shifted to GaaS/P on any device (including mobile) and performance consoles will just become more of a niche instead of a necessity to run the latest games.
I was paying 700-800 AUD for thr PS1 and
Saturn back in the 90s in Australia, which is the same price we are paying for the PS5 and XSX now. The inflation cost would be $1460.00 in today's money.
 

Mr.Phoenix

Member
I do agree with you that cost will more acceptably go up though. We are seeing people happy to pay $800-900 for a portable to play the same library of games when $600 for a high performance console would have been laughed at years ago. I would say that's happening already too where performance is paid for through a profit margin rather than them taking a loss.
Don't let the media or internet fool you. These $500+ Portable PCs... are significantly more niche than you can imagine. And if you want to move numbers like a console would.... yes, a $600 starting price for a console will Still be laughed at. like, do not use something that at best sells 2M units a year (high-performance handhelds) versus something that its own platform of products sells upwards of 35-40M (consoles) a year to make assumptions about what the 40M a year selling platforms future will be.

And no devs have not ready shifted to GaaS. Its a very popular genre and/or game mechanic, but like every gen, or every time when something new comes along, there will be a lot of publishers trying to be the next one to strike gold, and there would be a lot of gamers saying this or that is dead or the future. But its never the case.
 

rofif

Can’t Git Gud
Kinda weird that it feels like this gen hasn't even properly begun, but we're almost halfway through.

Each passing gen we're getting fewer and fewer games released.
It's getting less attractive to buy a new console.
Exactly m thoguhts. People been talking about Pro and ps6 since day 1. Its stupid.
The gen has barely started
 

PeteBull

Member
To ppl thinking no ps5pr0 is happening- think of it logically- will sony leave money on the table and present their hardcore/techsavy/graphics focused ppl with just base ps5 till 2030 w/o any premium option?

Sony isnt stupid and is very forward thinking, we are getting ps5pr0 4 sure- even early/mid 2023 game like Jedi Survivor performs terribly and has switch alike resolution even in fidelity mode- thats a clear sign for huge need for more powerfull mashine, sony isnt nintendo to leave their customers with very weak and outdated piece of hardware for long years .
 

Three

Member
Don't let the media or internet fool you. These $500+ Portable PCs... are significantly more niche than you can imagine. And if you want to move numbers like a console would.... yes, a $600 starting price for a console will Still be laughed at.
You're right, that's true. Remember though that I believe high performance consoles will become a niche. What you are saying is only important if you truly believe in generations.

like, do not use something that at best sells 2M units a year (high-performance handhelds) versus something that its own platform of products sells upwards of 35-40M (consoles) a year to make assumptions about what the 40M a year selling platforms future will be.
The thing is I don't think the high performance console will be the platform anymore. I think the ecosystem has become the platform and not the console. The high performance console I believe will transform to a for profit niche much like a portable is an extension of the Steam ecosystem or EGS, xbox/Gamepass, PSN/PS+, etc. I think the necessity to sell a loss leading high performance console to establish an install base to have games specifically support that platform alone is no more. So in the future I believe you can probably sell an expensive high performance console that plays the same games of an already established install base at a higher price/performance.
And no devs have not ready shifted to GaaS. Its a very popular genre and/or game mechanic, but like every gen, or every time when something new comes along, there will be a lot of publishers trying to be the next one to strike gold, and there would be a lot of gamers saying this or that is dead or the future. But its never the case.
I don't think it means other types of games are dead but the popularity of those type of games make the 'console as a platform' less viable and the game itself as the platform across all devices more attractive.
 
Last edited:
Next time round I’d like a high end SKU available at launch. Equivalent of the mid-gen upgrade. I’d pay up to £$800 on launch for such a device.
 

cireza

Member
To go forward in gains, we should spend the next 10 years actually trying to use current hardware correctly. You know, like when we managed to have games like Sonic running on Master System. Now this was using 100% of the capabilities, nothing wasted in an infinite stack of middleware and useless engines.

At some point, the only path forward will be optimizing again. Real optimization.
 

MarkMe2525

Member
Manufactures are at a crossroads as there are not many ways to lower cost while bringing more power. Nintendo is surprisingly well positioned in this new market due to their commitment to differentiating from their peers.
 
Last edited:

sono

Member
If you can't increase circuit density and cannot make larger chips due heat you go parallel
 

PeteBull

Member
Next time round I’d like a high end SKU available at launch. Equivalent of the mid-gen upgrade. I’d pay up to £$800 on launch for such a device.
Impossible, even base ps5 is max of console tdp , so 250W(theoretical/offical tdp of console is 350W and thats the psu it has, but actual peak usage is 220-250W, hell crossgen launchgame spiderman milesmorales could break over 200W easily already- measured by digital foundry ), at that time amd didnt even have gpu that would be 2x powerfull, that would fit into 250W tdp together with cpu, hell it doesnt even have it now, thats the whole reason we gotta wait for 3nm tmsc node and rdna4 gen gpu.

But price u are spot on, very likely ps5pr0 will be lowest 600/highest 800$ when it launches november 2024 or sometime in 2025(it might seem steep but its premium product, and even 599$ ps3 if u account for inflation is worth over 900$ now).
 

diffusionx

Gold Member
Whatever the cutting edge process is at the time, Apple will probably be buying up all the capacity, unless Samsung and Intel catch up to TSMC which they're both investing tens of billions to do.

just give me standard 60fps and performance mode 120fps.
this stuff is all relative man, these consoles can obviously do PS4 games at 4K/60fps but once you jack up the features then performance suffers. Same as it ever was and ever will be. You get more hardware, games will take advantage of it without a strict mandate by the console makers which they aren't likely to do. So the next round of consoles will surely be able to do PS5 games at 4K/60 but devs are going to want to use that hardware to do more.
 
Last edited:

Bigfroth

Member
Great tech, fast ram advance CPU/GPU, world class SSD storage will be standard. Also stuttering frames, pop in, Vaseline coated textures. Unreal 6 demos will be glorious to look at and dream. It will never change.
 

Wildebeest

Member
Possibly, real next gen will be so power hungry and expensive that game streaming services will be offered to the bulk of the playerbase. The long term trend is for mobile gaming to totally eclipse TV gaming anyway, so I suppose we see what form factor Nintendo choose and how they support TV gaming.
 

Knightime_X

Member
Pretty sure we had this problem last gen as well.
First 3 years were d
You think Sony and MS want to work with Nvidia to make a custom APU and who is gonna make the CPU portion of that APU?
Or come up with their own version.
Either way DLSS of some sort absolutely NEEDS to happen.

If nintendo works with Nvidia but not Sony or MS, it won't be long before Nintendo has the best looking and performing games thanks to the immense help of dlss.
This of course assumes dlss is even used.
 

truth411

Member
As always, every new console is tempered by two things, cost and heat.
To continue the march forward in gains, we rely on node shrinks in the main to keep heat and costs down.

In November 2027, where would we be with realistic node shrinks that would be cost efficient to use in a console, and how many more Compute Units would that node be able to house keeping it in the same neighbourhood as current console APU.

The thing I see is that node shrinks are slowing down, and the cost is going up, which is a big change from what the consoles have relied on historically.
However, I think there is the ability for the consoles to be more expensive and still not hurt sales. The PS5 going up $50 hasn't hurt it at all, and with inflation I think we could easily put $100 more on the cost of a console.
But again, the issue with just saying have a bigger chip, both companies really want to come in with relatively low power draws to look all green, which I think will be helped a little by lower power draws from the SSD, RAM and the disc drives being removed from the next generation of consoles.
However, it might not be enough.

Where will we be in 2027
Next Gen won't hit before holiday 2028, especially since the pandemic delayed software development a couple of years. It wouldn't surprise me if MS and Sony aim for holiday 2029 if mid gen consoles come out, thus:

TSMC 2nm (atleast)
Or
TSMC 1.4nm (whatever is after 2nm) in an ideal world.
 

Black_Stride

do not tempt fate do not contrain Wonder Woman's thighs do not do not
Or come up with their own version.
Either way DLSS of some sort absolutely NEEDS to happen.

If nintendo works with Nvidia but not Sony or MS, it won't be long before Nintendo has the best looking and performing games thanks to the immense help of dlss.
This of course assumes dlss is even used.
When you say DLSS3 do you mean Reflex and FrameGeneration/Interpolation?
Or the SuperResolution portion of the suite.

What about XeSS?
Or TSR?
Or a more advanced form of checkerboarding
Or FSR.
Or whatever clever techniques devs come up with for upscaling.

TSR already looks amazing going from ~1440/1800p up to full 2160p and it doesnt need any special hardware to run.

Cuz interpolation I 100% see AMD having their own solution sooner rather than later.

Reflex who knows....Microsoft supposedly already had their own version of Reflex.
But i dont know how true any of this Dynamic Low Latency Input shit actually is.....DLSS3s Reflex is legit though.

XboxSeriesXLatency_HERO.jpg
 

Knightime_X

Member
When you say DLSS3 do you mean Reflex and FrameGeneration/Interpolation?
Or the SuperResolution portion of the suite.

What about XeSS?
Or TSR?
Or a more advanced form of checkerboarding
Or FSR.
Or whatever clever techniques devs come up with for upscaling.

TSR already looks amazing going from ~1440/1800p up to full 2160p and it doesnt need any special hardware to run.

Cuz interpolation I 100% see AMD having their own solution sooner rather than later.

Reflex who knows....Microsoft supposedly already had their own version of Reflex.
But i dont know how true any of this Dynamic Low Latency Input shit actually is.....DLSS3s Reflex is legit though.

XboxSeriesXLatency_HERO.jpg

Deep learning super sampling

 
Ideally, 10th-gen consoles will utilize a mixture of advanced technologies that can help bring optimal performance alongside smaller case designs and better modularity for things like VR/AR. Node shrinks alone won't bring the required benefits; it'll be a mixture of things:

>HBM-PIM (Processing In Memory) or GDDR-PIM (this doesn't exist yet, but DRAM-PIM does and GDDR is based off DRAM so...)​
>Chiplet packages using standards like UCIe​
>Integrated multi-I/O data processing blocks (i.e taking PS5's I/O subsystem but with much greater expansion of functionality)​
>PNM (Processing Near Memory) elements (integrated into things like the I/O systems; again PS5's I/O subsystem has some of this but 10th gen can expand on it even more)​
>Faster SSDs (obviously) utilizing CXL 3.0 over PCIe​
>Targeted mixture of node processes for different GPU components (RDNA3 already does this; future archs should have better granularity for even more precise mixture​
of node sizes for various processor components)​
I can get a little more in-depth on what I feel many of these would enable for 10th-gen systems, but I think this also depends a lot on what Sony & Microsoft attempt for 10th-gen systems. And this is where I think they'll diverge a lot.

Sony: Some kind of console that integrates VR into the package by default. Controller can also function as a VR controller by detaching segments (for VR) and recombining them (for non-VR or VR with "normal" controls). Some AR functionality also built into the headset. System designed with multi-user VR gaming in mind (up to 4 people can play a game in the same space with VR headsets); VR/AR features also built into the system OS for QoL benefits.

Multiple VR headsets at Entry (default model included with all PS6s), Enthusiast, and Professional. Enthusiast & Professional headsets also have built-in hardware to enable cloud streaming of PS6 games via Remote Play and over PSN wherever Wifi service is available, and native play of PS4 (Enthusiast & Professional) & PS5 (Professional) games. Enthusiast & Professional headsets can function without a PS6; all three headsets can function wired or wirelessly.

Two SKUs; $499 with Entry headset, less internal storage & digital only. $599 with Entry headset, more internal storage and fit-to-form disc drive for physical PS6 games.

The VR unit can be docked to the console base to combine processing ability and offer "full" performance, or the VR unit can be detached and used wired (or wireless) with the console base, or on its own, with reduced visual fidelity (but still much better than what current PSVR2 & PS5 can deliver). System designed with multi-user local play in mind.

Microsoft: Transition Xbox to a mini-PC NUC-style gaming device brand; 10th gen Xbox premium device comparable in power or slightly stronger than PS6. Runs Windows. Custom CPU & GPU (separate chips/components; PS6 uses custom APU with integrated CPU & GPU) with at least upgradable GPU. Upgradable RAM. Console form factor, with controller, and console gaming UI directly taken from the Xbox side.

At least two SKUs; one with weaker CPU & GPU, less RAM, less storage, but priced at $499 (depending on options). Series S equivalent. Second SKU is the more powerful one, with more RAM, storage, priced at $799. Both have upgrade options, both offer much more gaming performance than typical OEM mini-PC/NUC devices available at time of release, at similar or slightly lower pricing.
 
Well yes.. kinda.

This is how I arrived at that 40-50TF Raw TF number.

I used the PS5 as the base, as I feel its a better representation of what the minimum w can get is. So 36CU. 4x density bump puts us at 144CU. If we use the same ~2.2Ghz PS5 clock, then we end up with 40.5TF.

As you can see, I was being extremely conservative.

What is likely is that clocks may be as high as at east 3Ghz... that puts us at the 50TF point. But then you consider some other things, like say, maybe the CUs aren't even quadrupled, maybe what changes is how many instructions they can run simultaneously. And maybe there is more of a shift and focus on RT and AI hardware.

There is just SOOOOOO much we do not or cannot possibly know yet. Outside some raw rudimentary TF calculations.

Oh... and AMD doesn't have to go down that path officially. All they have to do is add AI cores to their GPUs. Which they are already rumored to be doing with their next gen GPUs. At this point, they don't have a choice.

Before you start calculating TF, a few things to keep in mind:

-CUs aren't going to scale down linearly with reduced node shrink. RDNA2 CUs for example are 62% larger than the CUs in the PS4's GPU; that was mentioned back at Road to PS5

-Because of that, you won't get a 4x CU density increase with a 4x density increase. Other things like SRAM don't scale down linearly with node shrinks either, same with memory controllers, etc.

-RDNA3 utilizes dual-issue shaders, so you don't really need 4x the CUs of PS5 to get 4x the TF performance. Although, IIRC dual-issue is with more specific operations, whereas if the physical CUs were quadrupled you'd have those CUs available for all tasks long as you can saturate them with work.
 
Top Bottom