Predict: Next gen console tech (10th generation edition) [2028+]

[fehu]

I think that makes no sense to take that comment literally, and btw the jump to series x was really huge and didn't pay out.
In my opinion it will be something competent but not bleeding edge, hopefully infused with dedicated hardware for ai and what not, but not too much exotic because even internal developers won't waste time exploiting just one machine.

Plus
If it's true that they are returning to the dual sku scheme, they can't make a 40TF home console and a 4TF handled console, it would be a nightmare.
To predict the halo device we must define the low end one.
A starting point can be RX 7600M at 17TF / 2GHz / up to 90W / humongous 204mm^2.

 

[Globalisateur]

XSX already has exclusive hardware "AI tech" compared to PS5, how their studios are using it? More first party games actually use machine learning inference on PS5 than on XSX (using mostly FP16). Similar thing for DLSS. DLSS is mostly a software problem. DLSS exists because NVidia put a big amount of ressource (and intelligence) on developing the software. FSR sucks because compared to Intel or Nvidia, AMD sucks at software.

A big tech leap won't be enough if software and APIs are lagging behind the competition.

 

[cheapchips]

A big tech leap won't be enough if software and APIs are lagging behind the competition.

It feels like an even bigger issue for gen 10. As of today at least, if MS do get change vendors they get various bits are better than on AMD. Intel give them better AI upscaling and frame generation. Nvidia give them that and 'easy' to implement path tracing. That's nothing to say how these offering will evolve over the next couple of years.

Quite how much of that adds to any of MS's first party offerings? Do South of Midnight, Clockwork Revolution, Fable etc really look massively better than they do already?

 

[troyan]

XSX already has exclusive hardware "AI tech" compared to PS5, how their studios are using it? More first party games actually use machine learning inference on PS5 than on XSX (using mostly FP16). Similar thing for DLSS. DLSS is mostly a software problem. DLSS exists because NVidia put a big amount of ressource (and intelligence) on developing the software. FSR sucks because compared to Intel or Nvidia, AMD sucks at software.

A big tech leap won't be enough if software and APIs are lagging behind the competition.

What is this "exclusive hardware 'AI tech'"? There is nothing in the SoC. Sony and Microsoft have the same IP, the biggest difference is outside of AMD.

 

[cheapchips]

What is this "exclusive hardware 'AI tech'"? There is nothing in the SoC. Sony and Microsoft have the same IP, the biggest difference is outside of AMD.

Xbox added Int4 to the Series consoles. It isn't in PS5 / vanilla RDNA2.

 

[troyan]

Xbox added Int4 to the Series consoles. It isn't in PS5 / vanilla RDNA2.

So nothing. Int4 support is cheap and easy to implement but barely useful (i guess it is even useless). Gaming-Pascal supports Int8 and it is not used for gaming. There is a reason why nVidia went from Pascal to Volta within one year because "TensorCores (NPUs)" are much better for this kind of workload.

 

[cheapchips]

So nothing. Int4 support is cheap and easy to implement but barely useful (i guess it is even useless). Gaming-Pascal supports Int8 and it is not used for gaming. There is a reason why nVidia went from Pascal to Volta within one year because "TensorCores (NPUs)" are much better for this kind of workload.

To be clear, it wasn't me pitching Int4 as some mega AI thing!

 

[Globalisateur]

So nothing. Int4 support is cheap and easy to implement but barely useful (i guess it is even useless). Gaming-Pascal supports Int8 and it is not used for gaming. There is a reason why nVidia went from Pascal to Volta within one year because "TensorCores (NPUs)" are much better for this kind of workload.

Good to know. But my point was that MS was already promising big upcoming things thanks to exclusive int4 support on XSX. And we have seen nothing yet. It was already some kind of technological leap compared to PS5. But it was, how surprising, all PR BS.

DirectML – Xbox Series X supports Machine Learning for games with DirectML, a component of DirectX. DirectML leverages unprecedented hardware performance in a console, benefiting from over 24 TFLOPS of 16-bit float performance and over 97 TOPS (trillion operations per second) of 4-bit integer performance on Xbox Series X. Machine Learning can improve a wide range of areas, such as making NPCs much smarter, providing vastly more lifelike animation, and greatly improving visual quality.

Defining the Next Generation: An Xbox Series X|S Technology Glossary - Xbox Wire

[Editor’s Note: Updated on 10/21 at 11AM to ensure it is now reflective of the capabilities across both of our next-gen Xbox consoles following the unveil of Xbox Series S.] As we enter a new generation of console gaming with Xbox Series X and Xbox Series S, we’ve made a number of technology...

news.xbox.com

 

[function]

It was int8 and int8 and mixed precision, and basically DP4a compatible if I remember right.

So the Series consoles could run the GPU DP4a powered version of Intel XeSS, which seems to be a step up from FSR2 and very much within the performance envelope of both Series consoles. But so far, nothing like that for Xbox which is a bit disappointing.

I think that MS's requirements for Series consoles might be why PC RDNA2 supports DP4a while consumer RDNA 1 / PS5 doesn't.

 

[DavidGraham]

Anyway MS just said they'll have "unique hardware" as well as that "the biggest technical leap!" I guess the dual strategy I suggested is exactly what they're going for. Series S BC mobile and some ludicrous dedicated chiplet thing.

Maybe chiplets is indeed the answer, delivering biggest raw power leap within the constraints of APUs is not possible anymore, not with the current process nodes problems.

Funny I remember this thread 8 years ago, we were debating whether APUs will take over dGPUs, some thought this is going to happen by 2017 or 2018! I and others argued strongly against this, and here we are in 2024, and the strongest PC APU is on the level of a GTX 1650, while the strongest APU ever (Series X) is on the level of a RTX 2080.

dGPU vs APU spin-off

Why should the APU be any less capable? Perspective of silicon area budget.

forum.beyond3d.com

 

[Flappy Pannus]

Maybe chiplets is indeed the answer, delivering biggest raw power leap within the constraints of APUs is not possible anymore, not with the current process nodes problems.

Funny I remember this thread 8 years ago, we were debating whether APUs will take over dGPUs, some thought this is going to happen by 2017 or 2018! I and others argued strongly against this, and here we are in 2024, and the strongest PC APU is on the level of a GTX 1650, while the strongest APU ever (Series X) is on the level of a RTX 2080.

dGPU vs APU spin-off

Why should the APU be any less capable? Perspective of silicon area budget.

forum.beyond3d.com

While I was never of the mindset that APU's would 'take over' dGPU's, I sure as hell thought by the early 2020's we would have powerful APU's that would give at least the lower midrange cards a run for their money. Of course, 'midrange' is a moving target (at least in price), and that early belief was in part because I felt that HBM would 'naturally' get far cheaper. Welp.

Still, the pace of APU development on the PC has been relatively anemic - 2080 level performance in a PC APU would be revelatory. Maybe we'll get there with Strix Halo.

Also possible that the AI boom, and chiplets, at least fuel more concentrated effort in UMA's. Those VRAM budgets are very tight with LLM's.

 

[Sonic]

Yeah that's the thing I don't understand about AMD. One would think the market for APU's would allow for beefier ones on the desktop with a healthy margin. Strix Halo seems like a product I'd like to buy if they ever get around to releasing it. But still, one would think we could have had a wide array of APUs to choose from with varying performance levels. But that hasn't come to fruition.

 

[Frenetic Pony]

While I was never of the mindset that APU's would 'take over' dGPU's, I sure as hell thought by the early 2020's we would have powerful APU's that would give at least the lower midrange cards a run for their money. Of course, 'midrange' is a moving target (at least in price), and that early belief was in part because I felt that HBM would 'naturally' get far cheaper. Welp.

Still, the pace of APU development on the PC has been relatively anemic - 2080 level performance in a PC APU would be revelatory. Maybe we'll get there with Strix Halo.

Also possible that the AI boom, and chiplets, at least fuel more concentrated effort in UMA's. Those VRAM budgets are very tight with LLM's.

Zen 6 and RDNA5 are all chiplets, so AMD can theoretically put out arbitrarily sized APUs limited only by whatever the physical die has to be in/TDP. MX Ultra competitor should be available if one wants.

IDK what to think of "AI" as a "technical leap" in PR terms, then again people are eating up so sure why not. XDNA Ultra with 8 chiplets! Games now just run on OpenAI's Sora generating video in realtime, hallucinations like object impermanence are a feature

 

[RDGoodla]

We’re also invested in the next-generation roadmap. What we’re really focused on there is delivering the largest technical leap you will have ever seen in a hardware generation

That should mean most 'power', as that's the language we use. A technological leap between consoles is the technical difference, between PS1 and PS2, and PS2 and PS3. Which, let's be honest, is nonsense, unless the next-gen console is 1) coming out in 2036 or 2) is going to cost $2000

It is most likely that Ms wants to make most profit thereafter. So their hardware projects should be shifted from "high performance home console" to "hybrid handheld console + entry-level home console", because in this generation the entry-level series S sells much better than the high performance series X.

So technically Ms really can make a "most powerful handheld console " instead of a "most powerful home console" as long as it can surpass Swtich 2 spec.


I really don't think we will see another high performance home console or it only ends up with lower sales of Series X.

 

[McHuj]

I really don't think we will see another high performance home console or it only ends up with lower sales of Series X.

I'm hoping for the exact opposite actually. Once MS games are multiplatform, nothing is stopping them from creating a high-end, expensive console as a low install base is no longer a hinderance to software sales. Console and games sales can be decoupled. I'm hoping they take the approach of: you can play our games anywhere, but the best place to play is on our premium box.

 

[QPlayer]

I'm hoping for the exact opposite actually. Once MS games are multiplatform, nothing is stopping them from creating a high-end, expensive console as a low install base is no longer a hinderance to software sales. Console and games sales can be decoupled. I'm hoping they take the approach of: you can play our games anywhere, but the best place to play is on our premium box.

I would take it. However, is it worth it for MS to develop for a console that only buys a few million because of its expensive price?

 

[McHuj]

I would take it. However, is it worth it for MS to develop for a console that only buys a few million because of its expensive price?

I'm think more of it as a small all-in-one form factor PC. I think game development would still target the medium spec of across PC/PS5/6, but their own box would target the higher end of that spectrum. No one targets a 4090 spec either with their development, nor do that many people actually buy them. But it's highly profitable and by far the best card to play games on. I think if done right, MS could replicate that in the console space.

I would love for MS (or Sony) to re-engage with Nvidia. With Nvidia's new venture for custom chip design, the door is open for that opportunity.

 

[Frenetic Pony]

I'm think more of it as a small all-in-one form factor PC. I think game development would still target the medium spec of across PC/PS5/6, but their own box would target the higher end of that spectrum. No one targets a 4090 spec either with their development, nor do that many people actually buy them. But it's highly profitable and by far the best card to play games on. I think if done right, MS could replicate that in the console space.

I would love for MS (or Sony) to re-engage with Nvidia. With Nvidia's new venture for custom chip design, the door is open for that opportunity.

Sadly I doubt it, both are looking at mobile and Nvidia just doesn't have the CPU there. AMD trounces Nvidia's current CPU offerings and you need the whole APU package for a console. Thus why the rumors of both just going AMD again already seem entirely plausible.

 

[Shifty Geezer]

Sadly I doubt it, both are looking at mobile and Nvidia just doesn't have the CPU there. AMD trounces Nvidia's current CPU offerings and you need the whole APU package for a console.

Why is that necessary? Board costs skyrocket with more chips?

 

[QPlayer]

Maybe after the release of the PS5pro (does it really exist?) model, in December they will introduce an excellent technology as they talked about. If you really want the best console, you can use chiplet design and separate CPU and GPU.