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

But that’s just the cost of the wafers. 7nm vs 5nm has roughly a 45-50% density improvement so the same design on 5nm vs 7nm will require far fewer wafers (assuming similar yields) in addition to benefits in power consumption allowing cheaper cooling. It’s a pretty complex design and optimization puzzle.
 
Unless they're just after volume, do you think they can be more efficient than TSMC? (Intel might be after volume customers)
TSMC has around 37% profit margin, with plans to increase that next year. Intel and Samsung would probably accept below 5% profit margin to keep their foundries occupied, and consoles can keep them occupied for a long time. I don't think that the next consoles will have a cutting edge process nodes like PS5 and Xbox series, and it's alright, if they can keep costs down.
 
Last edited:
But that’s just the cost of the wafers. 7nm vs 5nm has roughly a 45-50% density improvement so the same design on 5nm vs 7nm will require far fewer wafers (assuming similar yields) in addition to benefits in power consumption allowing cheaper cooling. It’s a pretty complex design and optimization puzzle.
If it was cheaper overall, than PS5 slim would probably be 5nm. If they aren't using it, then it's sadly not worth it. I'm sure they did the math on those things.
 
TSMC has around 37% profit margin, with plans to increase that next year. Intel and Samsung would probably accept below 5% profit margin to keep their foundries occupied, and consoles can keep them occupied for a long time. I don't think that the next consoles will have a cutting edge process nodes like PS5 and Xbox series, and it's alright, if they can keep costs down.

Margin doesn't tell us about relative costs unfortunately. They may only be able to undercut TSMC slightly at lower margins if they're less efficient/low yield etc.

I'm not strongly disagreeing with you on consoles being on conservative nodes, but closer they can get to cutting edge would be interesting.
 
Margin doesn't tell us about relative costs unfortunately. They may only be able to undercut TSMC slightly at lower margins if they're less efficient/low yield etc.

I'm not strongly disagreeing with you on consoles being on conservative nodes, but closer they can get to cutting edge would be interesting.
I really hope they can undercut TSMC, or we're screwed with prices. Or maybe they can get a favorable agreement with TSMC, but why would they accept it? They got AI hardware to make for years 😪
 
But it's really awkward, because it partly invalidates the previous generation. There is a strong force to pot commit to pencil ray tracing, almost everyone is researching better ways to do importance sampling and post-filtering and have blinders on that they are on a fundamentally wrong road for real time rendering. RTX drove the entire industry down the wrong path.
Is it really "RTX" that sent the industry down that path, or Microsoft with DXR? Once ray tracing was added to the DirectX standard, it was inevitable that IHVs would dedicate transistors to accelerating it and game developers would use it more and more.
 
AMD could/should have pushed for exposing lower level implementation (AABB/tri intersection shader instructions for one). Much of the hardware is much more flexible than RTX/DXR exposes. But I already have a thread about that.
 
AMD could/should have pushed for exposing lower level implementation (AABB/tri intersection shader instructions for one). Much of the hardware is much more flexible than RTX/DXR exposes. But I already have a thread about that.

Pre building BLAS is pretty much a universal feature of RT on console now, which DXR doesn't allow (among other issues). That the API has issues matching console for BVH stuff is the biggest problem that I know of generally for matching PC to console. That Nvidia pushed Microsoft into such arbitrary limitations is definitely an issue.

But even moreso is that a lot of this hardware could be useful for generalized functions. Ray sorting is just as applicable to tracing SDFs as triangles through BVHs. As is things like smart prefetch hw as going through say, a sparse SDF should be just as predictable and latency bound as going through a BVH tree.

If AMD were smart they'd leverage their position as console provider to MS to include as much flexibility as they can, make the API conform to their standards instead of the API conformign to Nvidia's.
 
Last edited:
The most promising thing I can see for a mobile only next gen console is pushing the Switch/Switch Go model further.

Let's say you have Steamdeck 2 with a 16cu mobile only version for $399. "Docked" version could easily be a 32cu version (yay chiplets) with a solid state cooler for $499, the extra 16cu only activating in docked mode with the cooler on.

That solid state cooling is expensive power wise wouldn't matter, you only use it docked to a 65w charger or whatever. This also allows a specific low end mobile only chip version on top of mobile only physical casing.

Maybe it'd be $549 and 40Cu AMD wise (say, RDNA4/5), 4070(s?)ish Nvidia wise etc. as this is the hockey stick point where I feel like return on investment in visual quality and performance both really starts to drop off for modern games. You could go out and get a 5090 by the time Steamdeck 2 comes out, but the frank difference between 120+ and 60 is that 60 is above flicker/fusion threshold for human vision so 95% of people don't actually care. Similarly you can turn up the visual quality but noticeable difference just gets less and less, and you're still not going to get Hollywood vfx levels of image quality in terms of 0 aliasing/ghosting/etc. (at least on modern games).
 
Let's say you have Steamdeck 2 with a 16cu mobile only version for $399. "Docked" version could easily be a 32cu version (yay chiplets) with a solid state cooler for $499, the extra 16cu only activating in docked mode with the cooler on.
So 32 CUs in the handheld, but disabled in handheld mode and enabled when docked with a high-end cooling solution?
 
So 32 CUs in the handheld, but disabled in handheld mode and enabled when docked with a high-end cooling solution?
Yes, and a mobile 80-100 CU eGPU can be connected to the dock.

1, Xbox handheld that connects to a TV with the dock: $400
2. Optional powerful Xbox eGPU for handheld kit: $400
 
Last edited:
So 32 CUs in the handheld, but disabled in handheld mode and enabled when docked with a high-end cooling solution?

Yeah, but the cool thing (heh) would be that the cooling is built in. No "dock", an EGPU is almost entirely dependent on cooling not fitting into a thin and light which is already solved here, nothing to lose, everything is fully portable in the actual handheld itself which is nigh as thin and light as the standard version.

Solid state cooling is a new cooling solution with no moving parts, it's super tiny and thin but costs way more power (even a watt or 2 is super costly for a handheld) than a fan and can cool up to 50w at the highest end today. If you're plugged into an outlet you don't care about that power cost, go ahead and have your thin and light draw 50w+.
 
Last edited:
In fact, the dock is a simple holder that holds the handheld and the optional eGPU. Technically it can be solved and it can even look good.
 
WCCFTECH spoke with a game developer at Gamescom 2024, who confirmed PS5Pro specs is already in the wilds among devs.

a developer that I won't name openly said they had received the specs of the PS5 Pro and were confident that Unreal Engine 5 would run much better on the new hardware compared to the regular PlayStation 5

In a recent live stream, the Italian website Multiplayer mentioned that it had heard from a developer that they had delayed their game due to the upcoming launch of the PS5 Pro. I can confirm that I heard something similar during one of my appointments at Gamescom 2024.

 
Back
Top