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

Even the founder of Godot believes that virtual geometry is largely incompatible with RT ...


Alright since the only signs left of a standardized iteration on the current RT API has become stillborn with yet more extensions on virtual geometry to materialize on the way, we're quickly entering the prototyping phase of the next generation systems without having made any visible progress to solving outstanding issues with RT ...
 
Even the founder of Godot believes that virtual geometry is largely incompatible with RT ...
You phrase that as if he's the ultimate authority and if he doesn't think it can work (on Godot), it's a dead-end. Why should Juan Linietsky's opinion be valued above anyone else's (including the engineers at Epic and elsewhere)? Has he created a better ray-tracing solution for his traditional meshes in Godot? Only Godot games I've seen are 2D where it's beloved by indies.
 
You phrase that as if he's the ultimate authority and if he doesn't think it can work (on Godot), it's a dead-end. Why should Juan Linietsky's opinion be valued above anyone else's (including the engineers at Epic and elsewhere)? Has he created a better ray-tracing solution for his traditional meshes in Godot? Only Godot games I've seen are 2D where it's beloved by indies.
There are actual 3D content released with Godot (even simple games too) and in his case, he currently doesn't envision virtual geometry ever being a thing with RT especially for his own project ...

Tessellation with virtual geometry is already adding some performance overhead and the upcoming support for skinned meshes will only make the situation worse. I suspect that the next major milestone might be getting support for dithered transparency/opacity and at that point the only sensible option left is to give up on the idea of doing real-time ray tracing for those circumstances ...

Another thing that I have yet to see games do is combine emissive materials with nanite meshes but RT area lights with arbitrary surfaces that's composed of many lit triangles is probably another incoming performance cliff since unreal engine still can't support true rectangular lights for dynamic lighting to this day ...
 
Has he created a better ray-tracing solution for his traditional meshes in Godot?
He is actually planning a full path tracing solution for his engine relying on meshlets, so he completely ditched virtual geometry and any nanite like solution.


"The first thing to understand is that Godot is not Epic. The project does not have 150 veteran graphics engineers who can maintain an insanely huge renderer. [...] First of all, it should be ray/path-tracing only (with base raster pass). This saves incredible amounts of work, as shadows, GI, reflections, etc. would all be ray traced. Hybrids like Unreal or Unity HDRP are far too much complexity, and hardware is getting there anyway"

"As for Nanite. The problem with it is that it is also just too complex; it works as a hierarchical meshless LOD, plus it's not that friendly with raytracing, which is our main goal. Instead, a much simpler approach can be using more traditional auto-LODs with meshlets, all GPU-driven. This works great with raytracing (just stream LOD levels together). Has a bit more overdraw? Sure, but remember, we don't have shadow maps, so we don't care. Downsides? Probably slower for insanely complex, huge geometry if it spans hundreds of meters. Just divide it up (which most, if not all, games, do anyway), but for anything else, you still get great performance and detail. Also likely will use meshlet LRU less efficiently, but to be honest, unless you are an AAA studio, you don't care about these things; you have something close that solves a similar problem, and on the Godot side, something way simpler to do and maintain"


 
So when we have a working version and a couple of games that use it, we'll be able to compare the results. Although interestingly he'll be leveraging RT hardware, no? I'm not seeing anything in Godot's choice for smaller devs to inform us and the console designers on where to invest their silicon. If we had a superior software renderer in Godot showing it can be done, sure, but until then there's nothing I can see here saying RTRT hardware is a bad choice for next-gen hardware.
 
I'm not seeing anything in Godot's choice for smaller devs to inform us and the console designers on where to invest their silicon.
Juan is implying that console designers can't be on the fence anymore between the choice of virtual geometry vs ray tracing ...

Virtual geometry is both growing and advancing faster in terms of complexity than what ray tracing can handle on many hardware. Next generation hardware will never be able to catch up with RT if virtual geometry sees continual refinements which is why Juan intends to forgo the technology ever being an option for Godot in favour of RT ...
 
Last edited:
Virtual geometry is both growing and advancing faster in terms of complexity than what ray tracing can handle on many hardware.

What are you basing this on?

Next generation hardware will never be able to catch up with RT if virtual geometry sees continual refinements which is why Juan intends to forgo the this technology ever being an option for Godot in favour of RT ...

Never say never.

Lumen is a hack and Epic has admitted as much (they want to do RT but it’s not fast enough). If anything has an expiration date I would say Lumen GI and VSM’s days are numbered. The end state is RT + high fidelity geo (virtualized or otherwise).
 
What are you basing this on?
Constant enhancement additions like WPO, displacement, skeletal meshes only makes it more likely that future hardware will exhaust their new performance headroom before we even approach RT ...
Never say never.

Lumen is a hack and Epic has admitted as much (they want to do RT but it’s not fast enough). If anything has an expiration date I would say Lumen GI and VSM’s days are numbered. The end state is RT + high fidelity geo (virtualized or otherwise).
Arguably, we shouldn't even be considering software Lumen or VSM from the get go if the industry is serious about doing RT for virtual geometry since it's only going to get harder from here on out especially when nanite is a "moving target" which by itself has outstanding issues yet to be resolved such as fully deformable meshes, transparency, and god knows what else beyond that. A ~2-3x improvement in a new generation still isn't going to let us reach the endgame seamless integration of RT + virtual geometry ...
 
Constant enhancement additions like WPO, displacement, skeletal meshes only makes it more likely that future hardware will exhaust their new performance headroom before we even approach RT ...

That depends on RT performance of future hardware and APIs which are currently unknown. Also last I checked Nanite is an Epic exclusive. There are other engines out there.

Arguably, we shouldn't even be considering software Lumen or VSM from the get go if the industry is serious about doing RT for virtual geometry since it's only going to get harder from here on out especially when nanite is a "moving target" which by itself has outstanding issues yet to be resolved such as fully deformable meshes, transparency, and god knows what else beyond that. A ~2-3x improvement in a new generation still isn't going to let us reach the endgame seamless integration of RT + virtual geometry ...

Yep lots of moving parts which is exactly why it’s premature to declare that RT and high density geometry can’t coexist. Virtual geometry (and the specific implementation in Nanite) isn’t the only option for high fidelity assets.
 
That depends on RT performance of future hardware and APIs which are currently unknown. Also last I checked Nanite is an Epic exclusive. There are other engines out there.

Yep lots of moving parts which is exactly why it’s premature to declare that RT and high density geometry can’t coexist.
When we look at vendors spending hardware to make RT 'happen' only to next have it reversed soon thereafter do you think it's still sensible to double down again on that approach if the outcome remains identical ? It's precisely because of that uneven progression in graphics technology which can potentially impede advancements in other areas ...
Virtual geometry (and the specific implementation in Nanite) isn’t the only option for high fidelity assets.
You're right since we had hardware tessellation before but it only further begs the question that much like it why should we specifically justify extending RT to fervently believe that it'll ultimately "work out" somehow in the end ?

What exactly is captivating this time around about spending more die area on the RT implementation when it's a technical debt right now with no clear solutions to fix/mitigate it ? I predict that AAA console games shipping software Lumen or an iteration thereof in the future is still going to be an unavoidable reality when the next generation comes around. I don't think software Lumen was only ever intended to last for the duration of the current generation or was ever defined to be a stopgap or a stepping stone of sorts for hardware Lumen ...
 
@Lurkmass what exactly does software lumen do to bridge the game of 1:1 representation and continuous LOD that nanite has? It is even more amorphous, clumpy and non-supporting of skinning / transparency atm.
I think both solutions are imperfect.
 
When we look at vendors spending hardware to make RT 'happen' only to next have it reversed soon thereafter do you think it's still sensible to double down again on that approach if the outcome remains identical ? It's precisely because of that uneven progression in graphics technology which can potentially impede advancements in other areas ...

Which vendors have reversed their spending on RT?

You're right since we had hardware tessellation before but it only further begs the question that much like it why should we specifically justify extending RT to fervently believe that it'll ultimately "work out" somehow in the end ?

You seem very confident that lots of unproven tech will work out somehow. RT is no different and has much more of a track record than the alternatives you’re proposing.

What exactly is captivating this time around about spending more die area on the RT implementation when it's a technical debt right now with no clear solutions to fix/mitigate it ? I predict that AAA console games shipping software Lumen or an iteration thereof in the future is still going to be an unavoidable reality when the next generation comes around. I don't think software Lumen was only ever intended to last for the duration of the current generation or was ever defined to be a stopgap or a stepping stone of sorts for hardware Lumen ...

Nothing’s really changed on the RT front since 2018. The major APIs are pretty stagnant. However it’s still the most promising tech in terms of delivered results. That alone is sufficient reason to keep investing and optimizing. The alternative is to spend resources on half baked, incomplete and in many ways more complicated solutions which isn’t a great outcome.
 
Which vendors have reversed their spending on RT?
I meant that statement to be in terms of game integration ...
You seem very confident that lots of unproven tech will work out somehow. RT is no different and has much more of a track record than the alternatives you’re proposing.
There's a major difference between tech that's been extensively tested with more flaws being discovered over time in comparison substitutes that haven't been tried yet ...
Nothing’s really changed on the RT front since 2018. The major APIs are pretty stagnant. However it’s still the most promising tech in terms of delivered results. That alone is sufficient reason to keep investing and optimizing. The alternative is to spend resources on half baked, incomplete and in many ways more complicated solutions which isn’t a great outcome.
@Bold That's not how feature development works. Hardware vendors absolutely NEED to show us HOW they can realistically get there!

A promising prototype implementation are usually shown first often BEFORE we commit to adding hardware for it. When we look at examples like DXR, Nvidia didn't just suddenly simultaneously develop HW for the API on the fly when they originally did a preview demonstration of the technology with their Volta architecture to prove that it could be viable in practice. Similarly for Work Graphs, AMD did not spontaneously come up with the concept to support a better scheduling model for forward progress guarantees until Epic Games made their case with an internal demonstration on consoles that virtual geometry can be optimized with persistent threads. Intel was purely content for a long time with just emulating 64-bit atomics and having a brittle GPU driven rendering implementation but they quickly changed opinions after the introduction of nanite and games started abusing ExecuteIndirect in nasty ways ...

Virtual texturing is a really common technique these days yet we don't see hardware vendors clamouring to do better implementations of tiled resources when the technology was *exactly envisioned* for that use case. In many instances, hardware iteration follows these trials or observation of trends ...
 
meant that statement to be in terms of game integration ...

Even Cerny is surprised how fast RT got adopted in the early days of the PS5, he thought it's only going to get late adoption in the life cycle, but he was proven wrong. So he is doubling down on RT for PS5 Pro.

Every game engine under the sun has integrated RT support, game integration is high, path tracing integration is accelerating. The first half of 2024 had very few AA/AAA titles that launched, despite this we already received more than a dozen RT integrations.

Dragon's Dogma 2
Diablo 4
Atomic Heart
Suicide Squad
Homeworld 3
Starship Troopers
Skull and Bones
F1 2024
Ark Survival Ascended
Gray Zone Warfare
Supermoves
The First Descendant

You brought the creator of Godot into the discussion as a point against the adoption of RT, when in fact it's a point in favor of RT since Godot is moving to embracing path tracing.
 
Last edited:
Even Cerny is surprised how fast RT got adopted in the early days of the PS5, he thought it's only going to get late adoption in the life cycle, but he was proven wrong. So he is doubling down on RT for PS5 Pro.

Every game engine under the sun has integrated RT support, game integration is high, path tracing integration is accelerating. The first half of 2024 had very few AA/AAA titles that launched, despite this we already received more than a dozen RT integrations.
Cerny's taken aback because the transition has been anything but "believing in generations". You would not have much of a relevant list if those games weren't being released for last generation platforms or based their technology around foundations like UE4 ...
You brought the creator of Godot into the discussion as a point against the adoption of RT, when in fact it's a point in favor of RT since Godot is moving to embracing path tracing.
Godot's pretty much irrelevant for AAA games and even mobile games too. Not even Unity (Godot's closest competitor) has proper RT integration yet for PC/console platforms since their implementation half backed ...
 
Cerny's taken aback because the transition has been anything but "believing in generations"
That's your tainted interpretation of this. It's incorrect. His statement is quite clear in that regard. He is happy about the speed of RT adoption especially from 1st party titles (which went against his expectation).

You would not have much of a relevant list if those games weren't being released for last generation platforms or based their technology around foundations like UE4 ..
Almost half of the announced next gen releases are using RT, even UE5 games are using RT and path tracing, so that's another incorrect point.

Godot's pretty much irrelevant for AAA games and even mobile games too
You are the one who brought Godot into the discussion, not me! Now you are calling it irrelevant, if it's irrelevant why bring it up at all?
 
That's your tainted interpretation of this. It's incorrect. His statement is quite clear in that regard. He is happy about the speed of RT adoption especially from 1st party titles (which went against his expectation).
Is it really when they're releasing a Lego Horizon spinoff game on the Switch ?
Almost half of the announced next gen releases are using RT, even UE5 games are using RT and path tracing, so that's another incorrect point.
How many of those UE5 games feature high scene complexity (AKA aren't releasing on last generation or started development w/ last generation technology) ?
You are the one who brought Godot into the discussion, not me! Now you are calling it irrelevant, if it's irrelevant why bring it up at all?
Just because I don't see much value in their technology doesn't mean I don't see value in Juan's statement which was the principal basis I was bringing up ...
 
There's a lot in Unity that's lacking. Lack of inclusion in Unity isn't evidence of any level of viability for a tech thanks to crappy leadership over recent years. You're better off looking at smaller, more agile engines like Flax. Which also target non-HW RT GPUs, so the lack of HWRT inclusion in those engines also isn't indicative of a lack of viability for RT hardware. You'll need to look at which engines are being used for console and what they are doing. What's Insomniac doing? Guerilla? Naughty Dog? these are the people informing Sony, and MS will have their developers saying what they want too.
 
Back
Top