Μικρά νέα και rumors για το PS5 all-in-one Thread

[Leon_DiZ]

Από τα παιδιά του DF στο ResetEra

Was this real time or not, though?

Yes. Was even intended to be playable at GDC on PS5.

How many FPS is the demo running at?

It's 30fps locked at 1440p with temporal up-sampling to 4K

Σχετικά με το ray tracing. Είπε η Epic ότι το demo δεν χρησιμοποιούσε το RT hardware του PS5. Στην ουσία κάνει brute force GI το οποίο είναι πάρα πολύ όμορφο αλλά όχι τόσο ακριβές και βαρύ όσο το ray tracing.

Θα μιλήσουν για το πως ακριβώς λειτουργεί ο φωτισμός τους μέσα στην εβδομάδα.

Από τι στιγμή που φαίνεται ότι δεν χρησιμοποιεί το hardware RT αυτό μπορεί να σημαίνει ότι μπορούν να το χρησιμοποιήσουν διαφορετικά ή να βελτιώσει τις επιδόσεις όταν χρησιμοποιηθεί(??). Θα δούμε

https://twitter.com/BrianKaris/statu...08919044415488

Τα posts ενός indie dev πάνω στο θέμα του φωτισμού.

Spoiler!

They mean it's not demonstrating vendor specific raytracing extensions, not that it doesn't utilize raymarching. Different GPU manufacturers have different extensions in their drivers, this is why you'll sometimes see things like "better on Nvidia." What they are saying is they ignored all of those, and did raymarching through GPGPU. In other words, using a core profile, so to speak in GL terms.

Without a doubt UE5 will support vendor specific extensions.

when you do graphics programming, you speak with a specific language to talk to your GPU. To make games interoperable between GPU manufacturers, they support a common, agreed upon set of standard features, called a "core profile." In order to be, for example, Vulkan or DirectX 12 compliant, a GPU must adhere to a set of core standards regarding what kind of language features their card supports.

Beyond this core profile are what are known as extensions. These are specific additional features that cards have specific to their build. Not only are there vendor-specific extensions, like extensions that only exist on Nvidia or AMD, but there are even architecture-specific extensions, features that only exist on certain ranges of cards from those GPU makers. Example, many Nvidia architectures can support DirectX 12, but only turing cards can access their RTX extensions which handle raytracing in hardware.

What they said in their reply, is that the demo was built to ignore everything except the base standard. So no Nvidia or no AMD specific extensions, to give a better example of what the tech will look like across all hardware, instead of targeting specific extensions that (largely) get ignored in larger applications anyways. By using those specific extensions, they could push things even harder, and those extensions are inevitably exposed to the programmer, but the demo was built to explicitly ignore them.

The demo was already 1440p @ 30. Add RT and I wonder how it would perform then?

It would perform better.

So the demo they showed was using raytracing code, and offloading that to hardware that supports it explicitly would improve performance? Is there some sort of efficiency loss in writing your own fragment shader other than it being an absolute pain in the ass to do so?

yes, there is. There are different cores inside of the GPU to handle different tasks, like RT cores, Tensor Cores, etc. These cores are application-specific integrated circuits (ASIC) built to handle tasks through electrical engineering. When they talk about RT cores, they're talking about hardware specifically built to handle the specific kinds of math needed for speedier raytracing. Relying on general purpose cores doesn't gain the advantage using RT cores does.

Because the demo already feaures raymarching, there is dynamic global illumination happening all over the demo. Those aren't being done with the hardware specifically in place yet to speed up those calculations. Using dedicated RT hardware will speed up the things already being done in the demo. They're showing off the stuff RT cores are built for, being brute forced with non-RT hardware. Lumen is, essentially, software-based simplified raymarching.

Global Illumination describes any sort of illumination that is not the simple interaction of a light source and a sample or surface being shaded. Shadowing, for instance, can be seen as a form of global illumination because it relies on more than the surface you're shading and the light source that is affecting it.

Ray casting is the act of casting a ray into the scene, starting from it's origin and (usually) finding the first object that it intersects with. When this is achieved by 'marching' the ray forward step by step through a volume or area it's called ray marching. Ray casting is the basic operation of ray tracing algorithms which can then use the origin and the hit point of the ray in order to determine how the origin might affect the hitpoint's color or vice versa. An example of such an algorithm is Path Tracing which I will not describe here.

This means that when people use the term ray tracing they're generally talking about techniques that cast rays into the scene. We have been able to write ray tracing code that runs on the GPU for a long time now but only recently has specialized ray tracing hardware that helps improve the 'speed' of ray tracing been available in consumer GPUs. The demo today was not taking advantage of that specialized hardware and was instead using 'software' based solutions that (i presume) used a more general compute shader.

IMO, the term "raymarching" should be used these days for GPGPU software based solutions, while "ray tracing" should be used for implementations specifically using RT ASICs to avoid confusion. Raymarching is the technique, raytracing is the hardware-based implementation.

Here's a good cheat sheet of raymarching algorithms from shader toy: https://www.shadertoy.com/view/Msf3Dr

https://www.resetera.com/threads/ue5...ant-to.204591/


Επίσης έτσι όπως λειτουργεί η Unreal φαίνεται να βασίζεται σε μεγάλο βαθμό το performance στην ανάλυση του παιχνιδιού (κάτι που νομίζω ότι γινόταν και στην UE4 σε ένα βαθμό)

Βut it's Sweeney's invocation of REYES that I'm particularly struck by. The UE5 tech demo doesn't show extreme detail at close range, it's also delivering huge draw distances and no visible evidence whatsoever of LOD pop-in. Everything is seamless, consistent. What's the secret? "I suppose the secret is that what Nanite aims to do is render effectively one triangle for pixel, so once you get down to that level of detail, the sort of ongoing changes in LOD are imperceptible," answers Nick Penwarden. "That's the idea of Render Everything Your Eye Sees," adds Tim Sweeney. "Render so much that if we rendered more you couldn't tell the difference, then as the amount of detail we're rendering changes, you shouldn't be able to perceive difference."

And that, in a nutshell, is the definition of a micro-polygon engine. The cost in terms of GPU resources is likely to be very high, but with next-gen, there's the horsepower to pull it off and the advantages are self-evident. Rendering one triangle per pixel essentially means that performance scales closely with resolution. "Interestingly, it does work very well with our dynamic resolution technique as well," adds Penwarden. "So, when GPU load gets high we can lower the screen resolution a bit and then we can adapt to that. In the demo we actually did use dynamic resolution, although it ends up rendering at about 1440p most of the time."

https://www.eurogamer.net/articles/d...-playstation-5