DavidGraham
Veteran
What's strange in RTX effects being slower on non RTX hardware?
Last edited:
Next gen lighting technologies - voxelised, traced, and everything else *spawn*
- Thread starter Scott_Arm
- Start date
Ike Turner
Veteran
Because, as I posted above, the RTX path in OptiX6 which is now "recommend optimal path" by Nvidia for non Turing GPUs is actually the opposite... Not "optimal" at all..Why are results from the same people far lower than the OptiX 5 results?
Previous Benchmark, OptiX 5.
Titan V gets 108 M samples/s.
1080Ti gets 55.
Latest benchmark, OptiX 6.
Titan V gets 67 M samples/s.
1080Ti gets 27.
The only difference I can see is 'custom settings : 512 pass' whatever that means.
DavidGraham
Veteran
Did you even watch the video? The Titan V is faster with RTX ON, the 1080TI is either faster with it on, or the same speed.
Ike Turner
Veteran
? OptiX6 with RTX path on non-Turing GPUs is slower than OptiX5 & OptiX6 with RTX path off for the same end result.. What's strange here.. I wonder
DavidGraham
Veteran
Again that's not what the video shows, The Titan V is faster with RTX ON, the 1080TI is either faster with it on, or the same speed.
512 pass is lower quality I presume than standard.
Octane Render ... Results with a 2070.. 3.05x faster with RTX on!
Last edited:
Another alternative, very impressive: https://www.gamedev.net/forums/topi...g-on-a-2014-gpu/?tab=comments#comment-5401449
The thing I find most shocking is none of the RTX content using raytracing is looking that good. This is on a 750Ti:
"It runs at 40fps, 1920x1080 on my GTX 750 ti, with 3 coherent and 2 incoherent rays per-pixel (so about 300 million rays a second)."
No amount of RTX BVH hardware in a 750Ti would get results like this. It does seem a more game-friendly approach in terms of what's possible. Raytracing is the ideal for offline where quality is priority one, but if the volumetric approaches can look this good while being this fast, they seem a better starting point, working to solve their limitations in future hardware, than trying to get raytracing fast enough.
"It runs at 40fps, 1920x1080 on my GTX 750 ti, with 3 coherent and 2 incoherent rays per-pixel (so about 300 million rays a second)."
No amount of RTX BVH hardware in a 750Ti would get results like this. It does seem a more game-friendly approach in terms of what's possible. Raytracing is the ideal for offline where quality is priority one, but if the volumetric approaches can look this good while being this fast, they seem a better starting point, working to solve their limitations in future hardware, than trying to get raytracing fast enough.
But you could use RTX to make stuff that looks that good, ofc. And it would be faster.
On the other hand: You could also make this stuff faster using clever denoising, caching, whatever...
At the end is the question: How much RT work do we really need per frame, and how much chip area is it worth to dedicate for the speed up? Even if DXR/RTX would be to my liking, it would still ask for this.
The question then becomes, why don’t we see it in any game other than TTC? And even that didn’t look quite this good, save for the noise?
It's an interesting discussion. Back through this or the Turning impact on consoles thread, we found one game using volume-tracing and it ended up really slow in games versus early prototypes. But it was years old and there wasn't much (any?) momentum to develop and explore this tech. Also, it's just rendering the seen and not running a game with surface shaders etc. The 750 Ti released 5 years ago, so the hardware for this exact demo isn't new, and it's just been waiting for people to develop it. Which makes you wonder how much better at volume-tracing modern and future hardware will be.
DavidGraham
Veteran
The StarWars demo and Dancing Robot have higher quality. Imagine what you can do on a 2080Ti with BVH acceleration.
That's why I think games developed for ray tracing from the ground up are going to offer striking visuals at very high performance. We are limited to games with mostly rasterization elements for now.
Hahaha, i know why:
Because of FF raster hardware devs are so focused on using it most efficiently - they just accept the restrictions after some time. No time to go beyond - deadlines, risk of failure...
And now history repeats. We have nothing learned.
No. They use precomputed GI and RT only for reflections. The higher quaility comes from artist work and presentation.
No. RT alone is no solution for anything. I don't want games that show perfect reflections but still rely on baked GI, and GPU already struggles at small scenes. The raster elements are no limitation - they help to make it barely fast enough.
Why do you see raster as 'limited', although RTX is limited at least as much too?
If you want unlimited stuff, then you are welcome to join my philosophy
I think they both do a little more such as traced AO.
https://www.unrealengine.com/en-US/...ng-in-unreal-engine-4-with-ilmxlab-and-nvidia
- Ray-traced area light shadows
- Ray-traced reflections
- Ray-traced ambient occlusion
DavidGraham
Veteran
Dancing Robot uses RT GI.
https://www.allegorithmic.com/blog/...uture-nvidia-s-project-sol-textured-substance
It's limiting the amount of optimization possible for ray tracing, developers optimizing for rasterization only.
Last edited:
Games are very different beasts from
A tech demo like this.
I mean, the population is giving hell to BioWare for loading screens. Lol.
There is only so much you can do in a scene, you have all sorts of bottleneck limits per second.
Oh, ok, I see it. It updates as fast as my on stuff, if i run the CPU version of it.
Devs can not optimize fixed function hardware, so what do you consider as 'optimizations'? Removing smaller dynamic objects? Limiting to a single light and bounce? Mix with SSR?
That's no RT optimizations! It's only optimizing the application of it.
I contacted Boostclock who noticed the same thing and have approached the developer behind Fermat, so hopefully we get clarification soon.
Just because the CEO of a company says his product is awesome, it doesn't mean it is awesome.
Even if Jenson says he used alien technology recovered from the Area 51 crash site, the images speak by themselves.
MOD: Please don't spam image dumps. Present arguments with images as evidence.
Even if Jenson says he used alien technology recovered from the Area 51 crash site, the images speak by themselves.
MOD: Please don't spam image dumps. Present arguments with images as evidence.
Last edited by a moderator:
...ok (?)
Contains some valid points, but... RTX on does not mean everything can be assumed to be correct, because RT is only used for certain effects.
E.g. Metro uses RT only for the skylight, no other light sources, and smaller dynamic objects are not raytraced at all, thus the red circles.
The UE4 demo uses RT only for reflections i guess, so shadows do not change.
Some very nice voxel AO, they show GI / reflections too on webpage:
Contains some valid points, but... RTX on does not mean everything can be assumed to be correct, because RT is only used for certain effects.
E.g. Metro uses RT only for the skylight, no other light sources, and smaller dynamic objects are not raytraced at all, thus the red circles.
The UE4 demo uses RT only for reflections i guess, so shadows do not change.
Some very nice voxel AO, they show GI / reflections too on webpage:
Similar threads
