GART: Games and Applications using RayTracing

Status
Not open for further replies.
Thanks for the viddy Trinibwoy, it was easy enough for even me to understand and I found it fascinating!

One thing though, I'm not sure I always prefer how the ray-tracing looked over conventional. (just my personal opinion, but I like a bit of a gamey look I guess), and I am DEFINITELY sure that at this moment the effect isn't worth the performance hit. It's a great subtle effect sure, but 26% performance hit and that's probably best case scenario? No thank you, I'll wait until it gets a bit cheaper/more universal I think.

Well, unless someone wants to toss an rtx 2080 my way or something. ;)
 
Pretty neat shadow differences so far with RT shadows ON vs. Off in the new COD.

Shadow Maps:
shadowmaps2vgju3.jpg

RT Shadows:
rtshadows2uukeb.jpg

Shadow Maps:
shadowmapsufjdo.jpg

RT Shadows:
rtshadowseakxx.jpg
 
Why on earth would you have performance graphs for the latest big graphical game with RT benchmarked on a 6 year old Haswell 4Ghz CPU?
 
Why wouldn't you? It's not like it makes a difference in this title, even if it would be running at stock it would still yield same results as i9-9900K overclocked to 5.2 GHz according to these results:
At 1080p, 14 different CPUs across 8 generations and spanning 3-5Ghz, including all Ryzen generations, all have the same average and max framerate? Weird.

My comment wasn't really regarding performance though but rather the choice of CPU by Nvidia. A massive company wanting to sell their GPUs and either they a) benchmarked internally on some old Haswell, or b) picked a result from someone like the source you linked. Either way they chose that CPU? It's just seems stupid.
 
At 1080p, 14 different CPUs across 8 generations and spanning 3-5Ghz, including all Ryzen generations, all have the same average and max framerate? Weird.

My comment wasn't really regarding performance though but rather the choice of CPU by Nvidia. A massive company wanting to sell their GPUs and either they a) benchmarked internally on some old Haswell, or b) picked a result from someone like the source you linked. Either way they chose that CPU? It's just seems stupid.
You can also view it as "hey, even though you got old CPU you can still buy our top of the line GPU and get this"
 
Quake II RTX v1.2 Update: Download For Free For Improved Graphics and More Features
November 26, 2019

quake-ii-rtx-update-v1-2-004-v1-2-850px.jpg


quake-ii-rtx-update-v1-2-001-v1-2-850px.jpg

New Features:
  • Added support for dynamic resolution scaling that adjusts rendering resolution to meet an FPS target.
  • Added support for multiple reflection or refraction bounces.
  • Added light coloring by tinted glass.
  • Added support for security camera views on some monitors in the game.
  • Added god rays in reflections and refractions, improved god rays filtering.
  • Added a spatial denoiser for the specular channel.
  • Added support for loading custom sky (portal) light meshes from .obj files, and added portal light definitions for many maps in the base game.
  • Added triangular lights for laser beams: https://github.com/NVIDIA/Q2RTX/issues/43
  • Added “shader balls” to the shipping builds.
  • Added cvar `pt_accumulation_rendering_framenum` to control how many frames to accumulate in the reference mode.
  • Added cvar `pt_show_sky` to make analyzing skybox geometry easier.


Fixed Issues:
  • Fixed the stutter caused by Steam overlay by updating to the latest version of SDL2.
  • Fixed Stroggos atmospheric scattering (sky color) and overall sky brightness.
  • Fixed light scattering on the clouds.
  • Fixed the issue with overexposed player setup screen: https://github.com/NVIDIA/Q2RTX/issues/18
  • Fixed the sudden darkening of the screen: https://github.com/NVIDIA/Q2RTX/issues/26
  • Fixed the "PF_centerprintf to a non-client" warnings that happened on the "command" map when the computers are blown up, instead of informative on-screen messages.
  • Fixed missing GI on reflections of the first person model.


Denoising and image stability improvements:
  • Improved image quality and temporal stability of reflections and refractions by computing correct motion vectors for reflected surfaces and surfaces visible through flat glass.
  • Disabled the pixel jitter when temporal AA is turned off.
  • Added sample jitter to the spatial denoiser to improve the noise patterns that appear after light invalidation.
  • Improved image stability around blinking lights by using the light intensities from the previous frame on gradient pixels.
  • Improved stability of indirect lighting from sphere lights by limiting their contribution.
  • Added storage scaling for all lighting channels to avoid color quantization.
  • Fixed flickering that happened when the number of dynamic lights changes.
  • Improved sharpness of textured glass and similar transmissive effects by passing them around the denoiser.
  • Added multiple importance sampling of specular reflections of direct lights.
  • Replaced sphere lights that were used for wall lamps (mostly in the “train” map) with polygon lights to reduce noise.
  • Added an upper limit on sky luminance to avoid oversampling the sky in shadowed areas and thus reduce noise from direct lights.
  • Added light sampling correction based on statistical per-cluster light visibility. The idea is, if we see that a light is usually not visible, let's not sample it so much.


Material improvements:
  • Metals are now rendered much better thanks to the denoiser and BRDF improvements.
  • Over 400 textures have been adjusted or repainted.
  • Removed the nonlinear transform from emissive textures, and reduced the negative bias applied to them.
  • Force light materials to be opaque to make the laser lights in mine maps appear in reflections.
  • Restore specular on materials with roughness = 1, but make specular on rough dielectrics much dimmer.


Shading and BRDF improvements:
  • Fixed scaling of diffuse and specular reflections: https://github.com/NVIDIA/Q2RTX/issues/37
  • Fixed relative brightness and spotlight terms for different light types.
  • Hemisphere sampling for indirect diffuse tuned to make the results better match the cosine-weighted sampling in reference mode.
  • Trace more specular rays for shiny metals that do not need indirect diffuse lighting.
  • Fixed the misdirection of second bounce lighting into the diffuse channel when the light path starts as specular on the primary surface: https://github.com/NVIDIA/Q2RTX/issues/25
  • Replaced regular NDF sampling for specular bounce rays with visible NDF sampling: https://github.com/NVIDIA/Q2RTX/issues/40
  • Tuned fake specular to be closer to reference specular.
  • Added normalization of normal map vectors on load to avoid false roughness adjustments by the Toksvig factor.
  • Improved roughness correction to not happen on texture magnification, and to better handle cases like zero roughness.
  • Fixed the computation of N.V to avoid potential NaNs: https://github.com/NVIDIA/Q2RTX/issues/23


Misc Improvements:
  • Removed the multiplayer specific sun position setting, and changed the remaining setting to be morning by default.
  • Changed the default value of texture LOD bias to 0 for extra sharpness.
  • Use nearest filter for upscaling from 50% resolution scale (a.k.a integer scaling).
  • Made the brightness of sprites, beams, particles, smoke, and effects like quad damage be similar to overall scene brightness, to avoid having them washed out under direct sunlight.
  • Made the explosions and other sprites appear in low-roughness specular reflections.
  • Changed the flare lights to be derived from the flare model entity instead of the effect event, which makes the lights follow the flares smoothly in flight, and reduces flicker that was due to frequent light creation and destruction.
  • Removed insane and dead soldiers with `nomonsters 1`.
  • Random number generator now produces enough different sequences to cover thousands of frames for reference mode accumulation.
  • The `pt_direct_polygon_lights` cvar has a new meaning when set to -1: all polygon lights are sampled through the indirect lighting shader, for comparison purposes.
  • Moved the first person player model a bit backwards to avoid having it block reflections.
https://www.nvidia.com/en-us/geforce/news/quake-ii-rtx-v1-2-update/
 
Blender 2.81 Releases With NVIDIA RTX And Intel Open Image Denoise Support
November 21, 2019

On the topic of performance, it seems that the biggest gains to be seen in 2.81 are with Intel’s Open Image Denoise for denoising (surprised?), and NVIDIA’s RTX and OptiX ray tracing engine for speeding up renders. We took a look NVIDIA’s take in-depth back in September, and saw ridiculous (in a good way) performance changes. As we integrate 2.81 into our testing, we’ll also look to expand to include RTX benchmarks in the relevant content (such as an upcoming ProViz performance look).

Blender-2.81-Alpha-GPU-Tests-Classroom-CUDA-vs-OptiX-680x386.jpg

Here’s a performance example from our look at the alpha build of 2.81.

It’s easily seen here that RTX and OptiX can greatly improve rendering performance, something that wouldn’t just impact final renders, but also the viewport render mode. As we’ve found in the past, NVIDIA’s Turing architecture also powers the viewport quite well.

https://techgage.com/news/blender-2-81-releases-with-nvidia-rtx-and-intel-open-image-denoise-support/
 
In the professional visualization segment, any card is no match for an RTX Quadro, there are dozens of applications that RTX accelerated right now. A Quadro RTX is the standard to get now.
Do you have any list of ME CAD software packages (such as NX, CATIA, Creo, Inventor, Solidworks, Bricscad, SolidEdge...) that utilize RT cores in realistic view (not scene rendering)?
 
Do you have any list of ME CAD software packages (such as NX, CATIA, Creo, Inventor, Solidworks, Bricscad, SolidEdge...) that utilize RT cores in realistic view (not scene rendering)?
Not sure what you mean by "realistic view"? Do you have a specific software application in mind?
Some CAD software as AutoDesk Flame use RT cores to accelerate ML/AI algorithms.
 
I think he means the final render with all the effects applied, rather than a quick render you'd use while working on your scene
 
No, I mean quick render while working on model :) In mechanical design we don't use realistic models (for obvious reasons), but shaded with edges (99% of time). Though when design is completed, it is good to have quick realistic view, usually for documentation or quick sales support (there is no time for hours/days long renderings on our servers). E.g. Inventor has Realistic View, with Ray Tracing, but it is CPU only. It also has Studio environment for scene rendering, also CPU only (which sucks, but that's another story). Others have better scene rendering packages, with GPU support, but that is not what I would like (though it's good)

So we need something like this done on GPU

But as far as I know, there are no CAD packages who will utilize RT cores for such processes. I think they all still use CPU, except Solidworks (but only for Quadro/Pro cards, though there are registry tweaks for GTX). So I hope they'll implement it in Solidworks at least (most popular CAD software, like 50% marketshare)
 
Status
Not open for further replies.
Back
Top