Hi ace
In the thread you find a file
The metal under the plastic is very grainy and the Glas refection on the plastic handle reflects more noise too
as said trying to understand how the new system is working and supposed to be used.
Some said compositor denoise will yield better results too.
Adaptive sampling is still in, and itâs activated the moment you check âNoise Thresholdâ .
Also, threshold 0 means adaptive never kicks in, 1 is lots of noise.
.1 is quite noisy.
I think recommended value for final render is around 0.01, that leaves low residual noise, to be tackled by your denoiser of choice.
To be noted that the UI went through some redesign. This is because the future intended ideal workflow wants to be: âJust choose how much noise you want, and let Cycles do the restâ.
Optionally we can also set a rendertime limit
Where I am confused is then how noise threshold and max min sampling influences each other
The renderer will run for at least âminimumâ number of samples if not 0. Then, rendering can stop either by
a) Max samples reached
b) Time Limit, or
c) Noise amount less than threshold
whichever is first (if given and not 0)
Ah
So if I set max 10.000 with a noise threshold it will stop rendering at one point not when 10.000 is reached
Due logical - should have realized that earlier.
That is actually - Iâm convinced - what is the final plan of developers.
The future workflow where you just set the final quality (threshold) AND/OR a time limit. Never care about anything else. It would be the utlimate goal of the âremove buttons & tweaksâ crusade.
Thats actually the workflow im using with 3.0 im keeping the default 4k samples and 0.01 threshold and i only change the time limit, 10-20sec for simple scene and up to 1min for the complexe one, using a 3060ti
Ah that is super good to know !
This makes it actually super easy for beginners and very easy to teach to students !
Thank you for sharing and confirming !
The latest meeting notes for the rendering team
Attendees
- Brecht Van Lommel (Blender)
- William Leeson (Blender)
- KĂ©vin Dietrich (Blender)
- Brian Savery (AMD)
- Christophe Hery (Facebook)
- Feng Xie (Facebook)
- Patrick Mours (NVIDIA)
- Stefan Werner (Tangent)
- Michael Jones (Apple)
Notes
Cycles
- bake transparent shadows for hair is committed and provides a good speed increase. (Brecht)
- Rendering optimisations were commited for Shadow path compaction and Decouple shadow paths (Brecht)
- Committed Additive AO (Brecht)
- Restored render passes Depth denoising No AA on position pass (Brecht)
- Refactored cycles code organisation
- Resume work on pause/resume rendering and reducing memory usage (Brecht)
- Still working on Ray offsetting removal the results look promising but it affect to many things to get it in for 3.0 (William)
- Sobol Distance scrambling by Mathiue Menuet is committed (William)
- PMJ distance scrambling is still being worked on (William)
- HIP integration Work continues on AMD HIP integration. Currently to use this as yet unreleased drivers are needed on both Windows and Linux. Looking into what cards are supported. (Brian)
- Plan to submit MNEE manifold next event estimation code for 3.1 (Christophe)
- Working on the Metal Backend first fix kernel side then host side. Hopes to have this ready for 3.1 (Michael)
- Optix denoising issue caused by driver bug, hopefully fixed for 3.0 (Patrick)
- Working on patch for animation denoising needs rewrite for EXR (Patrick)
- Light groups work continues and is near completion (Brecht)
- Point Cloud rendering primitive almost ready for 3.1 (Brecht)
- Plan to submit a patch (based off Blender 2.9) for unflattening the vertices for ray intersection (Feng)
- Metal implementation currently used BVH2 on AMD HW and Metal ART on Apple silicon (Micheal)
EEVEE
- Attribute Rendering Committed by KĂ©vin Dietrich (Brecht)
Practical Info
This is a weekly video chat meeting for planning and discussion of Blender rendering development. Any contributor (developer, UI/UX designer, writer, âŠ) working on rendering in Blender is welcome to join and add proposed items to the agenda.
For users and other interested parties, we ask to read the meeting notes instead so that the meeting can remain focused.
- Agenda
- Google Meet
- Time: Tuesday 5 to 6 PM Amsterdam Time
- Next meeting: November 9, 2021
In shortâŠ
Also to note, William has just committed the scrambling patch for PMJ. This is perhaps turning out to be the most exciting year for Cycles since its inception.
Can anyone shed light on what this is and what itâs useful for?
Pretty solid update. We should see more of these as new Cycles is stabilized. One of the (more overlooked) goals of Cycles X was to simplify codebase and make new features easier to implement. The time for it is coming now with Blender 3.0 release.
No news of the texture cache patch ? There was a build available in may, nothing since
can someone explain what ray offsetting is?
When a ray bounces off an object, you have to âoffsetâ the start of the next ray bounce a little bit off the hit point so it doesnât self intersect.
This is normally done at a very simple level by adding a tiny 0.000001 etc floating point value to the ray start position. For too technical reasons, the coordinate space the 0.0000001 is added in matters.
thanks! i see. so this ray offsetting has disadvantages because of floating point precision issues? and a solution that works without it (with rejection ids) is better?
Are there any plans to bundle maketx.exe with blender ?
Are there maybe even plans to automate this workflow ?
I canât seem to get my renders to take LONGER than 2 minutes. Iâd like to let it go for like 4 hours and see how much better it is than 2 minutes but it always finishes at around 2 minutes.
edit : nevermind, I just realized the time limit is in seconds and not minutes. The should really say that in the tool tip.
you need to download the latest build
