E-Cycles - The fastest render engine for Blender. 3.2 release available now!

suggestions are welcome for naming. You can make a poll with several ideas and let user decide. I’m too much inside the project to have an objective idea here.
But it’s strange the March update is faster than the April one. The April one has a new option that allows for more speed, but the out of the box speed should be the same. If you can provide a file to reproduce, that would help. I monitor BMW, Classroom, Blenderman and Barcelona on 1080Ti and 2080Ti and have seen stable times there.

The new builds for Windows are up, with the corrections from @mib2berlin too.
To ensure everything works as expected out of the box, I made a shortcut to launch E-cycles with the new filmic log encoding version. So starting the blender exe will give the old filmic like in the buildbots. When it’s well tested, I’ll invert behaviours and make a shortcut to start with old color management.
Linux and Mac are on the way.

1 Like

well to be fair i just want to know on which version of blender im working on, either see that in system pref or splashscreen, or somewhere obvious would do it ? maybe i miss it ?

update the numbers with solo card. i give you the file per mail

You mean Blender 2.8 or 2.7? The UI shows it ? Or do you mean the minor revisions of 2.8? If so, it’s written at the bottom right of the UI like 2.80.53

All builds are now updated with all the latest fixes. To start with the new filmic log profile, on windows use blender_filmic.cmd and on Linux/OSx you can use blender_filmic.sh. It’s still release candidate level, if you find a bug, you are welcome to email or PM me.

okay nice to know, and how do i do the link between 2.8.53 and the release date ?

For the release date, I could write both E-Cycles and Blender’s one, but that would give a very long name. But to concentrate on bigger improvements, I was thinking about only doing bugfixes during the month. So I could take Blender’s release date simply for E-Cycles?

hmmmm

What about

« E-cycles + blender Version## + Official master release date + e cycles version## »

« E-Cycles_2.8_2019-04-10_v1 »
Then you do some bug fixes
« E-Cycles_2.8_2019-04-10_v2 »
Then more changes
« E-Cycles_2.8_2019-04-10_v3 »
Then you update the official build, going back to v1 ?
« E-Cycles_2.8_2019-04-15_v1 »

Maybe the release date is not necessary at all if we have the 2.8.## in the name ? It’s really more precise than just a simple date , and it’s shorter

« E-Cycles_2.8.55_v3 »

In This case the problem is that on the official download page we don’t know what the 2.8.## is , we just have the date, to know the ## we need to actually download the build. I don’t know if this is a problem or not, need more feedback from e-cycles users

I think date is better. The 55 in 2.8.55 is the api revision iirc. So it may stay the same over weeks, giving the impression it is the same as before. The buildbots have commit hashes, which gives also little information about how old it is ( unless you search for that commit in the git log to find the matching date).

So I’ll keep your first suggestion, date from blender code, then v1, v2, etc. if needed. Is that ok? I think the most important is that I only offer one version, which will be the case tomorrow when the beta is over. I already unified multi and single GPU and I just fixed EKlein’s bug, so the April update can be called stable and I’ll remove the march update. If someone need it, I can still add it back for a short time.

2 Likes

The 2.8x windows build with the bugfix are available (2.7x was not affected). The Linux and Mac builds will follow. Then I’ll update the doc and make a new video tutorial to show and explain the new features of the April update: filmic log encoding, the new parameter and the improvements of the AI denoiser.

Dear Mathieu,

Just purchased 2.79 version. I’m wondering if BPT improvement is already on the stable version or not?

Dear Anhdo,
Normally yes, but I had so much to do the last days, that I couldn’t extensively test it, so I didn’t made any official announcement. If you can test it, it’s welcome. It will surely need some fine tuning, but the in terms of line of code, the biggest part is done. I’ll try to tweak the heuristic depending on feedback.
This video for example was rendered with BPT and E-Cycles was above 2x faster than master to render it.

Of course, it will depend on the scene, The more feedback I get, the better I can tweak it, so your times are welcome :slight_smile:

1 Like

Hello, Mathieu, I know it’s a bit early, but anyway - are you considering to include the new adaptive sampling patch by stefan werner once it is more mature?

1 Like

Hello lsscpp,
I test the patch since some weeks now, but I need more thorough testing before including it. When the April update is polished with all the docs and tutorials updated, I’ll have time again for further development and I’ll keep you informed. I will also speak with @skw before deciding.

When can it be in the Master? What is it it’s status currently?

Sorry I dunno. My most optimistic guess is Blender 2.81, which I don’t know when will be out anyway.

That this happens does not only depend on bliblubli releasing the code. It also depends on the code being approved by Cycles developers. They are very reluctant to incorporate non-stable code, hacks, not stable for animation, etc. I’m not saying that this is the case with this code, but saying that for it to be incorporated into a master is not just a bliblubli decision.
Edit:
Ok, I misinterpreted. I thought you were talking about E-Cycles code.

2 Likes

Just doing the classic “How low can we set samples and still have the denoiser do its job” and some of the painterly strokes it produced got me thinking - could the ai denoiser have presets in various artistic styles?

1 Like

There is no such thing as presets for neural networks (except if it was developed for that from the very beginning). You would need to train or fine tune a neural network for each style. This might be suitable if you are working on a feature film where you could fine tune the denoiser for your film or even for certain scenes.
However, as the training pipeline (and dataset) of the Intel denoiser are not publicly available, that’s unfortunately not an option.

2 Likes

Thanks for answering

A-Ha! So you really are an insider! Got you!! :wink:

2 Likes