As E-Cycles and buildbots both got faster here are updated charts, now including the RTX cards with the 2080Ti:
4 Likes
Where can i find a release log of the last releases? I download a new version but have no idea what’s new.
If you will use Extra passes and hdr training with D-Noise it will be a different story. OIDN always use albedo and normals passes for denoising.
I always post changelog in the thread (here is the last one), but it quickly gets lost in the huge thread. So I’ll add a text file with changelog to the product page starting with the next release.
Apart from all the changes made in upstream Blender’s code, the AI denoiser addon got a much better usability and robustness:
- it adds a new mix node to be able to tweak the quantity of grain for your artistic needs.
- it works in many more cases with existing compositing node trees, file output beside composite output and many other setups.
- it will generate the node tree for the active render layer node if any, so that you can quickly denoise the layer view you want.
2 Likes
Great! It will help a lot!
Does E cycles denoiser work on individual render passes? Right now built-denoiser only works on beauty but if I want to composite render passes… Not so lucky
Yes it can denoise whatever you want, even textures or photographs. For this, you can manually add a filter -> denoise node in the compositor and feed it with your render pass or whatever you want to denoise in image and give the corresponding albedo and normal pass (optional, but gives better results).
want to say, last build of 26march for my two 2080ti is a bit better, from 8:06 to now 6:30
better ! but the beta build is still faster on 3:26
Yes the last weekly update introduced a speed regression which is fixed in the build from 03 26. But the speed regression introduced by animation denoising is only fixed in the beta version. It’s a huge change in the code, so there are 2 possibilities: It gets fixed in master or I do an update for the final 2.80 release. Note that you can use the latest AI denoiser addon on the beta version. Just replace the .py file with the new one in your addon folder and you will get the improvements in the beta.
Edit: wait, the single GPU performance should be unaffected. Are you sure about the Single 2080Ti times?
i am, those are test on old versions, if you want i can do benchmarks with versions of your choices
Ah ok, it’s indeed with the speed regression version. With the latest 20190326 would be great.
There is a new test version of the AI denoiser addon available at the bottom of the download page for those wanting to test. It has a new button to quickly remove the denoising nodes. It also restores the old connections when possible.
Very handy when you want to change the quality level of the denoising node tree for example.
Hi mathieu
im in a quest of finding the best way to work in huge scene in blender. so i tried the octane plugin on a 30mil poly+ scene with instancing, particle system ect… im not benchmarking anything, its purely based on the viewport interaction.
here is what i found :
-cycles take twice as much time to get the rendered view ready with pre-rendering, blender even could freeze at this moment for few moments.
-the quality of noise was the same (exterior scene) and the rendered view was quite clear for both at the same level, octane was let say twice as fast tho.
-in term of camera movement and rotating the view in the rendered mode, octane was ahead of cycles, but not that drastically
-in term of moving a huge object of 5 000 000 polys, octane was WAY above blender, cycles have this weird habit of re-loading the asset after moving it, so it could take 5second before having any sign of life of what your translation actually look like.
do you think we could see an amelioration on all thoses default above in the future of cycles / e-cycles ?
Are you using the Demo version of Octane?
yes, octane 4, not 2018 or 2019, not avaible yet in demo.
Everything is possible, but first I’ll concentrate on RTX. As far as I know there are ideas to better integrate Embree in Cycles, which should make BVH computation much faster. It’s in the GSoC Ideas, so maybe one student will pick it this year.
1 Like
Np, I have a lot of customers who are Octane user returning to E-Cycles because of the better integration with Blender. So I’m ok if people try it 
Most of my customers are architects or product advert people and the option in 2.7x of E-Cycles brings the preprocessing to 0 for their fly-through, so nothing can beat it 
2 Likes
For those wanting to experiment, their is a version Release_multi_gpu_fast_AIdenoising_3 that makes multi-gpu rendering much faster when using the AI denoiser. I also added some optimisations in it as it’s a beta anyway. So even if you have a single card, you could see benefit using it. You’re welcome to post your render times with it to see if it works properly and should be integrated in the next feature update.
This build also include all the latest improvement to the AI denoiser addon.
want to say, last build of 26march for my two 2080ti is a bit better, from 8:06 to now 6:30
better ! but the beta build is still faster on 3:26
dropped to 3:30 and with an amazing denoising !
amazing work ! cant wait for you future releases
about that, if you are working on rtx integration, isnt that counter productive if another official blender dev is working on it ?