E-Cycles 2020 is coming! Sales extended to 4th of January.
E-Cycles was improved a lot in 2019. The 2020 version will have all this improvements plus a complete rewrite of my code for even faster rendering. The first test build of E-Cycles RTX 2020 is now available and first tests show up to 4x faster rendering out-of-the-box.
Figure 1: Render by JLE Studios - New Zealand and made with E-Cycles. Original Tweet.
You can get E-Cycles with 50% off until the 28th of December with the code „holidays“.
E-Cycles 2.82 builds with Mantaflow, UDIM, USD export, random UVs per island, Light Groups and much more are already available!
Figure 2: Render Time comparison by user Hyesung using E-Cycles RTX 2020
New features and improvements since the first E-Cycles release:
Planned were 12 improvements for 2019, as of now, 30 were made. On top of the weekly builds, here is the changelog since January until now:
- up to 5% faster rendering in some scenes.
- new Dithered Sobol sampler for cleaner render at low samples (original Patch by Lukas Stockner).
- Cleaner Viewport renders in complex scenes.
Figure 3: Viewport comparison by user Stak between Blender on the left and E-Cycles on the right. Original Tweet.
- Intel Open Image Denoise Patch from Stefan Werner is included.
- Latest Filmic log encoding for top-notch post-pro as recommended by Bartek Skorupa https://www.youtube.com/watch?v=kVKnhJN-BrQ.
- New option based on the scrambling distance patch from Lukas Stockner, allows to render up to 40% faster (on top of existing speedup).
- Uses half the memory when denoising with all passes (transmission and SSS included).
- New AI denoiser addon which can now generate denoising node trees in one click, even in complex scenarios and with very high quality.
Figure 4: Comparison at 16 sample per pixel of the "denoise" node in Blender (long manual node setup with recommended settings, still blurry) and the AI denoiser default results in E-Cycles (added in one click).
- E-Cycles 2.8x got 2 improvements ported from the 2.79x branch. Memory usage is reduced in most scenes, allowing to fit bigger scenes in the GPU memory. Rendering is also around 10% faster compared to the April update.
- Both the 2.8x and 2.79x versions can render up to 2x faster when using the old denoiser (the one from Blender) on multi-GPU setups.
- Updated to latest Open Image Denoise, using 38% less memory to denoise.
- E-Cycles will automatically choose memory efficient code path depending on the scene, using up to 40% less memory than master/buildbots (BMW scene on a 2080Ti goes from 2GB on buildbots to 1.2GB memory usage in E-Cycles).
- An option to force one memory efficient option has also been added. It makes transparent shadows optional. Render times can be greatly reduced in some scenes.
- The AI denoiser tree generation has been improved and has a better auto layout now.
- Adaptive sampling is now available for CUDA and can bring some good speedup in selected scenarios. Own test on smoke+fire shows up to 2.5x faster rendering compared to the June update. (Although only CUDA is officially supported, OpenCL users also report noticeable speed-up)
- Based on the official 2.80 version.
- AI denoiser updated to 1.0. It allows to denoise renders of any size with low memory usage.
- New buttons for quick background command line render. You can then close the UI, which gives better stability, makes more memory available for your scene and also speed-up render in many case.
- Baking render speed was optimised in some scenarios.
- Completely reworked AI denoiser:
- faster, particularly on multi-GPU setups
- much better memory management thanks to a patch by Stefan Werner
- streamlined denoiser UI
- automatic optimizations of the denoising tree. Improves denoising and rendering speed.
- Faster baking:
- Baking is now up to 20x faster by taking smart decision for optimal performance.
- Baking will require a bit less memory too.
- Added a remove double modifier.
- Added an experimental bridge modifier.
- Added bevel and crease weight to the boolean modifier for non-destructive and easy to use bevel after boolean.
- Nebulae Build as an example of targeted optimizations potential.
Figure 5: Presentation during Gleb Alexandrov's talk at the Blender Conference 2019
- out of core Rendering for RTX (speed limited by system Ram, may require Quadros or Teslas for good performance)
- much faster rendering for RTX
- further optimizations for CUDA.
Figure 6: Render Time comparison by user Hyesung using E-Cycles 2019 november update. Original tweet.
- faster pre-processing in some scenes
- light groups patch added.
Figure7: Render of Sintel made by Ed O'Conell with E-Cycles and Grasswald. Memory optimisations allow to use higher subdivision levels and more particles. The AI denoiser allows to use low samples count even with strong bokeh.
You can get all those improvements plus the upcoming one including AI technologies I’m working on with 50% off this week using the code “holidays” during checkout.