Part of me wants to be selfish and wish that you and your team puts some work into OpenCL 
Still, what ever you do it sure moves Blender even further in the right direction. Keep it up
2 Likes
Some Blendernation readers still need time to get things working with Gumroad and Paypal it seems.
So I let the coupons there active for a bit more time.
1 Like
Image quality problems with glossy reflection with mutli-GPU and beta builds. Single GPU and older builds okay. Using the classroom scene with default settings look at the two closest desks glossy reflection. See below:
Single GPU (3/29):
Multi GPU and April Beta:
Another issue is on multi-GPU like my 2-rtx 2070 using blender denoiser is about 10-25% faster by turning off auto-tiles and having a manual tile size 256.
1 Like
Hi,
thanks for the report, I’ll have a look into it.
There are about as much user wanting the samples to be in the preset as those wanting it separate. To have the best of both world, I could make the sample preset only on the predifined presets as percentage.
For example, using scrambling distance or AO bounces greatly reduce noise in noisy areas, making the noise more even. So when selecting a preset that activates one of those, I could set the spp to 70% of current user-defined one for example, to keep the same noise level.
We are still in beta, as I still have to fix EKlein’s found regression in the betas. so I think it’s ok to try that before freezing it for final release. It would greatly improve first time experience. For experienced users, in worst case, it’s just one parameter to tweak, so still much faster than before the presets were added. I’ll provide builds with it tomorrow and let a week of test with it and then gather feedback per email.
About preset, there could be for types of scenes like « interior » « exterior » « volume » like vray have ? (Maybe organize this differently) Combining thoses per scene case preset + quality and speed preset could be interesting
You should really start testing your speedups on more variety of scenes, and especially scenes with proper true lighting, not evermotion scenes which are filled with ambient constant light and black, dirty ambient occlusion in corners. Those scenes always run fast… I’d still like to see a proof that E-Cycles improvements work in general scene, not on interiors with one very specific type of fake lighting setup.
Did you have a look to the blendernation article? There are several renders without AO there. I use the Evermotion scene because of legal reason. It’s the only one I’m authorized to show. You can also have a look to @BD3D 's renders.
You can look at my last Project :Boring Kitchen
6k samples, no AO, and speedup 2,5x, rendertime about 25 min on single 1080
1 Like
I also thought about adding Scene types for interior and exterior and product advert, as it covers 99% of the current E-Cycles users, even wanted to have it autodetected instead of manual. I’ll have a look to it again next week.
2 Likes
Thanks a lot for the good example. If you can write you rendered it with with E-Cycles under the render, that would be very appreciated
Yes, but there are no before/after examples. I have specifically made a scene, originally for Corona renderer, to test these kinds of scenarios. You can find it here: Cycles Performance
It will require a HDRI map, which I can not redistribute, however you can get it directly from here: http://illuminatedtools.com/Downloads/ext_LateAfternoon_Mountains_CSP.zip Just plug it in as an environment map and do not change rotation, and you should get an identical result. The scene itself is set up without any ambient fakes.
If you decide to test it, just please post a proper comparison. Meaning 2 images (regular Cycles vs E-Cycles) rendered with the exact same settings, exact same time limit, and both without denoisers. (We are comparing sampling performance here, not denoisers).
Best way to compare noise levels is to see two images rendered for the exact same amount of time on the exact same machine with the exact same settings. Anything else is just biasing the results.
I bought 2.8x e-cycles and now I am moving to the fully-featured 2.79 e-cycles build.
Mathieu worked right away to have all my issues answered almost instantly.
This scene is true light example. No AO.
Cycles:
4:06 pathtracing
E-Cycles:
2:25 pathtracing
1 Like
Thanks a lot for the feedback. I’m happy to hear it works good for you now. It seems you still have the slight bug EKlein mentioned in the beta as the lightning is a bit darker (it seems this bug reduces reflections/glossyness, still investigaring). The stable march version version work as expected, so a fix should come quickly.
Thanks, I’ll have a look when the april version is officially released, will help to check regressions. By the way, why not take a hdri haven one? You are allowed to distribute those with a simple notice it comes from hdri haven. That would help to have everyone use the same hdri.
I think it is not possible since with same settings e-cycles tend to be faster with same settings.
This is just what you are looking for:
1 Like
The one I linked is also free. It’s just a good HDRI, and since I’ve already done a lot of test renders with that one in this particular scene, then let’s just use this one for now. Let’s not make things unnecessarily complicated
Yes, it is very easily possible. You just open the scene with regular Cycles, do not change a thing, and let it render for let’s say 2 minutes. Then you open the same scene in E-Cycles build, do not change a thing, and let it render for the same 2 minutes. After that, you compare the two images. If E-Cycles is faster, then E-Cycles image will be visibly cleaner after 2 minutes 
That is the most reliable way of testing rendering performance - visual difference in noise after same rendering time. That’s what ultimately matters, when you send an image to a client. Simply how noisy it looks. All you want from a faster rendering is to either have less noisy image after same time, or have image of same visible noise level rendered faster
1 Like
Ok I got it.
But in the example above I can see clearly that in less time it have less noise.
And the file i opened and rendered with same settings.
6k samples ? are you crazy ?