Cycles Vray Comparision

Friend of mine, ([email protected]) has made this scene in Sketchup. Originally it was rendered with Vray using brute force with 16 subdiv and 2500 Light Caching samples with Fully Adaptive Subdivisions with min 1 - 120 samples. It took 16 hours to render on 2 Intel I7 machines (to yeras old). Obviously we tryed to render this scene using Irradiance map, but we had splotches problems in dark areas, under the bed. It was preety fast (more less 3H), but quality was too bad, that’s why we choose brute force and light cache.
As we are nowe trying to choose propoer workflow for us in monsunstudio, I have imported this scene to Blender and made simple material setup, to get more less same result.
There are some differences in the lightening, and I should do some twiking, but it still gives an good idea whats going on, how fast it is etc. It’s rendered in Cycles with 3000samples, rendered on Titan Black 9h 10min.


What do you think?

Attachments


Nice image :slight_smile:

The first (Vray) image just feels more real and pleasing.
Not that the Cycles one is bad, the Vray one just has that something extra.

Cheers

Yes You’re right. Main difference is in lightening. I’ve some changes in light setup, more like in Vray scene, tomorrow I’ll upload an upgrade.

The bottom image seems more saturated too… Or is that just me?

If you have all image textures uv-mapped, try Vray 3 RT-Gpu cuda render.
I would render you scene with irrandiance map + light cache.

The shadows seem better in the vray one but if the light sources aren’t consistent in both images then it is difficult to compare.

winner cycles ,with a litte bit color management.

It’s hard to compare but how many subivision in bf ?There is only a domelight+hdri image for lighting?In a scene with great windows i reach very clear render with bf 60 subd lc feault settings and dome lighr 30 subd and material reflection,for fix splotches in irradiance map you have to encrease hasph from 80 to 100 and interp samples from 20 to 50, but i think if you want obtain a really contrast scene you could set no mapping color linear color mapping and save in exr file,be sure the exr hdri is set to linear and set color correction from 0.8 to 0.7 (it depend from the image used),uncheck subpixel mapping an clamp, thick the botton rgb during the render(this not affect the saved image) but it isuseful as preview.Be sure also the gamma in color mapping is set to 2.2

Since you don’t provide your lighting setups there’s not much to say here.
We can’t really compare without more details.

My impression is that Vray looks a lot more realistic.

All the furnitures details are more clear in Vray version.

All the Vray materials look more realistic.

As mentioned before for this type of scene I would have gone with IM+LC and I feel that if you set it right it wouldn’t take more than two hours on a single machine.
(It’s a typical interior scene with extremely simple materials and geometry).

Also the cycles render suggests a complete lack of understanding of material setup and nodes. Look at the plant material it is totally washed out. Either you set a wrong diffuse material or an SSS material with the scale way too high. Also the lantern on the shelf has a different material… These important finesses are very important when doing scene comparisons. Which is why I always look at these comparison with scepticism. The OP usually has good experience with one render engine and little or none with the other. You really need to be a master of both to setup an unbiased comparison.

This affects constructive render engine comparisons. And I agree with the guy above, 19hrs for this scene on a titan. A freaking titan!

Something must be wrong here?!

It’s hard to do a proper comparative study on this.
Scene does not look good, neither designed, nor modeled, shaded & lit. Wonder how it is set for rendering? Those render times are things of past for what we’re looking at here.
Images are badly compressed.
Replication of materials, lighting and shading must also be made accordingly (exact or at least to same specs).
Maybe then we would be able to debate… now we can just share our subjective opinions.

Hi Again, I’ve made this test to measure time I need to render quite dark interior scene in sycles. Reason I did that was rather simple. We had problems in Vray with getting clean image using Irradiance Map + Light Cache, error pop up in highers resolution, normal 1920x1080 was nice and clean. But 4500x3000px gives as en strange behavior in shadows on reflective materials, especially under the bed. To solve this problem we’ve been trying to tweak Irradiance map settings, changing interpolations samples to 40, hsp to 100, increased global subdiv to 16 and sun and sky light to 12 if i remember correct. But ther error was still there. Using BrutForce as first and secound GI gave as good result with 32 subdives but took ages.
Then we used BF as first Gi and Light Cache as secound. BF was 12 or 16 subdiv, fully Adaptive Sampler was used with values 1 - 120. That did proper job in terms of Gi calculation. Vray render this scene for 16H on two machines, using Vray 2 for sketchup.
Design, texturing, modeling, light setup isn’n my…
Then I’ve imported scene to Blender, with cameras, lights, and textures. I’ve recreated light setup in similar way. There are additional planes as VrayRecLight, Sun where it was, and Hdri light as DomeLight. The major problem was with light strenght as it seems to give a bit different results.
I haven’t recreated all materials, like plants, fabrics (texture is mapped on simple diffuse material).
My point was, what time do I need, to get clean image at 4500x3000. I get more less clean image with 3000 samples and 9h of render time on single GPU. I know that materials are not perfect or even good, but we still can compare major one, wood, brick.
Anyway, I’ve made some changes in light setup, it’s rendering…

can you post me your settings? Entirely?

Nice comparison, thank you for that!
It seems to me that you are not using correct fresnel for the cycles materials - IMO comparing the vray render to cycles with diffuse shaders only can be a bit misleading because the glossiness usually gives the most noise and can slow down the render a bit. But with 3000 samples there is possibly not such a big difference in noise.
Anyway, looking forward to see your other renders.