What GPU for large exteriors?

[MANUEL_MOUSIOL]

Interesting comparison. You can see in the shadows though that some different contrast is applied. At least the huge Shadow in the back on the ground has to much so that we can really compare. I wonder if it would look similar if three tonemapping and contrast would match. Hard to tell...

[joconnell]

It's a fair point on the texture blurring and might be good to swap some defaults - we normally drop to 0.5 blur on all of ours. I'm not sure what the standard corona render settings are but it definitely looks like it's adding in a bit of a contrast curve - if you have something like nuke you could put the two images above through it's colormatch tool and it'll generate the difference between the two as a colour lut - you can load this in the vray buffer. Personally I'm in vfx and every camera has a slightly different curve to it so we always take the curve off the image to make it flat / linear, render our vfx to this and then reapply the camera curve to both afterwards. A big part of photorealism (aka making something look like a still off an slr / movie camera) is that one of those camera's job isn't to record colour and brightness accurately, it's to make a pretty picture. To do that they take the image data and accentuate certain parts of the result to make it more pleasant - if you take the same picture on a nikon, canon and sony you'll get three different images as all three manufacturers are going to choose different parts to feature. Vray doesn't do any of this so it can tend to look a bit flatter than some other renderers which add a little bit of a default look, some artists might really like that as they want to get a finished image right from the frame buffer, some others might want to have the purest data they can get as they'll be doing a lot with it after!

[Donfarese]

It's a fair point on the texture blurring and might be good to swap some defaults - we normally drop to 0.5 blur on all of ours. I'm not sure what the standard corona render settings are but it definitely looks like it's adding in a bit of a contrast curve - if you have something like nuke you could put the two images above through it's colormatch tool and it'll generate the difference between the two as a colour lut - you can load this in the vray buffer. Personally I'm in vfx and every camera has a slightly different curve to it so we always take the curve off the image to make it flat / linear, render our vfx to this and then reapply the camera curve to both afterwards. A big part of photorealism (aka making something look like a still off an slr / movie camera) is that one of those camera's job isn't to record colour and brightness accurately, it's to make a pretty picture. To do that they take the image data and accentuate certain parts of the result to make it more pleasant - if you take the same picture on a nikon, canon and sony you'll get three different images as all three manufacturers are going to choose different parts to feature. Vray doesn't do any of this so it can tend to look a bit flatter than some other renderers which add a little bit of a default look, some artists might really like that as they want to get a finished image right from the frame buffer, some others might want to have the purest data they can get as they'll be doing a lot with it after!

Yeah the blur sucks especially when you have loads of textures to correct. Sorry I do not have nuke nor use nuke. The curve or whatever you want to call it does add a lot more detail and depth to the render, I found this in many test renders I did. Corona and Fstorm both have a great curve for realism. I'm surprised that a curve can pull so much highlight into the trunk of the tree and the branches, if that's whats doing it. I would love to see a curve matched to coronas in Vray, so far I cannot reproduce.

[danio]

Is the blur really greater in VRay than Corona by default? CoronaBitmap and also max bitmap always defaults to 1.0. Regardless of render engine I often drop that to .5 or .2 (or .01 for opacity maps.) Is there something else forcing greater levels of blur in VRay by default?

[Donfarese]

Is the blur really greater in VRay than Corona by default? CoronaBitmap and also max bitmap always defaults to 1.0. Regardless of render engine I often drop that to .5 or .2 (or .01 for opacity maps.) Is there something else forcing greater levels of blur in VRay by default?

No both Corona and Vray blur is set at 1.0, but Vray for some reason blurs the texture a lot, especially at angles.

[danio]

No both Corona and Vray blur is set at 1.0, but Vray for some reason blurs the texture a lot, especially at angles.

I'd be curious if there is any way to modify this or is it at the engine level?

[Donfarese]

I'd be curious if there is any way to modify this or is it at the engine level?

To me this is something that needs to be fixed by devs. So 1.0 blur as default doesn't ruin textures.

[Morbid Angel]

The brightness in the corona / vray grey renders is due to color mapping difference, I'm almost certain of that. For the texture filtering I think corona just overwrites the bitmap blur value with its own, or a texture filter all together, if you use vray hdr loader you will get better results. I'm positive that corona / vray renders look exactly same in the same lighting / color space conditions.

[Donfarese]

The brightness in the corona / vray grey renders is due to color mapping difference, I'm almost certain of that. For the texture filtering I think corona just overwrites the bitmap blur value with its own, or a texture filter all together, if you use vray hdr loader you will get better results. I'm positive that corona / vray renders look exactly same in the same lighting / color space conditions.

Would like to know how we can mimic Coronas then. Cause that extra depth and detail make all the difference if that is what it is.

[MANUEL_MOUSIOL]

Actually in another thread this was discussed as well and it was also mentioned that Corona doesn't bounce rays as many times as vray, which gives a greater contrast, so to speak.
gi distribution should be the same though.
It would be great if the Corona guys and vray guys would finally get together and compare the coding and give us some insights.

[Donfarese]

Actually in another thread this was discussed as well and it was also mentioned that Corona doesn't bounce rays as many times as vray, which gives a greater contrast, so to speak.
gi distribution should be the same though.
It would be great if the Corona guys and vray guys would finally get together and compare the coding and give us some insights.

+1 to that

[Donfarese]

What do you think Vlado? I'm interested in knowing if we can get a better curve like Corona? If that is the reason.

[squintnic]

cant you do that in post or frame buffer?

[hoppergrass]

With regards to the number of GI bounces, many years ago and with an entirly different renderer (and gi calculating methodes) the developers were waxing lyrical about how great and accrate the GI results are because the calculating termination point only happened at 100 bounces.

I did a few tests and found that after 5 bounces there was little to no visual difference to that at 100 bounces and rendered significantly quicker.

I have done a few similar tests, over the years and different versions of Vray, with Vray as well by lowering the number of bounces to see what the differences were. Pretty much came down to the quality and detail in the GI solution dropped significantly but the overall contrast didn't change that much until the numbers came rediculously low (like 1 or 2 bounces). Which leads me to believe that its a lot more complicated than just the number of bounces being clamped to be the reason for the differences between Vray and Corona.

PS my test were fairly informal and not scientificaly conducted with detail analysis, but done more for fun and my own education

[Nejc Kilar]

With regards to the number of GI bounces, many years ago and with an entirly different renderer (and gi calculating methodes) the developers were waxing lyrical about how great and accrate the GI results are because the calculating termination point only happened at 100 bounces.

I did a few tests and found that after 5 bounces there was little to no visual difference to that at 100 bounces and rendered significantly quicker.

I have done a few similar tests, over the years and different versions of Vray, with Vray as well by lowering the number of bounces to see what the differences were. Pretty much came down to the quality and detail in the GI solution dropped significantly but the overall contrast didn't change that much until the numbers came rediculously low (like 1 or 2 bounces). Which leads me to believe that its a lot more complicated than just the number of bounces being clamped to be the reason for the differences between Vray and Corona.

PS my test were fairly informal and not scientificaly conducted with detail analysis, but done more for fun and my own education

Oddly enough, I keep getting the impression that a "pure" PT / PT renderer like Octane gets a lot more light in an interior than say Corona with its UHD cache. Obviously with Octane you need to keep the GI Clamping at +50 (or you know, the default 100000000) but even though it does only 16 bounces compared to Corona's UHD cache I'm kind of under the impression that it gets more light in. Its important to point out that Corona does clamp the GI at 25 by default if I'm not mistaken.

So I'm no rocket scientist but I'm guessing if you clamp the GI to a lower value you obviously can't get a ton of lighting info in there even if you do 100000 bounces. If thats true then I guess its pretty useful to have a balanced GI clamp / GI bouncing solution.

I know the above has very little to do with V-Ray but I presume similar principles apply.