Grain Grain Grain

[yeeha68]

We are attempting to switch to 1.47 from 1.09 however a few problems still trouble me. I have read as many posts as I can and still see no clear answer. The speed is the first hurdle...as I have come close to the same speed, the images are not of the same quality.

What happened to Simple two-level? Adaptive QMC isnt even close. The following test images are setup in both versions with comparative settings.

1.09 with Simple two-level and a vraylight, subdivs=10

1.47 with Adaptive QMC and vraylight subdivs=10


Notice how bad the line details from QMC noise are in 1.47...
1.09 with Simple two-level and vraylight subdivs=10

1.47 with Adaptive QMC and vraylight subdivs=10


Now, the grain on the walls can be solved by raising the light subdivs to about 30. Thus doubling the render time. The grain noise on the line details can be solved by raising the QMC min and max as well as lowering the noise thrhd, pushing the rendering time another 2x.

None of which really works for animation where time is everything. Can anyone give me some suggestions based on these simple settings. The object here is to keep rendertimes within the same zone as 1.09.

1.47 settings (comparative to the 1.09 settings)

[DaForce]

One thing i notice in the first renders, is around the light recesses on the 1.47.02 render there is some noticeable black squares, as if its not being smoothed. Looks like it could be co-planar faces? Do you know if this could be the case?
Also try turning off the IR map blur in 1.47 might help, or try turning it up.

[percydaman]

default QMC settings are not good with adaptive QMC antialiasing. There was a thread earlier about this.

Try setting your adaptive amount to .7 and your noise threshold to .002 and it should look alot better.

[vlado]

Here are some test I made with 1.09.03 and 1.47.03; I tried to get similar render quality and time for the same scene:





Here is also a rendering with the light cache as secondary GI engine in 1.47.03:


Here are also the two scenes, if you want to take a look:
http://www.vrayrender.com/stuff/spee...speed_comp.rar

Best regards,
Vlado

[DaForce]

Mmmm was that last link meant to be a link to the last picture, or to a zip file to download the test scene?

Cause it aint right

[vlado]

Sorry, it's corrected now.

Best regards,
Vlado

[SylvainJ]

Got a simple question,

Here, we are still running the 1.09.03r because we have a render farm of over 100 computers. We did not upgrade to 1.47 because this take a lot of time and we cannot afford a last minute problem with a beta release. So, can someone (Vlado... ) tell me when we will have a final version that we can use normally.

I've tried the 1.47.02 version and I really love the Light cache method.

Thanks

[yeeha68]

One thing i notice in the first renders, is around the light recesses on the 1.47.02 render there is some noticeable black squares, as if its not being smoothed. Looks like it could be co-planar faces? Do you know if this could be the case?

I noticed that too...they are not co-planar, and work fine in 1.09, but not in 1.47

Try setting your adaptive amount to .7 and your noise threshold to .002 and it should look alot better.

Here it is with those settings. Note how much the time increased and how it is still noisy.

[percydaman]

the longer you wait, the bigger a change its going to be. I would switch over the first possible moment that wouldn't impact your studios workflow.

The beta rocks over the older version, and it aint just the lightcache, which I love. Our studio makes uses vray proxies ALOT.

[flipside]

Re: Grain Grain Grain

We are attempting to switch to 1.47 from 1.09 however a few problems still trouble me. I have read as many posts as I can and still see no clear answer. The speed is the first hurdle...as I have come close to the same speed, the images are not of the same quality.

What happened to Simple two-level? Adaptive QMC isnt even close. The following test images are setup in both versions with comparative settings.

Notice how bad the line details from QMC noise are in 1.47...

Now, the grain on the walls can be solved by raising the light subdivs to about 30. Thus doubling the render time. The grain noise on the line details can be solved by raising the QMC min and max as well as lowering the noise thrhd, pushing the rendering time another 2x.

None of which really works for animation where time is everything. Can anyone give me some suggestions based on these simple settings. The object here is to keep rendertimes within the same zone as 1.09.

Hi,

When using adaptive QMC AA, you should use other values for the QMC sampler, but leave the min max at default values.

I usually use adaptive QMC 1/4 and the noise th in QMC sampler=0.001 (or 0.002). (don't use 2/4 and 0.001 because that will only render longer without much visible result)

Note that when you lower the noise th, your IR map calculation will also take longer. The new builds take more samples by default than the old one, so you can get away with coarser values for it to compensate. For example hsph can be lowered, and Clr & Nrm can be higher (altough they are not too low in your settings).

I would definatly try Lightcache to speed things up.

Hope this helps you a bit...

wouter

[yeeha68]

Thanks, Ill give that a try.
If anyone wants to experiment with this I have posted the files here
http://www.bluelimestudio.com/vray/n...t/compvray.zip

[flipside]

You don't have a 3ds version maybe (with boxes or something instead of lights)

[yeeha68]

You don't have a 3ds version maybe (with boxes or something instead of lights)

No, sorry, I was trying to use real details without getting too detailed for a more accurate result in time and quaility.
However, that being said, I am now experimenting with Vlados file...

[vlado]

I've been playing a bit with your scene; as much as I hate to say it, there is no way to get the same noise level for the same speed. The same quality result took about 4 mins (where the 1.09.03 version took 2m 40s). The main cause for the difference is because of the way the image sampler interacts with the area light samples.

However, when the scene starts to get more complicated (higher irradiance map settings, some texture maps, correct 1.0 secondary GI multiplier, better AA), the new version gets faster than the old one. Using the light cache also gives the new version an advantage.

Best regards,
Vlado

[ngrava]

Have you tried the good old Adaptive subdivision AA? I've found that QMC is just noisy by nature. It is Quasi Monti Carlo after all. It's supposed to simulate random jittered samples. Noise is the out come in a lot of cases. Adaptive Subdivision is just super sampling. Vlado, wouldn't setting the Subdiv's to 1 and 1 be the equivalent of the two level sampler? Or am I misunderstanding something?

-=GB=-