DMC Sampler: Vray2 & Vray 1.5 Comparison Test Renders

[pixela]

I rendered the same scene with Vray 2 and Vray 1.5 to compare the DMC Sampler Vray versions and I came up with interesting results.
I had longer times with Vray 2.

For all renders:

Sub-pixel Mapping : Off
Clamp Output: Off

Light Cache : 1000 Subdivisions Sample Size 0.02

For Irradiance Map Renders: Same precalculated IM is used (calculated in Vray 2 - same map is used when testing Vray 1.5 too)
For BF renders: 40 SUbdivisions

[pixela]

[BBB3]

Puzzling indeed... How did you solve the noise problem in the end?

[instinct]

I am not sure if sharing the IM between versions is a good idea actually. Did you try with seperate IMs ?

Regards,
Thorsten

[RSI]

Also if you check on spot3d,

"This is corrected in V-Ray 2.0, but it means that older scenes might render slower than with V-Ray 1.5 because more image samples will be calculated. In most cases, this can be fixed by increasing the DMC color threshold, as it no longer needs to be very low to produce a clean image.



If for any reason you need to revert to the old (and wrong) sampling method, you can create an environment variable called VRAY_OLD_SAMPLING and give it a value of 1."

Did you change the variable?

[zeronove]

I totally agree with pixela!!!
Something dramatic has changed in DMC with 2.0 or 2.1 release and, for my workflow, it increased my rendertimes a lot!
Till now I had no time to deep testing it like pixela did but after some months I'm pretty sure that something doesn't work as expected (for me) anymore.
Now I'm doing some tests trying to increase subdivs values a lot more than usual, the idea is to let the secondary rays solve the glossiness instead of DMC sampler, then DMC should only solve the antialias; I'm not sure it makes sense but I'm trying

[BBB3]

it would be great to get some workflow tips from Chaos about how to go about using the modified sampler. we are fumbling in the dark a bit.

[Neilg]

In most cases, this can be fixed by increasing the DMC color threshold, as it no longer needs to be very low to produce a clean image.

Have you tried changing this.

[pixela]

Puzzling indeed... How did you solve the noise problem in the end?

I could solve the noise problem by making reflection subdivisions to 100.
I am not any expert on Vray Render Theory but according to my experience with my last scene, the Universal Settings is not working as the same way in Vray 1.5 and it was not enoguh to put DMC Image Sampling min:2 max:100 to kill the noise.

[pixela]

I am not sure if sharing the IM between versions is a good idea actually. Did you try with seperate IMs ?

Regards,
Thorsten

I used the same precalculated Irradiance Map because I wanted the renders to be only effected by the DMC sampler and thought this would be a good way.

[pixela]

Also if you check on spot3d,

"This is corrected in V-Ray 2.0, but it means that older scenes might render slower than with V-Ray 1.5 because more image samples will be calculated. In most cases, this can be fixed by increasing the DMC color threshold, as it no longer needs to be very low to produce a clean image.



If for any reason you need to revert to the old (and wrong) sampling method, you can create an environment variable called VRAY_OLD_SAMPLING and give it a value of 1."

Did you change the variable?

Yes I remember reading these but if you check my test renders, you will see inverse.
The SAME scene renders faster with Vray 1.5!

[zeronove]

maybe I misunderstood something, but you didn't say anything about the clr threshold value

on spot3d is said that with exactly the same value for min/max DMC and clr threshold, 2.0 could be slower than 1.5; that's exactly what your test are proving.

what I desume from that statement is that increasing clr threshold in 2.0 should be similar or equal to 1.5 regarding rendertimes; in my scenes it doesn't happen!

[pixela]

maybe I misunderstood something, but you didn't say anything about the clr threshold value

on spot3d is said that with exactly the same value for min/max DMC and clr threshold, 1.5 could be slower than 2.0; that's exactly what your test are proving.

what I desume from that statement is that increasing clr threshold in 2.0 should be similar or equal to 1.5 regarding rendertimes; in my scenes it doesn't happen!

No my tests are proving that Vray 1.5 is FASTER with exactly same color threshold value.
Color threshold is 0.005

No my tests are proving INVERSE of that statement. And though it is same in all renders, Vray 1.5 always renders faster.

Not in this particular test, but during this scene, I made big noise problem and I made many tests with different color thresholds, adaptive amounts, different DMC Sampler maximum values, switching between BF and IM.
First of all, making the color threshold higher was not enoguh to kill the noise and making it lower is too expensive in terms of time.
That statement didn't happen in my scene.

Edit:
Allesandro,
I think I understand now what you try to say... But if I increase color threshold in Vray2, then render becomes noisy.
So still that statement is not true in my tests.

[zeronove]

sorry. see above, I messed the VR releases

pixela, I'm with you;
I'm sorry to say this but, from my experience with 2.0, the new DMC sampler is SLOWER and/or NOISER compared to 1.5 whatever min/max or clr threshold value you put in, period