insane render times with "use cached GI" on VrayHairMtl

[sharktacos]

I did some tests with the "used cached GI" button on the VrayHairMtl3 material and noticed that I was getting huge hits on the render times when this was disabled. I have attached the Maya scene which is a simple sphere and directional light with Vray fur on the sphere with the VrayHairMtl3 material applied. I am using Maya 2014 Vray 2.4.


As I understand the "used cached GI" it is supposed to render in BF when on even if the scene is set to IR. However looking at these tests it seems that it creates an incredible slowdown on the render times

All tests AA min1 max4

1) BF/LC (BF subdivs:20)
a) use cached ON

time: 2 min 2 sec

b) used cached OFF
time: 4 min 7 sec

Render time is doubled when on.

2) IR/LC
a) use cached ON
time: 18 sec

b) used cached OFF
time: 25 min 35 sec

Here the render times go through the ceiling.


3) BF/BF (BF subdivs:20)
use cache ON

time: 3 min 31 sec

This last test is as a control. If having this off simply means it renders the hair with BF then one would expect that the render time would be comparable to the BF/BF render times. However it is almost 10 times longer with option 2B above (25 minutes), compared to here with BF/BF (3.5 minutes).



In sum, when the option is OFF, rather than it helping, it instead makes the the render times shoot through the roof. Is this a bug? Surely this is not how this is supposed to work.



Here's the Maya scene I did the test with:

hairCacheTest2.mb.zip

[tashko.zashev]

Hi,

To be more accurate your test you need to set 50 subdivisions to the BF/BF setup, because when "use cached GI" is disabled the Brute force subdiv. will be taken from the Irradiance HSph. Subd. field.
Also it will be better to test with AA min/max 1/1.

I assume this will produce similar render time result as the one in point b)

[vlado]

This last test is as a control. If having this off simply means it renders the hair with BF then one would expect that the render time would be comparable to the BF/BF render times. However it is almost 10 times longer with option 2B above (25 minutes), compared to here with BF/BF (3.5 minutes).

The difference is in the subdivs used for the brute force calculations. When you use IM/LC, the irradiance map subdivs are used, which by default are set to 50 (=2500 samples), compared to 20 in your BF/BF test (=400 samples). So the time difference is perfectly normal.

In sum, when the option is OFF, rather than it helping, it instead makes the the render times shoot through the roof. Is this a bug? Surely this is not how this is supposed to work.

On the contrary, this is exactly how it is supposed to work. Using cached GI (when the checkbox is enabled) is always going to be faster than using brute force GI (when the checkbox is off), although the latter case is of course more accurate. BF+LC (and use cached GI enabled, which it is by default) with "retrace" enabled for the light cache is probably the best way to render hair.

Best regards,
Vlado

[sharktacos]

Thank you both for that explanation. It would be helpful if this information was included in the online docs

Based on this, I'd like to submit a "wish list" to change the behavior:

1) It is common to have the BF subdivs lower than the IM subdivs (as the default values of BF:8 and IM:50 indicate). So rather than it taking the subdivs value from the IM, it should instead take it from whatever the BF subdivs were set to before. Even better would be to be able to set the BF subdivs manually in the material with an override.

2) As Vlado said, the best approach is to use BF/LC. However, this option actually overrides both the primary and secondary GI to BF (BF/BF not BF/LC). It would be preferable I think if this only affected the primary GI, not the secondary GI.

3) Finally, it is a bit confusing that having the option ON means it does not override the GI and use brute force, and having it OFF means it does override the GI (ON=OFF, OFF=ON).

Based on all of this I would humbly propose something like the following GUI:



The default settings would look like this: