3.3.02 RT GPU issues

[super gnu]

1:

i can only do one render session. if i cancel a render, or let it finish, the next time i try to render with RT, i get "cuda error 719:exception occurred on the device while executing a kernel"

if i am running using the spawner locally, i just have to restart that.. if im using the in-process RT i have to restart max.

-as a note, not having to restart max after rt issues is a major reason to stick with the external spawner setup until RT is as bulletproof as vray adv.


2: ive noticed something wierd with the lightcache. if im using in-process, it seems to do a tiny bit of the first renderpass, before stopping, doing the lightcache, then starting the render fully.

if im using the spawner locally, it actually does the whole lightcache, starts the render, stops after a small part of a single screen update ( only centre few tiles are done) then it redoes the whole lightcache again before starting the render properly.


-if i am rendering an animation, it seems to only do this on the first frame.


3: displacement global switch doesnt work using rt gpu as production renderer. i didnt check the other global switches.


4: rounded corners edges tex works in the bump slot (great) but ive noticed it seems to have bad aa issues when its distant. i had a slow camera pan, and thin edge highlights were flickering a lot, even with a low noise threshold.

5: multisubtex works, but warns about an 11 slot limit.. i have 25 slots in my (brick) material. how does it deal with this? i didn't notice the wall looking odd.

6: falloff map doesnt seem to work with light/shadow mode?

[savage309]

First thanks for the reports !
1. Keep in mind that the 3.30 versions we had released so far were beta and as such you could experience more issues than we would like to.
On the topics - for the 719 error a scene would really help us check what is going on. We fixed a couple of issues in that direction, but I have to test to know if your one is fixed too. In the final version there will also be a checkbox to start the RT our-of-process (just as before).
2. The fact that when you are out-of-process could make the LC to be build twice is known issue, but we still haven't managed to get to that.
5. I think MultiTexture does not has that limitation, so you can use that. As a workaround for the MultiSubTexture you can add 3 of them.
I will add 3. and 6. in the bug tracker, but for 4. we would be thankful if you can share a scene to test with.

[super gnu]

just a quick update to point 01.

i have a titan x and a gtx 670. if i disable the gtx 670, i can do repeated renders without the crash on the second render.

[Vladislav Vulchev]

Hi, thanks for the feedback. Concerning the 6th question. We've discussed it and for now there is not a good way to make it work, because of big memory loss and many other things. The behavior of this method in the production renderer is also questionable. The first thing that comes to my mind for a workaround is to try to use the SSS material.

[super gnu]

im not sure i follow, could you explain how using the sss material can be equivalent to using the falloff map in shadow/light mode?

i need to mask a texture so it only appears when the material is not in direct light.

[tashko.zashev]

im not sure i follow, could you explain how using the sss material can be equivalent to using the falloff map in shadow/light mode?

i need to mask a texture so it only appears when the material is not in direct light.

Hi,

I think the best way to get the same result in GPU when Falloff map in shadow/light mode is used in the shading network is to Bake (single frame or animation depending on your project) the object that has this map applied and the reuse it as a bitmap file. Unfortunately this is the only approach that we could suggest until we manage to implement support for this mode. Excuse us for any inconvenience.