Daz 3D is part of 
Connect
DAZ Productions, Inc.
7533 S Center View Ct #4664
West Jordan, UT 84084
Licensing Agreement | Terms of Service | Privacy Policy | EULA
© 2025 Daz Productions Inc. All Rights Reserved.
Comments
The benefit gap widens with size of the render, but in general, I wouldn't advise a person to buy four cards. When I bought them we didn't really know much about Iray yet, including the fact that the effect of cards after the first one is greatly diminished. Having a second cheaper card to run the monitor isn't a bad idea, but it's better use of your money to get the best single card you can instead of two moderate ones. I would've gotten one GTX 740 (because they were like $100) and one Titan X if I had known; and when the GTX 10x series is Iray compatible I'll only be buying one of those.
...the payoff I see with multiple cards is Viewport response when working in Iray view mode. Last year when Mec3D posted video of her then, dual Titan X system (6144 cores total) screen refreshes in Iray mode were very fast. This is helpful as one doesn't have to stop, run a test render, make scene adjustments, run another test render, etc as the scene continuously renders as you work. It's wasn't totally "realtime" but pretty close.
Here are my results (with default camera):
Single mesh: 53 sec
Single mesh +10 instances: 1 minute 23 seconds
Single mesh +50 instances: 3 minutes 44 seconds
has no one realised that it isn't only lights but surfaces also play a huge part in render times.
It does change light placement and apparant brightness. Picture a photometric light is generated by a string of xmas lights. If they're all bundled tightly it appears brighter. As you change the geometry and size(s), the string is spread across the geometry and appears less bright, though the same amount of light is being put out.
Interested in this idea of creating instances of lights, I decided to try something. I created a plane and gave it an emissive surface. i then replicated it twice and placed it in 3-point lighting spots to see if it illuminated a human figure reasonably well on all sides. Since all the planes have the same intensity, I used distance to control the ammount of illumination on the figure from a given plane. Also used the trick of setting opacity cutoff to .001 so the planes were invisible. Set the environmental settings to "Scene Only" and made sure there were no other light sources, and rendered.
First render is a control: three planes (not instances) in the same location as the instances, with surface setting identcal to the instances. Instances are OFF, all lighting is from the planes. Render took 6 minutes 6 seconds to get to 300 iterations.
Second render is the one with the instance planes. Turned off the real planes, turned on the instances and they alone lit the scene this time. Took 5 minutes 18 seconds to get to 300 iterations. So faster.
The verdict: instances lit the figure up okay actually. Light from all three planes did illuminate the figure and it seemed to render marginally quicker than real geometry for some reason. That's good. But the negative thing was the iris of the eyes rendered black. I found that this was because when I set the planes to .001 opacity, they reflected black in the eyes. In the render with the real planes,I set the opacity on the one plane to the right (not visible) up to 100% and the eyes rendered correctly. I could not do that for the one with the instances because it would have made the others all opaque too and they would have been visible. So you get bad reflections this way.
For what it's worth.
@grinch2901
I suspect it rendered slightly faster bcause the instances have less to load into memory, i.e. the render starts faster.
Could be, but I do use the Aux Viewport set to Iray mode so it's typically already got the textures and geometry already pre-loaded into my card when I hit the render button. The dfference isn't much, the way to tell if it was a one-time thing would be to either start timing it with a stopwatch (or equivalent) from the time iteration 1 pops up or maybe tray a larger scene with more lights and see if the improvement scales. If it's simpler / faster for each instance then the more fake lights you have going on, it ought to multiply that savings. In theory. Of course, my rig is too slow to do a real test, it would take forever to render a decent size scene and I'm notoriously impatient!
...the face in the isntanced version seems to have less light on it.
I agree Kyoto, the light isn't exactly the same. It's lit, but one is more grainy in the face area, even though iterations were the same. Interesting how this works, isn't it?