Question about Iray memory use

MysticX

I recently made the upgrade from a 2Gb Geforce 960 to a 6Gb Geforce 1060 (I had trouble fitting busier scenes into the VRAM, and CPU rendering is glacially slow), and for my own curiosity, i screencapped the memory-use messages Iray gives when i start a render (Before it starts counting iterations), the numbers it mentions are:

Geometry memory consumption: 87.3477 MiB

Texture memory consumption: 5.46434 GiB

Lights memory consumption: 199 KiB   (I used "Sun/sky only" for lighting)

Material measurement memory consumption: 0 KiB

Material memory consumption: 485.516 KiB

The render is 1365x1024 in size and optimization is set for speed, not memory use.

The scene in question contains two clothed Genesis 8 females with morphs, both with the same skin texture, with two V4-era props used as background.

My question is: Did memory consumption jump up THIS much in the step from G3F to G8F?   In my simple (Probably uneducated, too) mind, 5.4 GiB seems pretty steep for just the textures, i did a render on the same system when i just got the new video card, which had 4 G3F characters wearing more complicated clothing material-wise (Metals and stuff), mesh lights, in an enclosed space, same resolution and that rendered fine, if i add two more characters in textures to the 5.4 GiB for the G8F-render, my system will go to CPU-rendering...

I noticed that as expected, G8F renders a bit slower than G3F for similar scenes (Mostly one character, simple prop background, sun/sky lighting), but the apparent jump in memory use really startled me.

Tobor

You don't actually mention if the GPU dropped out in the render.

The texture consumption value is for uncompressed textures. Iray normally compresses them internally and saves considerable space. You should do your render and monitor the actual memory consumption using a GPU monitor application, such as GPU-Z.

MysticX

Well, it didn't seem to drop out (But the memory vaules mentioned come up just below 6Gb), but i forgot to take texture compression into account, i'll get a monitoring tool to take a peek at actual VRAM use, thanks for that one!   :D

MysticX

Ok, i ran a GPU-monitoring program (i used "GPU shark") while rendering that same scene, and the results are in:

Basic VRAM use (From DAZStudio being open with the scene loaded) is 306Mb

Memory use during rendering is 2994Mb (From the 5.464Gib DazStudio mentions), that texture compression is pretty effective, shame DazStudio doesn't report on it...

I also tried doing the same render with optimization set to "Memory" instead of the default "Speed" under the render settings, and it then used 2944Mb, a 50Mb memory gain that is pretty neglegible unless i'm doing a render that only just exceeds maximum VRAM by a tiny margin...

There, now i added something intelligent to the thread too!   :P

Tobor

D|S only echos the render messages from Iray, and doesn't itself know what's going on. Getting the pre-compressed value is more useful because the compression is ongoing. By knowing the total value you can decide if you want to go back into your scene and either manually downsize textures to make them smaller, or elminate some what you might not need.

Memory/Speed affects geometry instancing, and comes into play for specific types of scenes. Sometimes it makes a huge differerence, sometimes (as in your case) very little at all.