VRAM / Out-Of-Core question

[da23ad]

Hi guys,

I am a future user of Octane Render (I am still using the Maya plugin and Standalone demo versions) and I am trying to figure things out about the VRAM limitation of GPU renderers in general ...

I currently have a GTX 780 Ti (3Gb) card and I have the second one on his way to get 5760 Cuda cores total with 3Gb of VRAM since 3Gb + 3Gb = 3Gb ...

I keep reading guys saying that VRAM depends on how big is a scene ... But what are we talking about when we talk about the size of a scene ? I am a bit confused because Octane is an out of core renderer so it is supposed to use the regular memory when it's running out of VRAM .... So even though it might be slightly slower than a fully VRAM render , this out of core feature is supposed to bypass the VRAM limitation ....

So what's the deal exactly ? What am I missing ? Why do people keep on telling people to buy a GTX Titan ?

[voon]

Because out of core is a brand new feature. Octane never had that before. I can't say how much slower OOC is... Maybe the impact is huge?

[glimpse]

as title of feature says: Out-of-core-texture - that applies only for textures (as far as I understand).

You still need to house Your mesh inside vRAM. Film buffer will eat some of vRAM depending on Your resolutions (if You go crazy enough..let's say 8k X 8k output will consume like 2+ GB of vRAM )

so is it good idea to get the same card? Well, yes in a way - You don't have wasted resources..

however, if You would get 6Gb instead of 3GB, You would be able to disable Your lower rendering card & then render on 6GB card only, allowing You to render bigger scenes with half of speed (if You take Titan/black instead of 780 - as both of them are based on the same chip & performs very close)

In the end You have to make Your mind, whether that extra investment is going to be useful for You or You just blow Your money..- think of output resolution, mesh density (try to load million, few of triangles to see the impact on vRAM..-I'm not sure now, but I think You're going to hit software limit of 19mln triangles faster then topping the vRAM as earlier almost max meshes could fit into 4GB card, but since then the limit was lowered)..

[da23ad]

Thanks for the answers guys...
So let's say that I wanna do mainly 3D things like 'Star Wars Rebels' or 'Big 6 Hero' ... Can that type of scenes fit in 3Gb of VRAM ?

PS : I think I'll have to run tests by trying to render scenes with X polygons then 2X polygons then 4X polygons...etc... and find out how many polygons is too many...

[da23ad]

I have just found the following numbers on another forum ... I guess they are relevant :

I was influenced by iray users who gave us some numbers about geometry vs memory.
I will list them here assuming that an octane user will be an iray user too (with the new free 3Dmax)

So for iray the video memory will receive 3 things:
-geometry evaluated in number of polygons (8 million triangles = about 1gb)
-textures+HDR uncompressed values(doesn`t matter that they are jpeg or png in the begining...so use medium resolutions)
-frame buffer (this is space for how muck your render will be in pixels)
frame buffer will eat your memory in the next amounts
1000 x 1000 = 425 mb
2000 x 2000 = 555 mb
3000 x 3000 = 750 mb
5000 x 5000 = 1125 mb
6000 x 6000 = 1200 mb


So iray is very dependent about the output size of your pictures(affecting the frame buffer value).

[glimpse]

what ever engine You're going to choose they all have the same base, underlying technology (more or less)..
GPU engines mostly written in CUDA..-so You're going to face more or less the same restrictions =)

[aoktar]

memory buffer formula for each frame buffer is width*height*16 bytes