Does a GPU rendered scene gets loaded in the GPU to get rendered there?

nicolasd

I have a half decent laptop which gets stuff done, but I've recently got into a redshift course and I assume my 1050 2gb is soon going to be useless once i get to do renders. I can't afford a full rig for the time being, but I can get a decent gpu and i want to connect it externally. But i'm unsure about one thing, is my weak laptop going to affect the render itself? I have read the scenes you want to render are loaded into the gpu, and then the scene gets rendered in there without the pc performances having any impact on it. Is this correct?

Also, do you think it's a feasible option?


PS: I have been googling in different ways to get an answer but i can't seem to hit the correct keywords so far

Yerus

The render quality isn't due to the laptop performance if it's rendered in unbiased (bucket) way. In this case, If the laptop performance isn't enough, it will simply not render at all. Only if it gets rendered in biased or progressive way that the performance may interfere.
What software are u using as a renderer?

gnoop

I am using Octane gpu render . it's biased one if I am right.  Video Ram  is the main limiter.  A scene should be loaded in video memory.   It could work with RAM too but getting super slow and unstable.  My 6 gb video memory is hardly enough  for what I render.

Form what I read in Octane forum people do use e-gpu with mac books. But I never tried.

Noren

The video card will do the rendering, but the data has to get there first. 2 GB is indeed quite low, but that of course depends on the scenes you are planning to render. Some GPU renderers (and apparently redshift as well) allow for data swapping or utilizing the system ram when video ram runs out, but this will slow down the whole process significantly and especially if your laptop isn't up to par. 

What external connection for the video card does your laptop offer?

nicolasd

Yerus said:

The render quality isn't due to the laptop performance if it's rendered in unbiased (bucket) way. In this case, If the laptop performance isn't enough, it will simply not render at all. Only if it gets rendered in biased or progressive way that the performance may interfere.
What software are u using as a renderer?

Redshift (on maya), which is biased as i have just figured out. The laptop performances will have a impact on the rendering then, but how exactly?

gnoop said:

I am using Octane gpu render . it's biased one if I am right.  Video Ram  is the main limiter.  A scene should be loaded in video memory.   It could work with RAM too but getting super slow and unstable.  My 6 gb video memory is hardly enough  for what I render.

Form what I read in Octane forum people do use e-gpu with mac books. But I never tried.

Octane is unbiased. Redshift can also use RAM for renders. I plan to get an 8gb card and i have no idea what limitations comes with it. What kind of scene can you render with it, and what kind can you not?

I'm getting skeptical about that and I might've not thought enough when starting that course

I might as well give my pc specs: 

i5 7300hq that can boost to 3.5ghz
1050 2GB
12 ram
SSD memory

Noren

I'm not sure what Yerus is referring to, to be honest. Unbiased is not a synonym of bucket and also not necessarily of quality. Maybe he thinks about a specific application that offers both.  Also render quality isn't due to performance in biased mode, either.

8 GB should be enough for most smaller scenes and if you are doing a course, I can't imagine them to expect you to have more.
Scenes can take more video ram than the file size of your scene and assets on disk, but roughly speaking if you don't plan to use a fully cinematic workflow with dozens of huge UDIMs per character/object and crazy high polycounts you should be fine.

EarthQuake

Generally speaking, it breaks down like this:

For GPU rendering, you need to fit all the content into VRAM. Content generally means: meshes, textures, special effects, plus room for render buffers for various passes and post processing effects. The video card needs to have the data in its memory to render it. Some applications may allow you to page memory to system RAM but this is typically slow. It's worth noting that size on disk and size in VRAM are not the same thing. For instance, you can save your textures as PNG to make them smaller on disk, but this does not save VRAM.

Speed with a GPU renderer will depend on how fast your GPU is. When using a GPU renderer, there isn't much work for the CPU to do, so having a faster CPU doesn't make a difference. Note that this is not always the case, you may have animations or particle systems that are calculated on the CPU. It really depends on exactly how the app works.

Quality will depend on your render settings. Higher quality = slower render times. Assuming you can fit the scene into VRAM, an older/slower GPU is only a limiting factor in that it will take longer to render the scene than it would on a better card.

nicolasd

That sounds reassuring. Will I be fine connecting it to a x1 PCIe slot though? 

Yerus

Noren 
I don't know what u talking about man. Unbiased and bucket are exact synonyms. And Progressive ir due to the GPU performance, otherwise he will stay forever in front of the computer waiting the thing to render.
Besides, it's what EarthQuake said: render quality will take on time. What are your purposes? Game, still image or film?
This video talks about how to improve render time by optimizing your render settings, it's made for Blender, hence many topics are translatable to other render apps. 

https://www.youtube.com/watch?v=8gSyEpt4-60

Noren

 @Yerus
"Unbiased and bucket are exact synonyms" Not sure where to go from here.  Agree to disagree? Else you'd have to help me understand where you're coming from. How do you define bucket? Are you using a specific rendering software? Something from Blender?

Yes, render speed is due to performance, but you initially implied biased render quality was due to performance, which would only make sense if you assume a fixed rendering time like it's offered (as an option) by some renderers.

nicolasd

welp, i'm going with a 1070 ti, and my external dock is arriving soon. Gonna go spend some time reading forums and blogs to figure out more details about how it all works. Thanks for all the answers, it did clear out some of my doubts. it's a good investment anyways.
On my render tests though my cpu is all on 100, how is that happening... if it's gpu rendering? 

Yerus

Noren , I would define bucket exactly like unbiased. As for the "bucket" is from V-Ray actually.
And yes, GPU performance that interfere on the biased rendering, because rendering forever is not an option.

nicolasd It most likely because your using mixed rendering settings.

gnoop

I could also say based on my experience  that GPU renderers are interactive only on simple scenes  and once you scene is getting huge with gazilion of instances, hi res displacement, hairs etc they are somewhat starting to get   less convenient than CPU ones.

  They still do final render quicker   but  getting less interactive than CPU ones which usually start to show you something noisy and pixelated almost instantly so you can get some visual  feedback and start to adjust materials and scattering.  

  And with GPU render  you see just black background first few minutes on a huge scenes  before the render  would pack everything into video memory.     Octane allows to move some object quickly but if you need to re-scatter particles, hairs etc,  it takes forever again.

So they kind of  loosing its advantages  more VRAM they need. 

Noren

@Yerus: Bucket in V-Ray is just the way the rendering is split up (as opposed to progressive). For GPU brute force rendering (which is close to unbiased if you don't limit it) bucket rendering mode wasn't even available in the beginning iirc.