All posts related to V2
#325113
I am also using Octane, contact me with pm.

My insight on the topic is that GPU is at its infancy, we have to wait and see. I am pretty sure if it grows to be a viable alternative, Maxwell will make use of it. I am also using it atm as a preview for what I want to render, or for small stuff. I can't really render big pictures.

Tassos Ringas
#325114
Yesterday i tried to make a comparison of Maxwell and Arion. But when i opened the scene in Arion my computer froze up. Even when i turned off my CPU for rendering it kept hanging.
It did render though, but not very fast. I am using a Quadro FX580. When I turned off both my CPU and my GPU it still kept rendering... That was the part where i gave up :lol:

I'll update my drivers and give it another try tomorrow.
#325115
tom wrote:Does anybody here own or have access to one of these GPU based engines, yet? Because, I'd like to share something small to test on it.
I would like to see it, I have Octane but with a non GPU Nvidia...I would like to make a test though...
#325116
Tok_Tok wrote:Yesterday i tried to make a comparison of Maxwell and Arion. But when i opened the scene in Arion my computer froze up. Even when i turned off my CPU for rendering it kept hanging.
It did render though, but not very fast. I am using a Quadro FX580. When I turned off both my CPU and my GPU it still kept rendering... That was the part where i gave up :lol:

I'll update my drivers and give it another try tomorrow.
It looks like you're working with a bad build.
#325121
A few days ago, I've revisited H.Jensen's frosted glass dragon scene with Maxwell Render V2 and
it turned out pretty satisfying in terms of quality I've expected. As you see, the scene is very simple
and it's only demonstrating rough dielectrics. The subject is not very challenging for most of the biased engines
but, it could be a real pain when it comes to unbiased cores unless they are not properly optimized/implemented,
especially for interactive and realtime rendering as their BSDF and microsurface approach is playing a huge role.

Here's the original work done by Henrik Wann Jensen (2003):
http://graphics.ucsd.edu/~henrik/images/raytrace.html

Image


Below is the frosted glass dragon rendered with Maxwell Render V2:

Image


So, I thought I should share the scene and you may give it a try with any other engine you have access.
Of course, this is a quality/ability test rather than being a speed-challenge.

The scene has a blue studio ground, a dragon, a small emitter plane and a dim skydome as ambient to match the original reference.
But, you can use either of them or both or more when the features of the engine don't match (only emitter, only skydome or only IBL or all).

Here's the scene OBJ files (all pivots at zero, no further relocation needed):
http://www.fileden.com/files/2009/5/30/ ... ne_OBJ.rar

Resolution: 800 x 640
Lens: 85 mm

You may also try FBX file instead:
http://www.fileden.com/files/2009/5/30/ ... dragon.rar


There are 3 goals in doing this:
1) Highly rough transmissive surface achieving blurry refractions.
2) Full caustics and TIR including the caustics on its own shadow.
3) Clear glass material core with no SSS/translucency or other effects is essential.

I want to leave this test open to any kind of render engine for comparing the quality of outcomes to Maxwell Render.
This is not a duel, so my respect goes to the ones which can at least succeed and demonstrate its quality in comparison.
And of course it's very sad for the ones which may fail awfully. Let's see... :) Awaiting for your renders...


COPYRIGHTS:
Dragon model by Stanford University Computer Graphics Laboratory.
Reference image rendering by Henrik Wann Jensen.
#325125
As I said, Octane doesn't seem to support blurry reflections yet. However, the following picture was rendered in 1 minute:

Image

As I mentioned earlier, it's very fast for test renders, or even small final ones. The first image I did was the one below, took me about 4 hours to get around the program, model the room, UV texture it and paint the maps. It rendered in less than 2 minutes, and apart from the fact that Octane doesn't allow the sunlight to pass through glass, I believe it's a nice start, to get a feeling and then take it to Maxwell for refinement:

Image

Image

It is by no means better than Maxwell, as the latter is years ahead. But it's nice to have realtime DOF and sunlight and decide what you want to do with the image.

Tassos Ringas
#325126
Ringas wrote:"...apart from the fact that Octane doesn't allow the sunlight to pass through glass..."
....er...really?

As Mihai sort of alluded to above it seems like most of the GPU renderers out there right at this stage get as much speed improvement from dumbing down the rendering capability as from GPU rendering. If all you were needing to render were images w/ the quality and features capable in most (all?) current GPU renderers (no displacement, sss,...um...glass?) then you wouldn't need a GPU render. Who cares about a rendering taking 1/10th the time if you're going from 10 minutes to 1 minute? When GPU renderers are capable of 90% of the features Maxwell has and are taking renderings down from 8 hours to 20 minutes, then I'll start clamoring for Maxwell to catch up. As it stands right now, it still looks like it's the GPU guys who are playing catchup.

I think it's important to remember that for all of us speed, however important, is still secondary to quality (if it's not, you're simply using the wrong software).

-Brodie
#325130
Brodie, I don't know when you started using Maxwell, but at the very beginning it also had its limitations; look where it is now. Don't rush to judge. Even MW had this problem with light entering through glass and it was solved in v2, Octane is still in beta.

Concerning what you said about 1 min vs 10 min renders; I do care, cause sometimes it's not a matter of how long a render will take but how many pics I can produce in a certain amount of time. In this case I would have 10 pics in 10 mins. And I can go home sooner or the client has understood what he's getting. I could even use it as a client-session application Flame-style, so he can decide on views and materials and not break my b...s afterwards.

At the end of the day, I am going to use the tool that produces what I need, in the time I have available. If a client or my boss doesn't understand what it takes to achieve this extra 5% that Maxwell can provide, they won't pay for it, and guess what; I am not providing it. Doing stuff of my own is entirely different and I can afford to use what provides this extra 5% whatever the cost, money or time wise. And just to be clear, for fast renderings and animations I was using Brazil and then Vray, but for the really nice stills Maxwell is the king. Now it seems I can look into accompanying Brazil and Vray with Octane, this is all I am saying. The way I see it, it won't catch up to Maxwell sooner than 2 years, and by that time Maxwell will be still ahead.

My 2 cents,

Tassos Ringas
#325131
The worst thing with GPU renderers is that they are very hardware dependable.... since GPU cards are something new, I believe we 'll see a lot of improvement fast...
Btw, don't concetrate to Octane, for example iRay may not have this "glass" or "blurry shadows" problem...
Anyway, Octane is very young and very cheap and still in beta so there are things that have to be forgiven...
Don't forget the RCs of Maxwell (where many of us wanted their money back!!!)
#325134
Ringas wrote:As I said, Octane doesn't seem to support blurry reflections yet. However, the following picture was rendered in 1 minute:
It's OK, Tassos but, Maxwell pulls the same thing in fractions of a minute. I understand they are not mature enough but, my example was not that complicated. It's something that any physically-based engine must have. So, you mean it does have a roughness control only for reflectance or it doesn't even have it? How do you deal with glass? Only specular (rough 0)?
#325136
So for now it seems GPU renderers are best for previewing purposes. How long does it take to have a finished render, also with refractive materials or reflected caustics? At what resolution? But even for previewing purposes how does the export/import work? If you want to just translate/scale/rotate any geometry, how long does the export take each time for a moderately complex scene before you can actually start to get the 'realtime' preview? Or say you want to reposition an emitter in the scene. How many of the GPU renderers currently have real emitters and not only IBL?
#325138
So for now it seems GPU renderers are best for previewing purposes.
if you take a closer look at Arion it obviously is not.
the ram limitation of GPU is soon over and the footprint of a scene needs much less GPU ram than normal ram.
there are still some limitations, of course, like limited number of textures and materials, but still for mayn scenes more than enough.

here is a nice example which looks very much production ready. all images rendered for 30min. - almost noise free. (resolution is 1920x1080)
http://randomcontrol.com/forum/viewtopic.php?t=5348

with 4GB tesla cards scenes with millions of polygons are no problem. all images rendered for 15min (2000x1500)
see here:
http://randomcontrol.com/forum/viewtopic.php?t=5415

hey, i don´t want to make advertisment for Arion (i don´t have it), but this sure looks good and i want NL to know what the competition is about.
  • 1
  • 3
  • 4
  • 5
  • 6
  • 7
  • 11
render engines and Maxwell

I'm talking about arch-viz and architecture as tho[…]

When wanting to select a material with File > O[…]

> .\maxwell.exe -benchwell -nowait -priority:[…]