zajooo

Hi,

I have decided to post this topic to get deeper into the rendering part ... and share experiences about viewport performance

I made quite nice optimalization for vieport performance on my HW athlon 1,7Ghz, 768 RAM, ATI Radeon 9600 XT 128MB

I started up with an Interior scene with 16.ooo triangles on 24 fps in viewport and ending up with 135 fps ... rising the performance over 5x

A was also getting realtime performance displaying 1. million triangles from another scene

... keeping these things on mind

1. Mesh is faster to display than EPoly ... the graphic cards are based on displaying triangles (When displaying convert to Mesh, when modeling convert to EPoly)
2. Groups, Instances and References of low poly objects lowers performance ... parse of the dependency tree is needed before displaying (Collapse object with the same material to single mesh)
3. Modifiers makes the models to display slower ... converting the modifier stack before displaying is needed and it takes CPU time (Collapse the modifier stack)

How is it with the optimalization of the scene for rendering ??? Keeping the triangle count unchanged.

1. How does the renderer GI access the scene in VRay? Is it better to attach the objects into one or keep them in groups?
2. Does the renderers ray shooting algorithms, exclude the non visible objects to the ray first, or parsing brute force through the triangles of the scene?
3. Is it faster to find the visibility parsing the triangles across single object or by accessing multiple objects through multiple groups?

Please share any links where I can find tips according optimalization for rendering.

Thanx a lot

Zajo

notamondayfan

another tip to speed up viewport viewing - if you have any mesh smooths or turbo smooths, right click the light bulb in your modifier stack and click "off in viewport". this will only turn on the smooth at render time. works for other modifiers too.

__________________
Pikcells Animator, Modelling & Environment Designer

sterealkey

I also used to try all sorts of things to make my viewport navigation a bit smoother.

I dont know what they changed in Max 9, but scenes that used to give me a headache in Max 8 I can now open in Max 9 and navigate around them like Superman on Red Bull.

I didnt change any hardware, just software. Whatever they did, it impressed me.

__________________
www.ocellus.co.za - innovation in design, illustration and visualization

zajooo

They used the D3D Xmesh caching feature of DirectX 9 ;-) ... and it makes a most out of it

zajooo

So I have spent the evening with some rendering tests ... some results are quite interesting ... at least for me ;-).

Lets take a better look on the scenes ...

Both scenes have 2.176.566 triangles

1) the first scene has 13.585 objects
2) the second is the same scene optimized for viewport and has 70 objects

Lets see the first VRay basic settings rendering statistics of 1)

• Max Tree Depth: 39
• Averange Tree Depth: 24,4
• No. of nods: 474.041
• No. of faces: 6.176.470
• No. of leaves: 221.988
• Memory used: 84 MB

Now take a look on statistics the 2)

• Max Tree Depth: 39
• Averange Tree Depth: 24,4
• No. of nods: 471.291
• No. of faces: 6.170.277
• No. of leaves: 220.312
• Memory used: 96 MB

1) Rendering time: 15 min and 47 sec.
2) Rendering time: 11 min and 36 sec.

actualy after rendering I have realized that one of the matterial is not the same so I have decided to override the material to eliminate the influence of the materials ... I used the standard VRay material

... and the rendering time
1) 5 min. and 50 sec.
2) 5 min. and 47 sec.

Leading to over 63% redering time savings

and one more test before some conclutions

I have changed the VRay material to Max standard material and tested on the first scene
1) getting the result 11 min. and 51 sec.

Conclusions:

After rendering the normal scene and viewport optimalized scene ... I predict and conclude

• the rendering statistics of the scenes are very simillar so I predict the foloving ... VRay builds its own parsing structure independent on model structure, probably it decomposits the models to triangles (or Polys) ... there is no need to optimalize the objects in the scene for rendering
• It is better to use VRay material, because it is optimized for rendering and can lead to half of the rendering time

Cheers, Zajo

sterealkey

Wow

I'm so pleased that someone actually took the time to do this experiment. Its something thats been on my mind for quite a long time.

Thanx Zajooo!

At least now we know its not worth spending time on simplifying your scene, coz it only saves you 3 seconds at the end of the day

__________________
www.ocellus.co.za - innovation in design, illustration and visualization

zajooo

... so lets continue

Another test scene, 16.ooo triangle, an interior of a shop, render res. 800x600 proving the qualities of VRayMtl, for rendering in VRay.
using default max materials 161 sec.
converting materials to VrayMtl 72 sec.
the render lead to 56% time saving.

Well at the end of the week I made some upgrade of my computer ... I have added extra 1GB of memory (leading to 1,5GB), and I was quite surprised with the rendertime downswing.

The test scene was as in previous post a building of about 2 mil. triangles, using XRef for details, at resolution 4000x3000, VRay renderer. The rendering time was about 4 hour before upgrade.

After the upgrade, and merging the XRef scene into one file (this was posible without caching thanx to the extra memory) ... I have achieved rendertime less than 1 hour, leading to 75% time savings

Afterwards I have took all od the objects and made 2 VRay proxies out of it. The test render using VRay proxies took 2 hours.

The both renders using proxies and loaded geometry have taken 1,08GB of memory. And I could not figure out why.
Everybody talks about saving memory using VRay proxies, how does this work? Does enybody have som explanation? Does the Proxies start to workout when getting over the memory limits for a scene???