Announcement

**^Lele^** · 06-10-2015, 04:32 PM

You're right: in general, for sub-pixel detail, one needs a high number of AA rays, and a low number for MSR, say 3.
More secondary rays (higher MSR) would likely get "wasted" by the time the AA managed to finish its job, which is going to take its due time anyway.
In other words, a large number of AA rays will be needed no matter what, so for each AA ray the number of secondary rays can and should be kept low, as on the overall average they will matter less, while the AA rays will do most of the meaningful work.

I opened the file, reset the vray settings, and changed only to progressive, 4-64 (16 to 4096 rays per pixel), no rendertime limit, noise threshold of 0.01, and ray packet size to 8 (less than ideal, but will help by the time i have fewer pixels to trace in the image, keeping cpu usage high. Things will change in this respect for the better, at some point.).
I set MSR to 3, turned on GI, leaving it at its defaults, and rendered.
The CPU on the laptop is a very decent Core i7-4710 @ 2.50 Ghz (four physical, eight logical cores).
Notice that this requires nothing else than the currently released VRay (3.20.02) to work like so, provided you also set the global subdivs multiplier to 0.0, so to discard local subdivs in shaders and lights.

Possibly if you want it cleaner still (although this would work a charm even animated, from my personal experience), lowering noise threshold would ofc increase rendertimes.
However, in my book, four and a half minutes, and seven, respectively, for this image quality are quite decent, especially if one scales up image size to 4k and CPU power to current render boxes.
Should it take an hour per frame, at very clean quality, i would think it's a pass for a hair pass (pun intended) even for VFX grade work.
But even if it took two, to be honest. :P
I have attached the files, but i don't think my OX demo saved it quite right.
In case, save out and reload the render settings.
Feel free to save the render settings in a preset named something like "Hair with GI". It ought to be a decent starting point for hair, fur and generally sub-pixel detail, no matter the shot, motion-blurred or not.

Static Image Render Stats (notice the number of CAMERA Rays Per Pixel : ~53)

Code:

================================
Tiled bitmap cache size set to 1000 MBPreparing renderer...
Using Intel Embree ray server
Using probabilistic light evaluation with 16 lights.
Max ray intensity is enabled: rendered result may have incorrect brightness.
9 render object(s)
9 render instance(s)
3 render light(s)
Found 4 materials/texture maps
Rendering frame at time 0
Non-camera view
Scene setup: 1.2 s
Using embree ray tracing.
Building embree static trees took 15 milliseconds
Static hair tree statistics:
  Memory usage: 117 MB
  Number of hair systems : 3
  Number of bezier curves: 664470
Tracing 1000000 image samples for light cache in 64 passes.


Light cache contains 241642 samples.
Light cache takes 40.3 MB.
Average rays per light cache sample: 5.79 (min 1, max 538)
Start progressive rendering, server mode 0, perform rendering 1
Progressive sampling base buffer memory usage: 27.66 MB
Additional progressive sampling filter buffers memory usage: 293.47 MB
Progressive rendering finished: specified noise level reached
Sampling level reached: 57.73 subdivs (3333 passes)
Number of raycasts: 287328713 (437.31 per pixel)
  Camera rays: 34790501 (52.95 per pixel)
  Shadow rays: 149952267 (228.22 per pixel)
  GI rays: 150296895 (228.75 per pixel)
  Reflection rays: 0 (0.00 per pixel)
  Refraction rays: 0 (0.00 per pixel)
  Unshaded rays: 8001232 (12.18 per pixel)
Light cache utilization: 33.21%
Number of intersectable primitives: 66308
  SD triangles: 66308
  MB triangles: 0
  Static primitives: 0
  Moving primitives: 0
  Infinite primitives: 0
  Static hair segments: 1993410
  Moving hair segments: 0
Region rendering: 272.0 s
Total frame time: 274.2 s
Total sequence time: 276.4 s
0 error(s), 0 warning(s)
================================

Motion Blurred Image (1f duration, 5 geometric samples, camera-only spin of 360 degrees in 15 frames)
Again, notice how it took proportionally more as the number of CAMERA rays per pixel increased: it's now 79, instead of 53, and rendertime went from 276s to 449s.
The proportion between RPP is ~1.45, that between rendertimes, ~1.62.
This is due to the fact that VRay automatically shaded roughly the same amount overall, although a bit more for the motion blurred one (437 total raycasts per pixel versus 471).

Code:

================================
Tiled bitmap cache size set to 1000 MB
Preparing renderer...
Using Intel Embree ray server
Using probabilistic light evaluation with 16 lights.
Max ray intensity is enabled: rendered result may have incorrect brightness.
9 render object(s)
9 render instance(s)
3 render light(s)
Found 4 materials/texture maps
BEGIN FRAME 5; TASK 1 OF 1; (NO SAVE);
Rendering frame at time 800
Rendering from camera PhysCamera001
Scene setup: 2.1 s
Using embree ray tracing.
Building embree static trees took 14 milliseconds
warning: Embree does not support multisegmented motion blur. Using regular raycaster for motion blur.
warning: Embree hair does not support multisegmented motion blur. Using regular hair raycaster for motion blur.
MBTree statistics:
Number of node faces: 16
Number of tree nodes: 15
Max tree depth: 4
Building moving MB trees took 5 milliseconds
MBSegmentTree statistics:
  Memory usage: 567 MB
  Number of hair systems : 12
  Number of strands: 98440
  Total number of segments stored: 7776760 (59 MB)
  Max tree depth: 43
  Number of tree nodes: 6030205 (138 MB)
  Number of leaf segments: 95939830 (370 MB)
Tracing 1000000 image samples for light cache in 64 passes.


Light cache contains 172678 samples.
Light cache takes 26.4 MB.
Average rays per light cache sample: 6.84 (min 1, max 764)
Start progressive rendering, server mode 0, perform rendering 1
Progressive sampling base buffer memory usage: 27.66 MB
Additional progressive sampling filter buffers memory usage: 293.47 MB
Progressive rendering finished: maximum sampling level reached
Sampling level reached: 64.00 subdivs (4096 passes)
Number of raycasts: 309626800 (471.24 per pixel)
  Camera rays: 51991768 (79.13 per pixel)
  Shadow rays: 153725884 (233.97 per pixel)
  GI rays: 155641163 (236.88 per pixel)
  Reflection rays: 0 (0.00 per pixel)
  Refraction rays: 0 (0.00 per pixel)
  Unshaded rays: 6171328 (9.39 per pixel)
Light cache utilization: 32.73%
Number of intersectable primitives: 7843080
  SD triangles: 66304
  MB triangles: 16
  Static primitives: 0
  Moving primitives: 0
  Infinite primitives: 0
  Static hair segments: 0
  Moving hair segments: 7776760
Region rendering: 414.5 s
Total frame time: 449.0 s
END FRAME 5; TASK 1 OF 1; (NO SAVE); 449s; 0 faces; MEM 2332 MB;
Total sequence time: 451.6 s
warning: 0 error(s), 2 warning(s)
================================

So, not only it's fairly quick, it's also very predictable despite the addition of Moblur.
Predictability will be even better, as well as the noise distribution, as soon as the new service pack will come out (don't ask, i have no idea of when that will be.).

Announcement

Ornatrix & Optimizations

Ornatrix & Optimizations

Comment