Key to improving VR rendering speed

The key to increasing the speed of VR rendering, this tutorial is much more important than previous tutorials, if you are just stepping into the learning and ascent stage then it will be something you have to look at and he will let you quickly upgrade your skills to reach a much better effect than you die reading Ben, not much to say on the tutorial

VR's basic rendering method has mastered, although the material and lighting aspects there are many shortcomings, but the most urgent need to solve the problem is the VR graph speed problem. It takes hours of rendering time to be very difficult to accept, whether it is practice or the real plan is a long wait, so it is inconvenient to modify. Therefore, it is imperative to understand the fundamental causes of the impact of VR speed, optimization parameters.

A few recent examples, GI time is not long, is anti-aliasing out of the picture for too long. Maybe I do all the small scenes, such as large scenes of light more when the uproar sub-chart is very slow.

Let's sum it up:

Graphic antialiasing aspect Image Sampler (antialiasing)

1, without reflection blur, image quality similar to the case, adaptive subdivision fastest

Compare parameters fix rand=4; Two-level Subdivis 1/4;adaptive Subdivision RATE-1/2

2, a large number of rough surface multi-bump map, simple two-level fastest

Compare parameters fix rand=4; Two-level Subdivis 1/4;adaptive Subdivision

rate0/2,threshold=0.05

3, if the scene has a large number of fuzzy effects, including motion blur, fuzzy reflection, high-gloss blur, depth of field, etc., fixed is the fastest

If there are a lot of bump effects in the scene, including bump map, displace, noise, etc., adaptive

QMC or adaptive two-level is the fastest. Other cases with adaptive subdivison will be faster.

It is important to choose different anti-aliasing according to different conditions!!

A few other key parameters:

Threshold This value has a very large effect on speed.

The adaptive subdivison CLR thresh defaults to 0.1, the test is changed to 1, the speed is improved a lot of quality also can accept

In the adaptive QMC, the noise in the RQMC panel

Threshold default to 0.005 test changed to 0.1, fast quality can also.

If you select Use QMC Sampler in adaptive RQMC image Sampler

Thresh, then the domain values sampled here are directly determined by the noise threshold of the RQMC panel.

If it is not checked then there is the CLR Thresh to decide that the default 0.01 test is changed to 0.1, fast quality is acceptable

Object outline:

When the scene has a large number of small objects, the antialiasing mode is adaptive subdivison, if you tick object

Outline, then the rendering will be very slow. This time we do not check the object outline and use the CLR threshold to control the image quality.

NRM Thresh: This is a specific normal area sample, if the shadow is not required to the case of the CLR

Threshold set a large value of 5, tick the object outline and NRM threshold such speed also has a small increase.

There are two parameters in the glow map related to rendering speed:

Interpolation type interpolation types and sample lookup sampling lookups

As long as the interpolation type is set to least squares fit (least squares adaptation method),

And the sample lookup is set to precalculated

Overlapping (pre-computed overlap) can satisfy most of our needs and speed quickly.

If the quality is not up to the requirements, try another parameter setting (remember that nearest is only suitable for sketch previews)

Check sample visibility checks the visibility of samples to reduce light leakage, but slows down

About Coolmoon's Posts:

Ways to speed up: Select all objects with smaller areas, turn their global GI off, and allow only a few objects with a larger area to calculate the global GI

He mentioned that the anti-aliasing speed has been analyzed before it is no longer summed up.

Posts about Tyrannosaurus Rex:

1. Don't bounce with too much light.

2. Do not use too high glow map settings to render large resolution graphs

3. Frequently Save Glow Maps

4. Get a bigger memory.

5. Resist teeth not too high

6. Do not use too high a reflection refraction level

7. Do not use transparent map samples too high

8. Mandatory reduction of GI sampling of unimportant objects

9. Do not use bump mapping and blur

10. Do not render images that are too high-resolution, just enough.

11. PS can be used PS, than you spend 3 days max rendering much faster

This is his exact words!!

The following are the empirical parameters:

1, wood floor fuzzy reflection 0.85 and 3, the number of reflections to 1

2, the final photon graph im parameter rate-3/0,clr threshold=0.2,nrm

Threshold=0.2,hs=50,is=20

3, with VR lamp simulation skylight is hooked on sotre with irradiance

Map speed will be faster.

4, calculate the photon map and then add to the scene less influence of the lamp

5, turn off the auxiliary lamp to calculate the photon map, to be formally rendered and then open, if you want to effect better point can play a low-brightness VR lamp simulation photon rebound effect

This figure runs photons for 36 minutes, renders 24 minutes, the machine is formulated Atholn 2200+/1024ram, the figure size 600x366.

The scene does not hit a light, and the brightness is raised by the skylight and color map.

Parameters: AA (fixed=3); IM (rate=-3/0,ct=0.2,nt=0.2,dt=0.3,hs=50,is=20)

Key to improving VR rendering speed (II.)

In the previous article combined with the Forum of the Tyrannosaurus rex, Coolmoon, henry0421 their posts analyzed some of the experience and parameters to improve the VR speed, this time we control Aparkey

Study notes to look at. Most of the parameters that he mentioned have been verified by myself, and some empirical parameters can be used for drawing reference.

Last time about the impact of anti-aliasing on the VR speed is much more, so this time I will GI on the impact of speed to analyze. Primarily for light maps and light caches.

One, irradiance map (Glow map)

As we all know, the rate of the most impact on speed is the min value, then we will be in the case of Max fixed first to test how to change the min value of time to spend the comparison

The rendering time is averaged 4 times because it is a test, so choosing a smaller scene will be faster.

max=-1,min=-10/9.9;-9/9.75s;-8/

9.73s;-7/10s;-6/10.3s;-5/10.0s,-4/9.8s;-3/9.85s;-2/10.45s;-1/12.55s

max=0,min=-10/17.3s;-9/17.3s;-8/

17.5s;-7/17.8s;-6/18.1s;-5/17.7s,-4/17.3;-3/17.4s;-2/18.3s;-1/22.5s.

Although only two sets of data were tested, it is not difficult to conclude that min=-4 is the fastest in the case of a decent image quality. So the test with min=-4, and the plot can also consider using min=-3.

The min value is a sample of a large area of flat area, and Max is a sample of the corner detail, and Max's increment and the time are linearly increasing, so we don't repeat the test here. Can be used in the grass relief-4

(No details can be used for 3), and the final plot can be used 0 (-1).

HSPH Subdivs: The higher the value, the slower the speed! Test 20, plot with 50, not more than 80

INTERP samples: The higher the value, the slower the speed! This value is the control of light leakage, the smaller the less leakage. General use 20.

CLR Thresh: System default fluctuation range 0.1-0.3. The smaller the value, the slower the value, and the more easily it leaks.

NRM Thresh: Control of Arc surface and multi-corner surface accuracy, fluctuation range 0.1-0.3, the lower the value of the slower

Dist Thresh: The smaller the value, the faster the light leakage is obvious, the default 0.1, coarse test 0.001-1 to the speed of the impact of a small

Options are generally not checked, the words will slow down the rendering speed.

Second, light Cahe (lighting cache)

Light cache generally as two times rebound, in fact, the VR two rebound absorption is not too good. Therefore, in the premise of basic quality assurance speed is the most important.

Subdivs: The main factors affecting the speed, test with 100, figure 300-500 (darker 500-800). Do not exceed 1000.

Sample Size: This value has little effect on speed, but the smaller the value, the darker the lighter, the larger the leakage is more obvious. General default.

Pre-Filter: Little effect on speed. The larger the value, the more blurred the detail, the smaller the value, the clearer the detail, and the more prone to dark spots. General 10-15

Filter under the Interp

Samples: The speed effect is not small, the larger the value of the more blurred details, the smaller the value of the more clear, easy to produce dark spots. General 5

Just hit a skylight, no optimization parameters, snail machine rendering, put up to mention a wake up

"Turn" http://www.zf3d.com/News.asp?id=2331

Key to improving VR rendering speed