Experimental Cycles render Normal map and displacement map examples

I think I have Blender’s massively awkward node system starting to bleed into my brain

Here is a picture of the brick wall I am experimenting with

Now I have taken this brick wall texture from http://www.cgtextures.com/ ‘cus most of these how do you texture examples always use a brick wall and that nearly always comes from cgtextures

Using GIMP and insane bump plugin (which takes some fiddling with a text editor to get it to work but that's another tale) we can easily generate a range of normal maps and a displacement map

And

So to get any of this to work we must first UV unwrap the thing we want the texture to go on, in this case one side of a wall.

Next for the displacement map we need plenty of mesh so we have to subdivide that wall a good number of times, I chose 30 in this case.

Now comes the tricky part making the nodes.

Blender adds the Diffuse shader and Material Output nodes for you but we must add in a UV Map (under Input…) and set it to the UVmap we just made

Then to each of these jpgs of the wall we have each gets its own image texture node, each one of which is connected to the UV Map node.

Which gives this

The Normal map can either link to the Normal input of the diffuse shader or to the displacement input of the Material Output

Which gives this

Or

The displacement map seems to only go to the displacement input of the Material output node

Which gives this

And as you know doubt have notice each mapping texture I have connected to a Math node (Add –> Convertor –> Math) which just needs changing to multiply so you can adjust the strength of the effect.

POST SCRIPT

Looks like you don’t need to subdivide the mesh for displacement!

And this node layout (I have inserted a neutral diffuser so all we see is the displacement, that's the lower one that is connected to the output)

KDE4 on Ubuntu kills performance

Further test runs (see previous post) have shown a surprising result for KDE4

KDE4 on fedora 20 AMD Athlon 5000 gets about 23.5 seconds a frame with a standard deviation of 8.7 seconds.

Same KDE4 on Ubuntu Studio x64 on the Pentium T4200 pummels performance back to i686 levels of about 40 seconds a frame with a massive standard deviation of 14 seconds

Meanwhile Ubuntu Studio x64 using Xfce on the AMD 5000 gives around 22 seconds average and an standard deviation of around 7seconds ,that is it is pretty much the same as Fedora 20 with KDE4 was! And it had the Matrix OpenGL screensaver running during some of this (its got an NVidia GFX 9800 with the NVidia driver)

Badly tuned KDE4 installs kill performance

Blender: faster on Linux or Windows?

Using the Network render option on Blender version 2.71 I was able to test the same machine on both Linux and Windows, so what was the fastest?

History

I was following this tutorial on making an island environment in Blender and I went down a bit of a rabbit warren when I thought “Hey Blender can work in meters rather than some nameless Blender unit I wonder if I made the island realistically big what would happen”. Well it just took longer to render because now I had to have four ocean squares. As I was in Blender render rather than cycles it was time to break out the network.

Network

My main machine is a Llano A6 3500 (passmark rating 2009) with 16 G RAM, 256G SSD, 1Tb HDD and a 24” widescreen display all running on Windows 7 x64. But over the years I have accumulated other machines:-

• an old XP running desktop with an AMD Athlon 64 x2 4000+ 4Gb RAM  (passmark rating 1285)
• another AMD desktop this time a 5000 Dual core (passmark rating 1228) running 32bit Debian (2GB RAM)
• a Dell Inspiron 1520 with an Intel Celeron M 540 @ 1.86GHz 4GB RAM (passmark rating 542) runs either Windows Vista or Knoppix, in this case Knoppix as Blender refuses to install on it under Vista
• a Samsung laptop with an Intel Core i3-3110M @ 2.40 6GB RAM (passmark 3067) running Windows 8.1 x64
• and finally another Samsung laptop with a Pentium Dual core (Intel Pentium T4200@2GHz passmark 1153, 4GB RAM) running Ubuntu Studio (32bit)

Set up and configuration

Blender’s network render consist of 3 different types of machine:

1. A Master which sends jobs it receives from
2. Clients to
3.  Slaves

Well the useless Celeron was made the Master with all the rest being slaves and the main windows 7 box (with the Llano) also being a client (a box can be more than one type of node on the network)

Results

The passmark ranking order (biggest number = fastest machine) would be

1. Samsung i3
2. Llano
3. Athlon 64 x2
4. Athlon 5000
5. Samsung T4200

Which I know from experience is very wrong!

Ubuntu Studio

The Athlon X2 was so slow as to not make any odds to the results and it kept banging up errors so I removed it, so here are the average run times from 3 initial runs on the Pentium Dual Samsung laptop running Ubuntu Studio (14) 32 bit times in seconds per a frame

Run 1	46.675
Run 2	45.9125
Run 3	45.7625

For an overall average of 46.12

Windows 7 x64

Same machine but this time running windows 7 x64 (I have two different drives which I swap out, so its not dual boot) Same job with 4 other computers on the network

Run 1	37.23333
Run 2	36.34167
Run 3	36.025

Overall average of 36.5333 seconds per a frame, nearly 10 seconds a frame faster than Ubuntu!
A 20% performance gain by changing the OS

Fedora 20 x64

Now lets install a different linux this time a 64bit version, I choose this because there was a promise of a tuned Blender, which did not happen but stock 2.71 running on Fedora 20 on the same laptop (same CPU, same RAM etc)

Run 1	23.43846
Run 2	20.93571
Run 3	21.00769

Overall average of 21.7939
A staggering 14 seconds a frame faster! I make that almost 70% improvement over Ubuntu and about 40% from Windows

Summary

Update from 32 bit to 64 bit as soon as you can

Is Linux slower than windows?

Doing some test with Blender and so far it looks like Redmond the edge…