Blender Internal and Cycles render on the Raspberry Pi

Hello, this is my first post on this forum and I’m about a year very busy with blender and animation.:cool:
I also really like ARM and Embedded/Single Board Computers and now I just happened to be using a Raspberry Pi 2B for school/simple work. It isn’t fast, not slow, but I really didn’t think I could possibly render with it.
Well, of course I tried to install blender, (I didn’t want to compile so I used 2.72 from the Debian Repository) and tried to start it. It really is not possible to work in this way, the GUI is very, very slow. But when I tried to render, it went actually pretty fast, not very fast but for a 4watt CPU it was top-speed; 1s for the standard scene in Blender Internal, so I tried something from the internet and that was done in 3 minutes in Blender Internal. So I tried it with the command-line and it was twice as fast:eek:.
This is the file: test.blend (304 KB)


I was stunned by this:spin:, because it is such a powerless device. (I’m not a computer scientist, but it must have to do with the CPU-architecture that makes it well suited for rendering)
I think that if you make a cluster of overclocked Pi3s and using it to render, it’ll be a great renderfarm.

I posted this merely to ask you to make an animation and try to render it on a Raspberry Pi and share the results.

I used a Raspberry Pi 2B 1000Mhz 4-cores Command Line Interface.
When Blender 2.78 is released I really have or someone else has to compile it on raspberry Pi 3B.

I haven’t yet bought a Raspberry Pi 3B but I’ll surely do and I’m going to try to render other self-made or blends from others and Cycles too.

Greetings Animajosser:D

Edit: Just bought one


This is a less pleasing result, but still it’s possible to render on the pi.
I took tile-size 128x256, not very low, because this ARM cpu seems to act a bit like a gpu.
for comparison to my Hyperthreaded Xeon 5650 (6 cores and 12 threads) 16gb workstation, which did it in 11:35 minutes. The difference is around 23x and in the test above was the difference around 15x.
But I still have to find the ultimate tile-size compile the newest version on a pi and use the pi3, so I can get the best results.

Greetings Animajosser

This is really impressive! I have never heard of these raspberry mini computers before. I used to program with PIC controllers, where I was also surprised of the potential power of those micro controllers.

Without having any real technical knowledge of renderfarms:
I think its a nice fun project to build a renderfarm out of mini computers. But if you consider the commercial point of view, i would guess that buying a bunch of GPUs is more efficient.

Very interesting. To compete with Xeon, there should be about 40+ raspberries. That’s a big pile, but would it require less energy? And how much it would cost compared to Xeon.

To consider
Small amount of slow RAM per board
Total power consumption, total for each board plus additional cooling as they would be running at 100% when rendering
The hassle to put together and manage.

These cheap board computers on not really cost effective for a render farm, a basic computer and gpu will be much more flexible and cheaper.
More info http://raspberrypi.stackexchange.com/questions/50351/does-raspberry-pi-cluster-make-sense-for-a-small-rendering-farm

I understood that this is not the traditional “easiest and cheapest allSolution” option. Rather, is it possible and with what contraints?

Yes, less memory, but as seen, bmw is rendered, it is not impossible to render certain projects.
And it seems the power consumption is not a problem even with full load: http://raspberrypi.stackexchange.com/questions/43285/raspberry-pi-3-vs-pi-2-power-consumption-and-heat-dissipation

The “hassle to put together and manage” is part of the hobby I think.

If somebody has 300€ or more to put into rendering, he/she should consider gpu, or maybe even commercial render-farm solution, instead of 6 raspberries. But if somebody want’s a tinkering hobby, and have some rendering magic with it, the Raspberry is very exciting possibility.

Ten million Raspberries sold: https://www.raspberrypi.org/blog/the-impact-of-ten-million/

I find it amazing that the PI can even run Blender well enough to actually render something (though with 2.8 bumping the OpenGL specs. I’m not sure if it will still apply in the future).

It’s not bad for such a cheap machine, but it’s definitely no replacement for a proper PC.

I agree, it’s not a replacement for a good computer, but for writing and simple web-surfing it’s enough and it use way less energy so then it’s a perfect replacement.
About the heat, the raspberry pi 3b needs a little heatsink, when using extensively, but it doesn’t use that much energy in the end.
I was trying this merely for fun, Although there is also the Pi Zero which I want to try, but I think It isn’t the best way.:stuck_out_tongue:
GPU’s have fast, but little memory, so I think I would go for CPU’s. The best thing should be to make special render-GPU-computers or, if using Pi’s, only use the CPU’s and get much, fast RAM.

But it’s fun rendering this way.

I got a Raspberry Pi 3B and it’s a lot faster than the 2B. This is my last test with tile-size 128x128:



This pleases me more: 16 times slower than my Xeon 5650.

I also tried to compile Blender 2.78 on the Raspberry Pi. Unfortunately I was not able to compile OSL and when I try to render bmw27 with it I get a head-aching Bus Error, so I thought I should quit and wait 'till someone else finds a solution for that.
Although It does work with simpler models, I think it isn’t very usefull, but I still uploaded it to Sourceforge for those that want to try it.

The Raspberry is starting to look more impressive now, I could see them wiping out the market for Intel’s ultra-low end Celeron processors or even their Pentium class processors (good riddance too).

The bargain basement (as far as computing goes) looks be moving beyond the under-powered machines of yore (that can barely run anything). I do like seeing just how fast the low-end is advancing when it comes to processing power :slight_smile:

Yea, as far as I understand; ARM has been developing new instruction sets and x86 is mostly the same and so it’s old and not power efficient. Also ARM focuses on being power-efficient and x86 on speed and power. I saw some good low-power Intel CPU’s, but don’t know how good and how much power they consume. The CPU in the Raspberry Pi is clocked very low, that’s why it consumes so little power. ARM already rules the mobile-phone market. But for software, when designing it smartly, you can do a lot with less powerful hardware and even make it fast, when using a low power GPU you can make things way faster because it was designed especially for processing images. I think that’s why the Raspberry Pi Is able to render that well, for his power; I think that the technology for the CPU is a bit like that of a GPU.

hi,
Blender 2.78 seems to be included in the Arch Linux arm repository. I have a Raspi 2 with Arch and LXDE desktop. I will give it a try for rendering.

Cool, I didn’t know about that. Maybe I should try arch for a time, the Debian repository is pretty outdated for some software. Although I could get it to work on Debian, but the dependencies keep breaking important packages. But nice:yes:. I’ll shut down my compilation I think, later, unless I can get it to work out of the box.

Honestly, current high end GPU’s are like many RBP’s, but all connected in one device with a high speed reusable memory (between cores) and far more power efficiency, and better performance vs $$.

1 GTX 1080 only needs 8GB memory, 100 RBP3’s would take 8GB x100 to store the same thing, cost a lot more, use more power and require extensive networking hardware to connect.

Energy wise, except whilst gaming/rendering, my PC with an i7/SSD’s etc uses less power than the monitor I use does to display the image.

Honestly, I am quite interested to see when updated versions of systems like Cavum’s 48 Core Arm CPU, looked at in this article here:

Note, in the use case test for these systems, the Xeon CPU’s still beat the ARM CPU’s in power efficiency vs performance. There’s a lot more to the performance vs power of ARM/X86, and overall X86 generally comes out on top (for high performance usages, which is what I assume we are discussing here). For now.

Interesting stuff you have there about the GPU’s I didn’t know that you could store render-information in a different way for a GPU, but also have to admit that I just saw GPU’s with 12gb Memory. I really do believe GPU’s must be used for rendering movies and other animation, because they indeed are power-efficient. But CPU’s are very difficult to compare, because CPU 1 is better at rendering and CPU 2 better at math (just an example). I think arm does already beat x86 in power usage vs performance, because ARM (e.g. cortex A72) is becoming pretty powerful (not nearly as good as new Intel CPU’s of course, but already better than old ones) and uses very very little power. Maybe some very powerful CPU’s do beat arm at power usage vs performance, but they use very much power. But I wouldn’t use Raspberry Pis to render for real, would take too long. But If I want to render a picture of a project, I would really consider doing it a night on the Pi. But this ain’t a plea for using it at businesses and the like.:slight_smile:

I think that would be interesting (for testing purpouses) make a cloud distributed render farm like Sheep It, but for small arm devices. But i don’t know how to do it.

Thea render can run on a Pie and make use of the GPU. Pies are cheap low power consumption and scale nicely. With Thea those are sweet render farm computers one could set up.

As far as I can see does Thea render use Their Presto engine, which does only support NVIDIA CUDA GPU’s. The Raspberry Pi has a Broadcom VideoCore IV. So when starting the program it will only use the CPU, it says. But it still is very nice they support rendering on the raspberry pi/ on raspberry Pi’s.

That’s great for rendering but does anyone have any experience getting the interface to run?

Yes; Just put on the experimental OpenGL driver for the raspberry pi and it will work like a charm, although it does run without it (not like a charm:no:).