Important things when using render farms: whether your final result is rendered correctly?

In this post let’s talk about the important things when using render farms. Some of you could argue that cost is the most important factor. The lower the price, the better. Some others may think time is more important. The faster render time, the better. Me personally, they are all crucial, but what matters the most is the problem related to my final result, whether it’s wrong or right. So it leads to one concern whether the render farm is able to render the project correctly.

The render farm could easily render your project. The problem is if you miss any textures or assets, will they know it and notify you? Or they will continue to render and what you receive later is something you cannot use?

To differentiate this problem, I have to divide it into two cases. In there, I will help you to understand what could affect your quality in the process of rendering on farms and why it could turn out wrong. It took me many hours to investigate and come to a conclusion. 🙂

The two most famous ways to use render farms now are IaaS and SaaS. If any of you don’t know what they are, here are some reference links from vfxrendering.com site and radarrender.com. I know you are lazy, so I will not post much here but send you some links to read later if you have time. 

Okay! Now we come to the two parts I said above. Let’s talk about the factors that affect the quality of rendering and how it could render wrong or right. You always should look into the hardware and farm manager.

In a SaaS type of render farm, sometimes the hardware could be not the same, and they only have one farm manager and you are not the person who manages it. For example, they have your files distributed by one farm manager to many nodes. Their nodes could be the same (like same GPU, same series, same manufacturer, etc.), or could be different. The worst case will be different hardware, where your file could easily be rendered wrong. 

I asked for permission from vfxrendering.com to use their video and image they have rendered on SaaS farms, and got approved. Let me show you the wrong rendered image. (I did a quick crappy grid layout to demonstrate what I will explain later, so just ignore the logo of vfxrendering and all the other)

So you see that SaaS farms rendered the image wrong (wrong color, not the grid lines I did). The object on the left (the moon?), the sand, the rocks (or big jewels), the armor of the guy all are wrong in color. That’s the weakness or the risk you have to take when you use a SaaS type of render farm. You don’t know what GPU nodes they will use. 

Just imagine they distribute 12 nodes to render your images, each node contains one GPU and will render each part of the image (so in total 12 nodes render 12 parts at the same time and combine them together, that’s why I did divide the image above into 12 parts to demonstrate my example). They GPU did render each part separately, but they still needed to load all the data of the project to be able to render. In this case, it seems like the system cannot detect what is wrong (or missing) with the render settings, that is why they rendered wrong.

In an IaaS type of render farm, you can easily track the specification of the GPU you are using. So once you can render and it turns out correct, the whole project will be correct. The sand is light brown, more like a beige color (not in snow white); the moon (?) is in grey, the rocks and the armor are all correct.

And moreover, you are actually the farm manager yourself (you can install thinkbox deadline and distribute the renders by yourself). Or the easier way is that you can work directly with the GPUs without the farm manager tools. Let’s take Blender and Cycles as the example above. You rent an instance of multiple GPUs on IaaS render farms, and are able to work with all the GPUs of the instance itself. It mitigates the potential errors when working through a farm manager like SaaS type of render farms.

When you build a render farm on your own, you pay attention to the hardware. That seems not to be the case when you use a render farm. Like I said above, some of you will think about the cost or render time more, and assume that the result turns out like what you want. In fact, if you are not sure about the GPU used in the project, it’s likely that your result will not turn out as expected.

In this post, I only discuss GPU rendering. VFXrenderfarm will explore CPU rendering later. And for GPU rendering, I came to the conclusion that if I want my project to turn out correct, IaaS would be my choice. I’m not so sure about SaaS.