A modern asset pipeline: How much time is spent optimizing?

This is the third post in a series about a modern asset pipeline. Here are the previous two posts:

Introduction

7 reasons to optimize content

Background

Over the last generations of hardware asset complexity has increased a lot. Geometry, shaders and skinning are increasingly complicated due to more computing power. There is a linear correlation between the complexity of an asset and the amount of time that has to be spent on optimizing it. Optimizing a rock is rapid task, whereas making a low-res version of a million polygon dragon is a significantly more time-consuming effort. Hence the time spent on optimizing assets has gone up by a lot if we assume approximately the same amount of assets are in a game today as in games in past generations of hardware.

Team sizes and game budgets are getting extremely high. There are cases of 100MUSD+ production budgets and teams of upwards of a thousand people working on a project at peak times. I would like to stick out my neck saying that we are nearing a peak of what is feasible. With this in mind we need to be smart about our usage of time and money, even a small percentage improvement in workflows is a significant saving in time and money when looking at a full production.

Combining the rise of asset complexity with the peaking of budgets and teams sizes it should be quite easy to deduct that automating optimization work will be hugely beneficial. However, we need to strengthen the case.

Stats

Prior to writing this article I did some non-scientific research, talking to number of peers in the industry trying to put some figures together. To be honest, my stats are coming from a rather small number of places, so it’s in no way a complete dataset but it should at least give an indication.

Firstly I wanted to find out was a rough estimate of how much time was spent on asset creation in large scale productions today. The answers I got back were divided into two ranges, 30-40% and 60-70%. The differentiator between the ranges were if the studio was using in-house tech (30-40%) or using an external engine (60-70%). No matter what, a significant portion of the budget is spent on asset production. A small improvement in the workflows would have a great impact.

How much time are spent on optimizing content?

Here are some stats that was shared by a medium sized AAA game.

According to these stats about 26 man months were spent on optimizing art assets in the game.

Here is another set of stats that was shared to me from a large AAA game.

In this case they had done some research into how many times an art asset was touched during a production. This turned out to be on average 3.6 times. It’s hard to estimate how much time will be spent on optimizing the asset for different use cases as they vary so much. If we make a crude (and probably low assumption) that about 1 hour is spent every time, we end up with a figure of around 110 man months of optimization work in a large project like this.

Another random stat that I heard from a techincal artist in my research is that according to his findings roughly 30 % of an artist time is spent on optimizing content.

Finally, in our backend we can see how many optimizations are run over a certain time period. Here is a sample from a couple of productions:

If we take this numbers and make some crude assumptions. Each optimization run saves 10 minutes of work and a project spans around 200 working days, we end up with these stats:

Which means that on average there are about 40 man months of optimization work in a two year AAA project.

As I stated earlier in this discussion, this in no way a scientific approach but the different stats all points to the amount of work spent on optimization in an AAA project is in the range 20-100+ man months. Which is a significant amount of time.

In the next post will look into what can be done to reduce this.