[Visit the original post at Unity Performance: CPU Time Slicing]
The common wisdom says "profile, optimize, repeat". Let me warn you: this won't always work. Optimizing low-hanging fruits can gain you a whole millisecond in a day. But when those are gone, speeds of 0.1 milliseconds gains per week can become normal. When that happens, you need other tools such as the Unity CPU Slicing technique.
Interlaced-Logic-Thumbnail
In this blog post, you'll learn how to apply:
Update Batching: reduce the costly overhead of Unity calling your Update functions
CPU Slicing: Split your CPU load across multiple frames to multiply your scripts' performance
Quick Navigation
Unity Performance: What's CPU Time Slicing?
How To Apply CPU Time Slicing In 3 Steps
2. Link Your Scripts to the UpdateManager
How Did I Get Into This Mess?
2019 was a tough year for me.
I was in charge of porting a high-demanding PC/PS4 VR title to Oculus Quest, a mobile VR platform.
It was a daunting task for many reasons... and performance was surely the toughest objective to accomplish.
So I had a few months to make the rendering of each frame take less than 13 milliseconds. If you're a veteran, you know that doesn't really give you much room for fancy 3d graphics and complex gameplay.
And 13 milliseconds was a very distant goal from the 40+ milliseconds I started with.
At the beginning of the project, I was confident I'd pull it off. After all, I had done huge mobile optimizations in the past.
Weeks passed by optimizing every single aspect of the game I could think of. Days where I saved a whole millisecond were kind of common. And those days brought me closer to my objective relatively fast.
However, by the middle of the project... I wasn't that confident anymore.
After optimizing all the low-hanging fruits, the CPU performance gains were becoming much rarer. I was used to gaining whole milliseconds within single days, but my speed dropped to the painful figures of 0.1-millisecond gains per day — if lucky.
I remember how excited I got when I jobified the audio system we used to update 200+ audio sources. I was really proud. But after careful examination, that optimization only got me 0.3 ms ahead but stole about 3 days of my budget.
... And I still had 3 milliseconds to go with little time left.
That was really a big problem considering the optimization speed I dropped to.
Draw calls were fine. Physics were also very optimized for this type of game. And the whole game logic was already pretty well optimized and partly multithreaded.
I had just a few weeks left and I didn't have any idea on how to approach the situation... and I knew the traditional wisdom of "profile game, optimize script, repeat" wouldn't get me there in time.
If you know me, you can probably guess what I would do in these type of situations...
Radical solutions.
I opened my notebook and started an unusual brainstorming session that would end up with a crazy idea.
As I re-read the Oculus Quest guidelines, I saw that indeed I had to render the game at 72 FPS.
But this is the key I realized back at that point: rendering at 72 FPS doesn't mean you must execute everything at 72 FPS. In fact, physics already execute at a different pace.
So I asked myself: what if I run the logic at lower framerate?
I quickly stood up from my seat, grabbed a sugarless double espresso and went straight to the drawing board.
At the drawing board, I started by making an inventory of all the expensive gameplay functions (funny how code that stole milliseconds from a CPU would steal hours from my sleep in the previous months).
Cactuar
The "Cactuar" Performance Group
Then I divided the gameplay functions into three groups:
Cactuar group: thousands of inexpensive scripts that, when combined, created a terrifying panorama. The name comes from the deadly 1000 Needles attack from Cactuar in Final Fantasy, each needle making just 1 point of damage.
Serious troublemakers: about ten scripts, each taking an average of 0.1 millisecond per frame.
Final bosses: massive monster scripts you wouldn't want to mess with. Each took about 0.5 milliseconds.
Was making these groups critical?
Probably not, but I had fun doing it... and it helped me with the next step.
Script groups in sight, I then re-arranged the scripts into two new separate groups that I called Group Alpha and Group Beta. I moved them around with one goal in mind: to make each group take about the same time to execute, i.e. 1.5 milliseconds per group.
The idea was simple: to execute the logic of Group Alpha in frame 1 and the logic of Group Beta in frame 2. And then I just had to repeat the cycle to literally halve the per-frame CPU cost of my scripts.
Within an hour I implemented a prototype to split the execution of my logic into these two groups.
Naturally, the next step was to test it. As soon as I went into gameplay I saw the game breaking into pieces from 8 different angles. Even Unity crashed (but that was nothing new).
Another sugarless double espresso gave me the well-deserved caffeine kick that helped me tweaking my scripts. I had to make them less CPU-attention sensitive. They'd need to let go of all the CPU love they used to get each frame to be content with half of it.
After some time and tears, I got it all to work.
I reduced the CPU time I spent each frame executing scripts by using what I call logic interlaced execution. I think the whole internet calls it CPU Time Slicing... so I guess I'll stick to that name.
Traditional Execution vs. CPU Slicing
Traditional Execution vs. CPU Slicing
Slow down, though... this system comes with its side effects.
I excluded scripts driving noticeable visual elements, as alternating their execution would make the game kind of jittery. I found it funny but my client wasn't that enthusiastic about these.
The benefit of this system is that eventually I could add a third, fourth, fifth group depending on time-critical the scripts were.
Also, by using a centralized update manager I got rid of the overhead that comes with having too many Update functions in Unity.
This system helped me get the last few gains that I so desperately needed.
Back to you... let me share with you the trick about implementing this powerful strategy in your game.
Unity CPU Slicing: Setup
Unity CPU Slicing: Setup
Unity Performance: What's CPU Time Slicing?
If you read my story, you might have a vague idea about the meaning of Unity CPU Time Slicing. But just in case, I'll share an analogy that I like with you.
