Aggregating Ticks to Manage Scale in Sea of Thieves | Unreal Fest Europe 2019 | Unreal Engine

This Unreal Fest Europe presentation by Rare's Senior Software Engineer Jon Holmes covers the techniques employed to efficiently manage the scale of Ticking Components and Actors within Sea of Thieves while also diving into a few real-world examples of how the team optimised these systems and detailing how it all plugs in neatly inside Unreal Engine 4.
Learn more about Unreal Engine at www.unrealengine.com

[@BenVlodgi]

Loved this! Good balance of examples, info, and reasoning.

[@dyanikoglu]

Had chance to attend this one live, one of the most useful presentations in whole fest :)

[@Jambax]

underrated talk! Learned a lot from this which I can actually apply, GG!

[@yonjuunininjin]

Actually a really good talk! Thanks!

[@calmsh0t]

1) You guys at Rare are geniuses! 2) Epic, make this a standard in the engine please :< I would also love to see further implementation of that system. Like how it was solved for stuff like the movement component, where the manager will end up etc. ... would be lovely to see a live training introducing this method in the future

[@lifeartstudios6207]

movement is one of my hugest hurdles right now. very curious on how world war z handled their ai

[@Internetzspacezshipz]

Very interesting talk. Extremely informative and useful.

[@abdullahamrsobh]

I like how much effort he put on these presentation slides xD

[@jameskoss]

The real issue, more than the manual labor and difficult maintenance, is the complexity of the code. The more machine friendly it is, the less human friendly it becomes! Great work :D

[@vegitoblue2187]

I still don't get the prioritization part. I understand how they reduce tick rates at a distance but how would this look in code and what potential components could this be applied to?(AI, Character Movement, etc.)

[@zeon3d755]

Anyone tried this round robin scheduling and ticking in UE4? Any examples or tutorial or video?

[@zeon3d755]

Let's say you turn off all the ticks then use multiple timer functions to do the work of ticks.

[@Deck_Dynasty]

The main takeaway from this video is: You can immensely save on cpu budget by turning off tick on all actors and components and instead registering similar ones to arrays in a service which manually calls tick on them by your own prioritization rules. This is not because we are reducing the overall workload by changing what's in the tick functions or making fewer calls (in fact, we're increasing it). It's because the structure and pipeline of cpu caches are more conducive with batching.

[@alexandrm3073]

Will you be so kind to make an example with pseudo code and/or diagram? It would really help me to understand your point.

[@Deck_Dynasty]

@@alexandrm3073 sure man, no problem! It's a word diagram which goes like this: If you're too dumb or lazy to understand what I said or watch the video, consider another vocation. Perhaps custodial services or president of the United States.

[@alexandrm3073]

@@Deck_Dynasty First of all English is not my native language, so I didn't understand well enough. Second, when I'm elected as a president of the US this November, such rude comments will lead to the death penalty.

[@vegitoblue2187]

so where is this aggregation done? I mean what component when this finally reaches blueprint

[@masterneme]

Isn't this similar to what Unity is doing with ECS/DOTS?

[@calmsh0t]

Not too much into Unity, but if it is then we have a good chance of epic bringing this system to unreal :p

[@tom.looman]

Yes it is indeed. A big gain of Unity's ECS relies on the cache performance. It's good to see that Rare used it without seemingly having to re-design the standard Unreal way too much.

[@masterneme]

@@tom.looman I hope we see more stuff like this in more detail 😉😉

[@calmsh0t]

@@tom.looman Do you have any idea how you would aggregate something as dynamic as the movement component (since they listed this), where n objects are not depending on one other object, like it is the case with the shown water example? The only idea that came to my mind is caching all controllers and looping through them, but it doesn't look to me like this would result in a noteworthy performance gain. Would love to see one of your great videos tackling that topic, really liked the one showing the gameplay ability system btw :)

[@JonathonMHolmes]

Hi @@calmsh0t, that's a great question. Caching all movement components and looping through them manually can have a surprising perf improvement. If you consider the movement component instructions cost is well over 30% of your instruction cache (as I mention in the talk) making sure you do them all when the instructions are hot in the cache is important for cache coherency in a single frame. There's also a surprising knock on improvement from fewer ticks to queue at the start of your frame (especially if you're talking 10's of characters all moving at once). That's only the start of it though. As with most things we've optimised, aggregating is just the first step. Once you have a list of things to transform it's now possible to change your algorithms and get the real speed ups (as per my example with the water heights). Granted that is tricky with the movement components because they're so complicated, but we haven't ruled it out yet! :)

[@RR-kn1tg]

Parent calls childrens tick. problem is solved

[@vosdraug4628]

Funny how this is just a basic ECS method.

[@eusuntflorin6984]

I love epic games end fortnite the best

[@neontriangle274]

we don't want more content right now, we want more player capacity and fixed server issues, content can come after yeah? make the game playable yeah? maybe that's an idea yeah? don't make me daily for an hour and not find a single person yeah?

[@kyoai]

Yeah?

[@Unforseenak]

Sea of snores.