My favorite part is definitely when he gets down to L1 cache amount and explains the exact amount of memory needed. It appears a Unity transform takes 36 bytes (1 float = 4 bytes, position, rotation and scale take 3 floats each, so 12 x 3 = 36) of memory in L1 cache. (not that it matters, since it takes up the entire L1 line of 64). With ECS, you calculate take what you need/want to change. in 3d games that is usually only 3 floats for position (xyz) and a rotation around Y axis, for a total of 16 bytes. (he is only using 12 bytes because his game is 2d, so he also does not need the y position). Things like scale rarely change, so you can have varying scales on your entities, but that doesn't matter, because you aren't changing them, so set up your system to only calling the entities xyz positions and their y rotation. Yeah, I think that's when the lightbulb on this ECS thing finally went on. Thanks Simon Eliasson.
Excellent presentation. I can't believe that this was his first time presenting. Very calm and collected. Didn't rush or stutter. And they packed in so much information in a fairly short time. I haven't even watched the whole video, but I'm just very impressed. Bra jobba!
Right!? Is his audience asleep? I get that things like this are often contrived examples, and 2250x performance gain would not represent any actual real-world results. But still! If all of this could realistically give a simple 10x performance boost over the traditional methods in Unity and effectively use all the compute power in a high-end multicore device...it could totally transform Unity. His audience must be asleep. *yawn*
@@tiredlocke this is called respect, they wait for the presenter to finish his presentation before applauding in Europe interrupting someone who presents his work is seen as disrespect
@@Hoowwwww Lol. Oh, right. Because people in "Europe" are homogeneous and do everything the same way, and they are so different from Americans. All Europeans are just so classy and respectful.
This makes a lot of sense - one of my first questions that came to mind when I started game development was about the overhead of each of the "on_update/_process/_physics_process" call added per game object.
there was mentioning that the pathfinding is done on server side, so the client just receives a current and.a target position for the enemies and interpolates the position between them. no need for pathfinding with ECS in this case.
The 27:40 reveal made me gasp. I'm trying to learn DOTS asap. Any game build using classic OOP in a few years will feel sluggish and unresponsive compared to what a DOTS-based application can run at.
Chris -0 also you’re assuming the application uses heavy gpu work. Some simulation games don’t need much graphics work at all but invest heavily into simulating thousands of independent objects interacting. Also some machines, such as most MacBooks, don’t even have GPUs, they just use CPUs to compute graphics.
not just performance, even if it is elephant in the room. I actually like the way of writing the code. it is much more intuitive than oop (with games this is case, not sure about softwares) if you have right mindset. want something to behave that way?add that component, which holds data to achieve the behavior. system will do the work.
awesome. I've been examining the use of DOTS for a mount and blade / Total war style hack and slash...though I'm having some trouble with wrapping y head around the animation component implementation as well as how to get the characters to attack preferred targets with an equiped weapon...
Nice presentation. Just.. I was a little confused when he was talking about CPU caches and 3 floats and in the picture there are 12b (bits). Images are wrong. b = bit, B = byte. He is talking about bytes.
One of the ways ECS gets its performance benefit is from memory alignment and data locality. As C# classes are reference types, they can each be stored anywhere in memory. C# structs, however, are value types and due to this when they are stored in linear collections they are tightly packed and located next to each other in memory. This allows the hardware to pre-fetch entire buffers of data at once into a higher order of memory built into the CPU for quicker data interaction. There are other things DOTS does such as avoiding lots of virtual functions but data locality is definitely one of the big ones I would say.
@@kayomn I'm aware of that - I was mostly talking about the first half of the talk where he mentions the performance of using Instantiate to make his trees, which is obviously a bad idea. I'd be interested to see a comparison between the ecs and a more traditional pool. I'm still blown away by the burst compiler in the second half though.
I think he's talking about per frame update performance, not creating objects. So pooling won't work here cause it's totally about looping all these objects.
The hillarious thing is they could just take the n of enemies down and have destructable enviroment Have interesting mods that could make the guns behave more like weird poe or diablo skill manipulators. And have the same performance maybe? Like a shotgun whos bullets explode at the end of their range. Or maybe a lazer that leaves a trace of fire a gun that fires guns that explode when empty crazy things
This is about DOTS and a new, thousand-fold improvement in code execution time kenny. No one cares about the pedantic details of game design terminology at the moment.
Bird's eye view, satellite view, side view, or over the shoulder, is all "third person" The location of the camera is not inherent in the concept, just that the camera is not the perspective of the character being controlled. You are likely thinking of over the shoulder, which is a specific third-person perspective, but not the only one.
First person means from the perspective of the person. Third person means from an outside perspective. That includes birds eye view, it's a third party view, as if you were a bird watching.
The moment when you realize that the fancy sh*t everyone talks about is something good programmers do since years.. lol. I guess the world would be a better place when people would start learning programming in assembler and not unity's gameobject/component system.
This talk it's baseline good. But it doesn't say anything new or that we couldn't very easily abstract from other more 5 or 10 official talks simply about the same theme: "our implementation of the ecs system" others have even been more in depth. I would prefer more talks about syntax and hands on tutorial that you can follow coding along the presenter. I applaud every new ecs related video, but the title it's very misleading. Like a vulgar RU-vid tutorial with very low interest in actually sharing the knowledge but only interested on mechanically completing the task. It's should be more like "how we gained more performance with our dots implementation". Three title make it look as a tutorial, but they deceptively put it knowing that ambiguity. It just makes me angry, this presentation could have been much better, I can see that the presenter is not lacking in ability. It just feels like he's morally down for some reason. *Sigh* I'm just expecting I don't know.. like lucas meijer coming up and talking about his new anti boiler plate way of writing the dots way. They guy could have posted the video like on RU-vid on his personal channel. But I don't think what he showed and how he showed isn't on part on what one should expect on the unity channel. After everything I consider just something positive unity opens his doors to people to talk about ecs, but I would expect for him to put more soul into the damn thing. One thing which is due it's his page about memory chunks and I can see he has a real compromise with understating what's he's using and it's one of the downfalls of him in this presentation.. I feel he let him that take him away from his main plot line. He kinda dropped down the ball mixing the two senses; 1 of understating the architecture and two of making the game. This guy could have been more better, but I understand it's actually, some times really hard to be soo good. I'm, you know, just kinda mad he had the ball !! wish him success with his game and he better gets his presentation better the next time ! >:). He should've taken more risks explaining things in a different way and talking more about things he doesn't yet not understand completely. even if they could have make him look dumber. You know maybe I should check that discord.. And I don't like having to use discord. When I get to understand this performance system, I'll make my OWN RU-vid video and it's going to have quirky jobs as genius development, along with heroic architectural archivements. And in going to be cool and oh soo cool about it. You'd wish you be me! And my hand it's going to be super fun, and the story it's gonna make you cry but you'd wish you live longer to experience this better view in life you gained playing my game. The best, man. I tell you!. And it's going to have a thousand likes. Just so you know.
