NVIDIA, AMD, Intel Explain How OpenGL Can Unlock 15x Performance Gains

Hmm even if they cant get this Great Performance with OpenGL support then, they / you uber still want to hit the technical goal of 1 million units in a game? I think this is a good number, performance will be great hopefully ...

It's really wrong question in topic about renderer. When you're not actually see all those units they don't affect rendering performance and I doubt @neutrino ever meant that you'll able to see all units in same time.

I believe he's alluding to the simulation handling 1mil (ie. if they can get there with the current overheads from rendering), though yeah it's kind of a separate problem. I kind of imagine the server side will crumble long before the client side does though (with everyone having multicore processors these days; albeit I don't believe the game actually takes advantage of more then one core at the moment).

Both server and client use advantage of multiple cores when generate planets, but yeah after alpha started and @neutino said about "one server = one core" nobody of developers said anything new on how server-side works at moment.

Anyway last time there was actual discussion about simulation performance (don't have link unfortunately) there was notice that multi-planet gameplay give PA great advantage here. Real multi threading of some tasks in simulation is hard and expensive to implement it's much easier to run simulation of different planets across multiple cores.

Rendering 1 million units isn't really the goal, but rather games with 1 million units in them. At 1920x1080 resolution there's only just over 2 million pixels, and it'd be a bit silly to render a million units when they're on average only two pixels wide. At that distance they'd be strategic icons anyways if they're anything.

And yes, trees, as well as smaller rocks, and metal bits on metal planets, and metal spots, etc. all turn in to 6 polys when you zoom out far enough. Here's an image showing some trees roughly at the point they switch from their normal models to the generated LOD models.


We've discussed the possibility of rendering units in the same fashion, but it's a bit more complex to handle than a tree or a rock. Trees are single "things" that when they rotate the entire object rotates, completely rigid. A unit, like a tank, it can move it's main body in one way, the turret in another, and the barrel in yet another. Do we render them at just a static pose, or do we render out the body and turret & barrel as two separate parts so they can turn, or one per each moving part? It's not that bad for a tank, but for a more complex unit like a bot with a lot more moving parts?

It might be just the simplest version with units rendered in their base idle pose is enough, but is there a reason to render them in the first place? On the planet you're focused on they'll be displayed as strategic icons already, so why bother? On planets not your focus should we render units at all?

Render them in some form. It certainly looks epic when you have a teleporter on two planets and can see both simultaneously in freecam. You see the bots enter the one closest to you, and on a distant planet you can see the dark grey flood forth from their teleporter in a really sweet style. Don't compromise that.

Brilliant involvement from Uber in this thread. I was initially posting to say "man comment trains on articles like this get retarded fast" (so much MS hate), but damn if I didn't appreciate some of the work Uber have done on PA in the process

EDIT: my vote is render them!

I'm one of guys who use range dependent "show terrestrial" with "orbibal shell" enabled options so when I go back to orbital layer there is no icons for terrestrial.

From my point of view "icon wars" it's worse that happen with SupCom.

I sure there is reason to render stuff on different planets in some form.

When you see moon (or main planet) in the background and there is stuff blowing up it's actually looks epic.

I actually sometimes go a step further and completely disable strategic icons when playing vs the AI just to appreciate the beauty of the game. It is so much more nice without strategic icons - they were the main reason I never really played SupCom.
So I think we need two things:
- Units should always be rendered, also on different planets that are not in focus but visible (even if handled with an LOD system)
- We should be able to specify a height at which strategic icons appear (they appear way too early on small planets) and how transparent they are

I want to play the game with a few strategic icons as possible.

Totally support this.

If some players want to play icons war then it's will be much better if they'll get "thin client" option for game. I doubt that people who prefer icon wars for better performance actually want hundreds megabytes of surface textures or some rocks models.

I actually play with strategic icons at a minimum. I forget if they're actually off or set to really high zoom. Frankly it's easier to see units that way and judge the size of a force. And of course it's nice to actually see the graphics as they really are.

That said I do play fairly close to the ground and am the type that would also appreciate some different camera positions as well; especially for some of the higher altitude zoom cameras which currently seem to just go towards a boring top-down view.

When will we be able to actually play million unit matches? (rhetorical question, the answer is "soon")
How about: will we be able to play in million unit matches at release?

How soon after release will you start adding features from the newer versions of OpenGL? (immediately? half a year? a year? 2 years?)

How about, is there even a point to playing a milion unit matches?

Sure it might be fun the first time but that's probably it.

Dont think a million units will ever happen, its more like 75000-100000 units.

That was a technical goal...

Yes I know, but there seem to be some discussion about playing with a million units.

I really don't see it ever getting anywhere close to that. I think we could reach games with tens of thousands, but hundreds of thousands?

Why no? In theory servers can have up to 4 sockets and each CPU might have up to 15 low-end cores. If each core can efficiently handle simulation of planet with 5,000 units then we can easily see 200,000 units in one game.

In reality server on EC2 with 32vCPU and 60GB RAM will cost like $2.4/hour, but you can only pay for 4-5 hours.
So it's totally possible and CPU time goes cheaper over time so...

In reality now there is number of problems that slow down simulation a lot and I'm made topic about them:
https://forums.uberent.com/threads/just-notice-about-physics-in-simulation-performance.58214/
Still if Uber can fix them performance of games will be improved a lot.