In the past I was often concerned with the polycount of my static meshes. I always aimed to keep it as low as possible.
Today, I played a bit with the Berlin Flat content from the Market Place.
It uses a 500000+ poly candle holder and some well over 200000+ poly curtains and bedsheets and what not. :o
And it runs absolutely nice and smooth.
So, as long as its not supposed to run on an iphone or under Linux on a Tamagotchi, (and PC is the platform Im aiming for with my projects)… I should be fine with whatever I create.
Im curious. How well does the Berlin Flat perfom on your systems?
Well you would think that and moving polygons has never been a problem for “stand” alone hardwear rendering.
For example a test I did way back when on a 8800
v=IQ_BWblteQU
For visualization not a problem…until you load up a server with players on a network connection and every one starts complaining about frame rates when a 750k player model shows up.
The test for good or bad performance can only occur under practical playing conditions and in some cases server loads. There have been more than a few times that I “thought” such and such addition would not be an issue…until the complaints started
I also get a pretty good performance -> but as it’s just a small level + nearly no actors/particles/other game stuff you can use such a high poly count
I Think that for Solo Games, with no dynamic world generation it’s fine, until it’s multiplayer or if there’s a possibility that a high number of poly appears on the screen
Yeah. I plan for games that are singleplayer. If multiplayer, then with probably 8 players .
I just dont understand enough about network play, so I cant see the difference, performance wise, if in a deathmatch scenario a pawn is controlled by AI, or a remote player; assuming that Bot-AI calculations are not significant in comparioson to the graphical tasks. Can somebody explain that to me
How about mesh particles with 10000 polys per dust particle …?
Issue isn’t about the AI or remote player, issue is about unknown conditions in multiplayer. Meaning when you make a multiplayer that doesn’t have a set number of players. AKA not a deathmatch type where it’s always 8vs8. But a free MMO style like WoW.
You can’t tell if you will have 10 people show up for 20,000 polygons, or 10,000 people show up for 20,000,000 polygons on the screen. 20,000 GPUs snore through, 20million they start to cough a bit. Then you start to add all the scene stuff and dynamic particles that are causing lighting and dynamic lighting to be calculated, the fact those 10,000 people show up and all dump a billion spells / rockets in the area for a few million more polygons, and the GPU starts yelling uncle.
So that is where polygon counts come into play, where you don’t know how many people or AI bots might show up to the party. If you have a static world, with static number of players, you can go much higher. polycounts cause you know you what GPUs can handle so that range is super high, cause it’s a static number, my scene always has X polygons in it and that never changes. With multiplayer that changes to my scene has from 1,000 polygons in it up to billions, and I don’t have a lot of control over that. Without putting things in like queue lines, and instancing of areas / worlds, etc.
Ok, that makes some sense. But i thought this could be dealt with LODs. If there are 10000 players in a game, even onscreen (the king looks down on his crowd), most of them will be far away. They could get lower detail models. The players in the crowd would get high detail models to see for the other players immediately around them and then fall off towards the distance. Like what LODs are for (I thought).
Or make it dependent on the visible actor count. Although I imagine it will be tough to make these transitions unnoticable… For example when I look at two other players in with high poly models and a third player steps into view and all three are “downgraded”…
Well network play has and will always be a mystery but the reasons for keeping low counts is not just about performance alone or counting on the client having the resources in hand as benched-marked under controlled conditions.
Historically all engines could perform beyond the accepted budgets but the reasons for lower than acceptable counts was so not to bloat the release version as to whats needed to fit everything onto a single CD. Granted it might only cost a buck to span to a second but with a million units sold your starting to talk about real money. So if High detail is a habit your game as a deliverable could become so high that it would discourage others from downloading on impulse.
Personally I would love to ignore counts just as much as the next guy but excessive counts leads to serious problems down the road and creates problems for the other guy. Textures and Materials for example depend on shape and form and at 200K your eating into their budget as to things like fill rates and draw calls not to mention UV mapping would be a <inset B word>. Toss on an environment shader kiss your frame rate goodbye.
Graded this is all based on my own experiences, but I’m sure others have theirs, but lets just say as a test, we like testing, that common items don’t need to look 8 bit and that an oil drum should look like and oil drum and a door should have a door knob is acceptable to have enough detail to support the material and textures with out having to look like something from 1990.
On the other hand.
UE4 is the best at environment development through iteration of any engine I’ve worked with and even though a 200k candlestick is excessive it’s inspiring to be able to throw place holders or proxies into the mix just to see the fit or just to please youself.
The fact that UE4 can even render a 200k “stick” is inspiring.
LOD’s help a huge amount but they still come at a cost. If you have 100 people all cross over that “LOD” distance you just slammed a persons GPU with the next level of LOD polycount at the same time, Also means additional drastic change in the drawcall. Also as you point out you really get a distance only option, as you can’t say downgrade Polycounts. Now there have been a few games that coded in auto performance setting changes. This is where it watched on the client what the FPS was client side, when it noticed it was starting to falter it would auto reduce the settings one of which was forcing a whole LOD layer out of the stack to say. It’s a fine line between what your end goal is and how good you want to make it. One thing you will quickly find out about gamers, is we really don’t care “All that much” about graphics, do we want things to look great? YES do we want them to be pretty and cool, yes. Do we really care the guy has well defined knuckles? Or he has an extra bit of definition in his boot? No not really, cause we rarely look at those items if ever. Point is make your polygons count when they need to and let them go when things aren’t really needed. That frees you up to have amazing looking games that don’t cost a lot of polygons to do. For example a low poly count on a candle stick isn’t going to bother a single soul. Especially if it means you used those polygons on that center piece demon statue that is a core part of your game.
I guess that depends also on the personal taste. For example, I really liked the idea when flies start buzzing around corpses in Quake 2, which were pixel sprites at the time.
In Unreal, the flies then were real meshes. And that was one of the (many) wow effects of the game for me back then.
You say people dont notice such details, but some of us do.
How would you then explain the GPU arms race?
To me it seemed that graphics became all that mattered.
How else do you explain the deline of point and click adventures?
