Reading through the post, what do you think the actual battle size on screen will be compared to the original? I know we had some discussion that just kind of fizzled as to why 3-D was not necessarily slower than 2-D, but I'm looking at the numbers you posted here, and it's a fairly significant difference.
It's not really much difference in terms of the earliest fights in the game; honestly in my opinion these feel slightly bigger, and that's without the ships properly zooming around in their squads.
I'm talking about north of 60fps on a waaaay old machine on max settings on a pretty sizeable resolution on a goodly-sized-but-not-remotely-largest battle, and I was surprised by that result. That combination of things was not remotely something I expected would work (max settings plus that resolution plus etc etc etc).
As for the actual upper bound of things, I just really don't have a sense for that yet. From a computational standpoint, this simulation from every angle should be able to kick the butt of AI War Classic. From a network performance standpoint, the same. If you're thinking of "far zoom only" and we're talking only icons and shots and explosions on the screen, then again this would blow the former out of the water.
When we start talking about a mix of things where you see lots of large ships, and icons for further-away ships, and it's fully animating all the squads, and so on and so forth... it's really going to depend on the hardware. Past some certain level of hardware, the AIW2 approach to graphics starts winning out and is future-proof, whereas the Classic approach doesn't really get any faster. It's too fundamentally constrained on something that doesn't really evolve over time, which is the gpu bus. The GPU itself, and its amount of RAM, and even things like pixel fill rate, have evolved far faster than that.
So either way, given more modern hardware, this can push more ships than the original game. The question is definitely more gray-area when you go back a few years in hardware, but what I found surprising is that on a really non-gaming-oriented machine from early 2013, with a notoriously ugh GPU, it was performing really surprisingly well in early experiments. That's really the extent of what this original post was about, so far; hence why I was not really wanting to make larger-scale predictions about how that hardware will hold up under the full simulation later, or larger battles, etc. It's promising, but not definitive.
ALL THAT ASIDE.
What I have observed is that, in my opinion, a smaller number of ships in this game feels like many more ships in the Classic game. The reasoning for this is what actually gets shown on the screen.
In Classic. there were a couple of options:
1. You're all the way zoomed in, and can't see all that many ships because... well, you're zoomed in.
2. You're all the way zoomed out, and the icons combine with nearby icons so they remain legible. You see large numbers on the sidebar and elsewhere, but you don't really get the full "feeling" of that vast number of ships in many cases because the ratio of zoom to icon combining is such that it had to combine a lot for you to see.
3. The battlefield is really spread out and there's just mess everywhere. Icons all over the place, not grouped together much, and so everything feels really huge and epic. This happens some when there's FRD against a large force, or you've just got such superiority you can overwhelm the enemy, or when you have turrets and similar all over the place. But it's not exactly the all-the-time scenario. This is the biggest feeling of scale I think the first game gave, though.
4. The last one is kind of a hybrid of 1 and 2, where you keep zooming in and out and thus it is combining and uncombining icons. You zoom out and see fewer icons, and it doesn't look as large. But as you zoom into an area, the icons separate and you see just how deep the rabbit hole goes. So by virtue of zooming in and out, you get a greater sense of scale than any particular single image could convey -- if that makes sense?
That's my interpretation of the first game, anyhow. With the second game, things are different:
1. Icons don't ever combine, because squads are inherently already combining icons. So when you're very zoomed in, you see fewer icons rather than more.
2. However, the elements of individual ships within a squad are a lot more apparent even when you are far out, because you see shots and contrails and so on all over the place.
3. There's a much larger variance in scale in general, so with some of the really huge ships like the Ark you wind up getting a different sort of feeling of scale that never really happened in the first game. There's a sense of "this is a really big thing," not "that's a 4x4 chess piece" or what have you.
4. In general, because of how the squads work, there winds up being a bit more of a spread-out feel to the battles, ala case #3 in the first game, even when there's not a bunch of guys in FRD everywhere. It's just the nature of squads needing more room so their ships can fly around, and thus they are more naturally spaced a bit better.
5. And so I guess my overall conclusion is that it feels larger or equally large to me because there is stuff going on in as much or more of my screen, or as much or more of my gravity well. There's a similar amount of "dazzling my eyes" with lots of things happening with a smaller number of ships compared to a larger number in the other.
Again, just in my opinion. And that's not to say we're aiming for smaller ship counts, or will have smaller ship counts, or anything else of the sort. It's simply an observation of mine that the battles feel bigger when they are smaller, so I can only imagine that will continue to be the case as the battles are progressively larger.
It's particularly misleading right now since all of the ships in each squad are presently stacked on top of one another rather than swarming, and we only have but so many ship types at all yet. Art takes time, etc.
The TLDR is that I think you've got nothing to worry about and will be really happy. The numbers I'm talking about here are just numbers from the early fights, and it's notable because of how bad the hardware is, how high up the quality settings were turned, how large the screen resolution was, and so on -- not inherently how large the unit counts themselves were. AI War Classic ran a lot rougher on the same hardware with similar ship counts, with things locked to 20fps, with a downsized resolution, etc -- and that was due largely to the fact that the GPU in question is constrained primarily by the GPU bus and the pixel fill rate above all else, to a degree I hadn't even fully appreciated until today.
If that makes sense? I think you seem very anxious that every time I mention a number of ships, that's the maximum size for battles or something -- it's just the first battle in the game, before you've even captured any planets.
I know that when we moved from Classic 3.x to 4.x we took a big drop in unit counts there for a while, and on some hardware in general. I'm guessing that's the root of your worry? That was a mixture of the sprite batching from Direct3DX9 extensions not being available in unity, and then the overall mono runtime being slower. It's a bit apples and oranges with this move to 3D. We're still in the same mono runtime, but Keith has been kicking its butt all over the place in the multithreading department. With the shift to 3D, we're gaining more batching and making more savvy use of the graphics hardware, rather than less.
For example, AI War Classic had a huge amount of overdraw because of partial transparency in every last sprite, and then drawing from back to front simply because of having to. So it might draw the same pixel 10+ times, or even more, if there were a lot of icons near one another. Here we're drawing mostly opaque geometry, so the z buffer prevents a pixel from being drawn more than once. It does do a lot more work with things like lighting and normals, but even a bad GPU is massively parallel and so that's not a big deal -- the overdraw is a sequential process, not a parallel one, but geometry and lighting rendering can be done in parallel when properly batched, which we are doing.
We do of course have partially-transparent things like explosions and shots in AI War 2, and those cause more overdraw. But they take up less of the screen on an average frame than all the ships did in Classic (plus all the shots and explosions there), if that makes sense. And I've been mulling a switch to shots being based around opaque geometry with a big emissive glow, because I think that might look actually a bit sharper visually as well as speeding up the rendering enormously. Explosions will always be in the transparent pipeline by nature, but it's not too big an issue apparently. Same with contrails.
Put another way: I expected to have to switch off things like the glow, the antialiasing, reduce the texture quality, the contrails, and so on with the poor quality laptop GPU. I expected to have to do that to get 30fps in a basic non-fighting sort of situation. Instead during an early game fight all of that is ON and I'm getting somewhere far north of 60fps, which far exceeds my expectations. We didn't have antialiasing at all in Classic, for example. Or something like contrails. Or post effects like glow and bloom.
Turning off the whiz-bang on older hardware shouldn't be a crisis, so it's not been a big worry to me. The exciting thing today was finding out how well it performs in early-game busy-scene scenarios with all the whiz-bang ON, even on old hardware. It's not really a remark on long-term unit counts or what I think hardware of that sort should have turned on or off in order to run late-game battles at acceptable framerates. To the former, the answer is "if you can't run it on old hardware, turn off some of the special effects or mod it to have lower unit counts, but we're not designing the game around that;" and to the latter, that really just remains to be seen as this evolves during the alpha period.
I'm just really, really surprised and pleased by this particular batch of early data, though. It's more than two standard deviations more positive than I expected, given the set of input criteria of this specific test. That's... substantial. Really, really substantial. It points to even better things down the road, and is in no way a commentary on unit counts in a more generalized sense.
Wow I repeated myself a lot, sorry about that. Hopefully that was clear, though. I just see that worry coming up again and again from you in particular, though. I
completely understand why, particularly given the leap from Classic 3.x to 4.x. So I've just tried to be as thorough as possible in hopefully reassuring you. Bear in mind that things will only get better, too; I haven't even bothered putting in the ability to tune most effects thus far, because frankly they haven't been needed yet even on my worst testing hardware. That's also kinda notable. But we'll have a lot more data, rapidly, over the coming couple of weeks.
