Well, from a test I did a while ago, in early May, taking out my blog again, here are my results and conditions in my experiment:

skin size (16x16 versus 2048x1024), object size (4x4 by 8 high versus 1024x1024 by 512 high), polygon quantity (4 versus 4096)

skin size (S for 16x16, L for 2048x1024), object physical size (S for 4x4 by 8 high versus 1024x1024 by 512 high), poly count (S for 4 polys, L for 4096 polys)
S S S = 48 fps @ 208 objects, 35 fps @ 312 objects, 26 fps @ 416 objects, 18 fps @ 624 objects;
L S S = maxed fps @ 27 objects;
S L S = maxed fps @ 64 objects, 38 fps @ 256 objects, 22 fps @ 512 objects;
L L S = maxed fps @ 27 objects;
S S L = 50 fps @ 9 objects, 28 fps @ 18 objects, 15 fps @ 36 objects; 8 fps @ 72 objects;
L S L = 50 fps @ 9 objects, 28 fps @ 18 objects;

So, about 360 objects on screen at once gives about 30 fps. From this, with a super high poly model, 69632 polys is the limit for 30 fps. This is with my Radeon 9600 XT on a 1.05 GHz processor with 512 MB 333 MHz RAM.

I don't understand how you were able to get a million polys....

In any case, the key is to avoid lots of entities and combine them into just one entity. My world has 25,000 polys total (and can be reduced to about 18,000 or so) and it renders at 30 fps. It consists of only six entities. This, however, does not include the platform levels which uses many entities (but this can be reduced drastically if the external textures feature gets implemented).

Can a moniter have anything to do with frame rate? Cause our moniter isn't the greatest cause we got it right when LCD were comming out and I know those arn't to great of ones. Max refresh rate of 75. I think that is why it isn't running good.

Refresh rate shouldn't have any impact on the frame rate the engine runs at. The actual frame rate you see is limited by your monitor, but the engine could easily run at twice the refresh rate provided you don't have a lot of "clutter" on the screen. If you see nothing at all, then yeah, you could get 180 fps with ease (and I've had this happen). I use 60 Hz refresh rate on my monitor and I've noticed that the frame rate has gotten up to 90 fps when I've got nothing showing up on the screen, except textual panels. Texture quality has almost nothing to do with the frame rate and I've run an experiment (see above) to prove this. The draw rate is intensely high, probably several terapixels or even petapixels per second (a trillion or quadrillion pixels per second) or something. It's hard to test this without having to have high poly models with an oversized skin image.

alright thanks. And for some odd reason my fps on a simple little test went from 63 to in the 200s. Probebly my comp got an update today.


And yes I saw your test results, interesting.

Quote:

---

The draw rate is intensely high, probably several terapixels or even petapixels per second

---

Sorry that's simply not true. Did you just make this up? The fill rate is limited to the 3d card. The peak(max) fill rate of a plain GeForce 6800 is about 3900 megapixels. The fastest card for fill rate available now is the Radeon X800, which peaks at around 8320 Megapixels. A petapixel card is absurd. No card will ever reach that kind of speed in a standard scanline renderer. By the time processors are fast enough for that, they probabably wont use pixels anymore but something else, like voxels, vectors or rays.

Quote:

---

Texture quality has almost nothing to do with the frame rate

---

Again, not true.. it all depends on your video card, not the A6 engine..it doesnt matter what your experiments show, its dependednt on your video card. Larger textures will always impact framrate, sometimes its neglibile, sometimes not.,. it also depends on what you doign with the textures. In my own D3D engine experiments, using multiple render targets and shaders with many texture reads, a differnce of 256 and 512 can make a HUGE impact.

The true fact of the matter is, the A6 engine is a standard D3D engine, and is able to acheive similar results as most other engines. I imagine the core rendere is very straightforward.. so in general, it's performance will always be broadly dependent the target platform. This is the case with nearly all modern engines. this is why i think this a bad quetion to ask "how many polys an engine can handle".. On every 3D engine forum i've been on, this same question is asked over and over.. the answer is always the same. Use a search function..

All that said, if you are using advanced shader techniques such as per-pixel lighting, certain optimazations can be done on an engine level to take take advantage of the z-buffer and draw order to improve performance. This is not currently availble in the engine. So the Doom 3 engine will be able to draw more per-pixel lit polys than the A6 engine. The point of A6 is not to meet every specific need of a certain game, but to meet general needs.

As general rule, dont worry about polygon count, worry about performance. If your 10,000 polygon model seems to run fine, than use it...there is not limit to the number of polygons D3D can draw per scene. A typical D3D render loop looks like this:

Device-> BeginScene()
...
//infinite length rendering calls, infinite number of draw primitives, etc...
//you could draw a billion meshes here if you wanted.. it would just take all //day for a frame.
...
Device-> EndScene()

Just try to make it run well on your target platform.. thats all you can do.

Not true eh? Well, try this. Create a model like a 9x9 terrain. Now give a simple 2048x2048 texture on each of a set of 2 faces. This would be about 4 megapixels. Since you have 64 areas where this can go, that's 256 megapixels. If your card does 2 gigapixels per second, you'd need about 8 of these to get 1 fps.

In my experiment, I used a 2 megapixel texture on a 4-poly object (a square pyramid without the base) and I had 27 of these on screen. This is 54 megapixels worth of objects. From the ATI website, my card has a fill rate of 2 gigapixels per second. Taking 2048÷54 and you get about 37.9. I was running at 60 fps (the "maxed fps" note), explain that. I'll run it again if you want and I'll find the maximum.

Okay, I just tested it. I used that same 2048x1024 texture and placed it on the sets of 2 squares each on 16 possible. This gives about 32 megapixels per model. I've got 24 of these models in there giving 768 megapixels and I'm still getting 80 fps! That's about 60 gigapixels per second.... To go further, I'll need to make the model bigger with more polygons or just have a bigger skin size. This is no joke ya know. I'm running the experiment right now as I'm very curious to what the limit actually is. 60 gigapixels (and this is with all of the objects on the view at once ya know) is way higher than ATI is showing.

Edit: Okay, I've expanded on it further and I'm still getting 70 fps with 3 giga pixels. The poly count is getting too high so it's further limiting my experiment. 210 gigapixels per second....

Edit #2: Well, I've reached my hardware's limit (video memory mainly and max texture size) and with 7 gigapixels for the entity skins (14 of these (I could go 15 though at the absolute maximum, but I might run out of video memory entirely (I have a 128 MB card; 15 of these would use up 120 MB, then I got the screen display and other stuff and at 14 objects (112 MB), I should be safe.).)), at 70 fps (even peaking into the 80 fps range), that's still an impressive 500 gigapixels per second. If you want, I could send the experimental level, the entity I'm using and you could see for yourself. I'm not joking either.

ulillillia, I don't know much about this, but common sence tells me your doing something wrong in your experiment. If ATI is saying their product has a fillrate of 2 gigapixel, you can be pretty sure that that is the ABSOLUTE maximum, tested under the best circumstances. It is very unlikely to ever get that in a real game, let alone 250 times as much..

ok the fastest Radeon is the x850 platinum-- it has a peak fill rate of about 8.64 Gigapixel/second. It can never exceed that..Your mistake is that you are calculating your fill rate by how many total pixels in each texture in each entity.. this is simply not how renders work. you rarely render all the possible pixels. Remember the textures are usually mipmapped or scaled down or up. The card ultimately renders pixels, but the engine internally doesnt think of that way.. it uses texels which arent exactly the same thing..no matter how big your textures are, you card has means of reducing the number of interpolated texels to draw to each pixel ( maybe i'm not describing this with scientific accuracy).. Whiles its true that a 2048*2048 texture scaled to actual size would take take longer to draw than a 128*128 tex scaled to the same size, rarely is a texture ever actually drawn to size.. usually its smaller, or larger, and in many cases when its larger it's not fully on the screen.. and often its shown through several mip maps anyway.

The point is this, its not as simple as you make it out to be, and your tests are flawed because of that.. no card will perform much over the listed peak rates... usually it will be under that for other reasons.

And to be honest, this seems to be a silly thing to investigate, as it has little bearing on what you do, unless you are really trying to optimize a multipass shader or some such.

Well what I read before is the way procesors are made is they test the procesors, if they feel a simple test they will put the speed down. So they might do the same with video cars. They will say it can run at such speed but it really can go higher, sometimes at high risks, or very very low risks.

But to what Matt just said, I still migh be right anyways.

Also engines can be compaired in polly count, that is why I asked what other people could handle. I wanted to know 2 things, how good the engines was generaly, and how good my comp is.

Acording to others my comp is behind for some odd reason. And this engine seems to be as powerful as others, but alot more stable than others as well from what I can see.

The only other improvment I would want, is an even better modler. But that is asking a lot and that is anouther thread. I think I have gotton what I want from this thread.