Low Poly is a waste of time and creativity.

Low Poly is a waste of time and creativity.

I?m now running 10 x 40,000 poly figures fully animated and vertex blended at 20+fps on as low a spec as an Nvidia 5200 using directx. All with full game play character smarts, collision detection, gravity / physics effects (none of which effect graphics performance anyway much) and special effects, and networked with 4 co-players with a 150,000 poly scenery figure models with 1000x1000 plus textures and the background with 3000x3000 plus.

Keep in mind the Nvidia 5200 is 10 x slower than the 6800 and 20x + slower than the 7800, ps3 and xbox 360.

This means WHAT DOES LOW POLY MATTER!

It's a waste of development effort and stifles creativity. Hasn?t anyone in this forum heard of MOORE?S LAW?

Admittedly it is my own code not some stupid inefficient engine. SKINNING ? what the heck is that where you want to get a chicken without skin? It's soooo inefficient and over complex ? tip for the month use geometry blending 1 level only (any more may be inefficient and is unecessary anyway should apply vertex weights for smoothness) and animate materials using transform.world and joints transform.world1 ? I shouldn?t be telling you this secret you don?t deserve my love. It's like the time when I was younger when I discovered how to unchain the 256 colour VGA mode 14 years ago to allow for paging in the only colour mode over 16 colours and shared it with everyone ? no thanks then and will get no thanks now just more gutter sniping from the inexperienced.

Tip - start designing larger models - low poly will mean nothing if the coders are any good in a couple of years. Although, that assumes good coders who don't have to rely on inefficient engines???

mcdrewski's picture

No sniping at all from this inexperienced gutter dweller - I'm happy giving you the benefit of my (somewhat jaded) doubt.

Where can I download the game or demo? I'd like to see just how creative you've been!

3DArty's picture

Yes Please Do Show.

Mdobele's picture

pffffft YOUR models are so low poly! Its you who are behind. Get with the times dude, my models are all 4million tris a piece. And its running on my uber Voodoo 3. All you have done is show how much of a n00b you really are bragging about such low specs. [:p]

unknownuser2's picture

What a joke, moderators - cull this fools babble.

/my two cents.

Anuxinamoon's picture

You did not show us what you preached in your last post and you still will not show us in this post. How will you expect us to belive you?

Makk's picture

lame
here is a tip just for you........shut up

poser's picture

Well I suppose I should comment - I'm not going to get personal like some - pity to see someone who sponsors a site wanting to cut off debate.

On the more constructive side - yeah sure I can give a demo.
ALSO - I have a good idea - if you want to email me 40-50,000 poly models with some type of joint animation file I can convert to my format to test in my environment - this way you know I'm not cheating. Otherwise, how can I prove what I'm saying.

For my demo I will have to encrypt artistic content using crypto api first - everyone has property rights eh? Can expect a demo next month.

Anyway, sounds like some of you don't care and think that 4 million triangles on a Voodoo card is the appropriate standard at which to be technically impressed - interesting idea where do I get a Voodoo card from now days so I can prove the technical impossibility of this ridiculous suggestion.

But the point of what I was saying is LOW POLY is USELESS or should I say DEAD is good advice take it or leave it.

conundrum's picture

quote:Do not feed the troll...

http://members.aol.com/intwg/trolls.htm

poser's picture

Why everytime a debate is lost by you dominant group on this site do you finish with that childish comment?

unknownuser2's picture

This isnt a debate - your not asking for opinions your stating what you believe to be fact without any evidence to back up your reports.

In a nice way, put up - or shut up [:D]

You posted about a month ago with a similar post - touting polygonal goodness - I emplore you to make me look the fool and show us your masterpiece.

Until that happens - your just another one of the many *many* people who claim grand things and never deliver.

For the record, Ive never once stated you cannot do what you claim, but you must admit stating something like this:

"1000x1000 plus textures and the background with 3000x3000 plus"

Sounds to me like you have absolutely no idea what your talking about - that is nothing to be proud of at all - unless you told me you had that running on a PS2 I wouldnt bat an eyelid at those specs.

I have a couple of questions for you.

1) What exporter are you exporting your meshes into your engine with?
2) What raw texture format are you using ?

quote:so I can prove the technical impossibility of this ridiculous suggestion.

Forget that for the moment and focus on prooving what you speak of first, then afterward you can dazzle us with your technical prowess by solving Mdobele's very achievable challenge. [:)]

Also lastly - to answer your question that you have so thoroughly researched " What does low poly matter ? "

It matters threefold:

1)Why would anyone want to spend the time modelling, UVW mapping, texturing, weighting and animating such a high res model when I could build 5 other models in that time that would a) cost the same amount and b) look no different in silhouette from your 40,000 poly one?.

2) Rome total war I believe can handle hundreds - if not thousands of characters on screen, including dozens of unit types. This genre of game where the camera is a long way from the individual models thrives on budgets.

3) Investors and Publishers are smarter than you think - developers must develop a product that is financially feasible. Forget the rest. And simply stating " I can push more pollygons " is only going to drive development costs up. Why would any investor or publisher in his right mind pick up a project that takes advantage of your uber engine and massive polycounts, when a similar project is on offer that would cost less than half the amount and have an end reuslt that looks pretty much identical ?

poser's picture

If it is of any interest to anyone -

Animation Format -
bvh motion files which I believe are supported by
3dsmax, lightwave, lifeforms, motionbuilder and poser - not sure about Maya.

Texture format -
jpg / dds for transparent
(but can convert pretty much anything)

Model format -
not surprisingly I prefer .x but can convert from obj, 3ds and about 20 other formats - not maya though which probably annoys anyone from Krome reading this but hey who cares?

rezn0r's picture

Why would you use a compressed texture format for your super super optimised engine? [:P]

Scott.

poser's picture

Well actually you make a good point rezn0r as directx expands jpg's to be the same size in memory whether you use bmp's or any other format. This is for speed of rendering in memory - I'm pretty sure the consoles work the same way. But hey why not save a few disk mb especially if you're trying to email or download stuff to people who don't have ADSL - yeah ADSL, so good to be able to download at 1.5 mega bits I don't have to wait anymore - yeah.

WiffleCube's picture

Wow you fellows can be mean when you want to be! It's
like poser jumped into a field of sheep and they all
turned out to be rams [:p].

I'd like to add another factor to Moore's equation; the
higher the processor/memory speed, the sloppier the next
version of Windows will be, clogging the system to a crawl.

For an objective opinion, try looking at the same title on
PC and XBox to see the difference in speed and quality
using what is essentially the same hardware.

denz's picture

hehe. well looking forward to the demo.

mcdrewski's picture

quote:
quote:Do not feed the trolls
http://members.aol.com/intwg/trolls.htm

Comment by poser

Why everytime a debate is lost by you dominant group on this site do you finish with that childish comment?

Okay, I'm trying not to take this off-topic... but since I was the original poster of that comment in your other thread, I figured I'd explain.

Firstly - despite the somewhat confrontational tone of your post, I was actually impressed. I recall that last time you posted you had a rendering/animation only rig, and no 'game' code. The fact that you've gone off and put in AI, collision, networking etc suggests that you're actually a pretty talented coder, and that you are listening to what people here say and are taking it on-board.

The reason why people might immediately [url="http://members.aol.com/intwg/trolls.htm"]dismiss your post as a troll[/url], however is that at first reading it simply looks like one.

To explain, imagine that you walked into a job interview and in response to a question, read your post word for word (lots of your potential employers DO read sumea). Do you think you'd get the job talking about "stupid inefficient engines", asking if anyone has ever heard of moore's law, or complaining that you got no thanks for something unrelated 14 years ago...

I certainly wouldn't employ you based on that. People here also won't welcome you with open arms for the same reasons. They'll more likely assume that you're trying to start a fight, ie: you're a Troll - even if you're not.

So - good luck - I'm really looking forward to seeing the demo - after all, you've made some pretty impressive claims!

conundrum's picture

There are a few things that really make me skeptical

1. His name is poser
2. The first topic was using obvious Poser models (bad ones at that)
3. Most importantly, if his engine is really that much further ahead of the pack, why is he spending his time arguing with people on a relatively small forum rather than working with a big company.

My opinion is that he is someone who actually does know what he is talking about and is just trying to piss us all off. The best example is using Poser models, if there is one thing that is sure to set off a game art froum its someone posting Poser models and claiming that they're great

i seriously do not buy it

Mdobele's picture

Just send me a compiled demo Poser that backs up your claims and I'll hapily eat humble pie. Untill then I will simply not believe a word you say because my experience shows me otherwise.

poser's picture

For those interested in technical stuff.

More on the technical specs of what's being done - I have purposely left out discussion on the other most important part of 3D graphics performance ? shader throughput as this was not what the discussion was about - i.e. it was about low vs high poly counts...

Testing on NVidia 5200
Fill Rate (pixels/sec.) 0.5 billion
Vertices per Second 31.5

Achieved -
10 x 45,000 polygon character smooth animation with total texturing exceeding 2000x2000 each
1 x 110,000 polygon environment background model
15 frames per second.

Testing on NVidia 6600
Fill Rate (pixels/sec.) 4.0 billion
Vertices per Second 375 million

Achieved -
70 x 45,000 polygon character smooth animation with total texturing exceeding 2000x2000 each
1 x 1,000,000 polygon environment background model
20 frames per second.

It's very interesting that these two figures 1. verts per second and 2. fill rate are a great predictor of the throughput or should be if you're not using a slow engine. If you do a simple calculation you can see at (10x45,000 + 110,000) x 3 (verts per polygon) x 15 = approximately 25 million verts per second which is quite close to the raw limit quoted by Nvidia of 31.5 million. Similarly this seems to hold for the 6600 with (70 x 45,000 + 1,000,000) x 3(verts) x 20 = approximately 250 million verts per second.

The small extra loss (about 30%) is for operations of game environment, animation, blending, gravity, collision, networking, network play and user input and a giving up of some processor time for other programs (only required for fully windows games).

NVidia 7800 (ps3 and xbox 360 are similar)
Fill Rate (billion pixels/sec.) 10.32
Vertices/sec. (million) 860

Estimate -
200 x 45,000 polygon character smooth animation with total texturing exceeding 2000x2000 each
1 x 2,500,000 polygon environment background model
20 frames per second.

Other cards of interest for comparison
NVidia GeForce FX 5950 Ultra
Fill Rate (pixels/sec.) 3.8 billion
Vertices per Second 356 million

NVidia 6800
Fill Rate 6.4 billion pixels/sec.
Vertices per Second 600 Million

ATI RADEON X850 XT PLATINUM EDITION
Fill Rate 8.6 billion texels/sec.
Vertices per Second 810 Million

ATI RADEON X850 XT
Fill Rate 8.32 billion pixels/sec.
Vertices per Second 780 Million

ATI RADEON 9600 XT
Fill Rate 2.0 billion pixels/sec.
Vertices per Second 250 Million

poser's picture

Ps. I have been corrected the PS3 RSX is said to be...

Vertices Performance: 1.1 billion vertices per second

And a shader operation throughput from hell.

unknownuser2's picture

Definately looking forward to seeing the demo Poser [:)]

conundrum's picture

just out of interest why do you seem to be aiming for such a low framerate, i believe it needs to be at least up to 24 and probably 30 for it to be a reasonable speed for gameplay.

Rahnem's picture

Sounds like a perfect world kind of scenario. How many textures does your 1 environment model have? Try this as a test, take your 1 high poly background environment model and chop it into about 200 different models with different textures on them and see how your engine preforms then. See how fast your code can deliever that info to the GPU. I think you'll find that the CPU spends more time on the model than the GPU does rendering the thing.

Malus's picture

Poser: Hmm, a post full of specs....
Doesn't really mean much since you haven't show anyone a piece of practical or tangible proof.

I'm with Hazard, Show us a demo, its not really that hard is it??

poser's picture

Hi Rahnem,

what would be your reason(s) for breaking up the model?

Regards,
Mark

poser's picture

Ok Rahnem,

I asked you a loaded question so I won't wait for the answer as I know your reasons.

1. To animate the parts of the model so that you can blow things up destroy and generally create havoc on my peaceful scenery.
2. To selectively distance or field of view cull.

Well I really don't know why I keep sharing this pearl of wisdom but it is all in the materials. I.e. animate and change and hide / show the materials don't do separate models or skinning as these are slow and actually not even as flexible as working with materials. This is by far the fastest way based on at least the way directx does things and I suspect it is the same for the consoles. E.g. if you have a lamp post you want to knock down give it a separate material and rotate / animate that / then can even hide / show a broken / good lamp post material ? if you want to cull it just don?t show that material. Boy I shouldn't be telling everyone this - it's the unchained VGA mode all over again.

Sorceror Bob's picture

Higher poly models are (generally) much nicer to look at than the lower poly equivelents. However, more is not always better.

For a developer, low poly is simply more cost effective, with lower development times and expenses due to the relative speed in which low poly assets can be created. Add to this the fact that developers are trying to deliver their product to the widest possible consumer base. This means delivering a working product to people who haven't upgraded since 2000.
Why do you think valve did those surveys before half life 2 was released? It gave them a fair idea of what systems their target audience owned, and how they needed to scale their software.

Of course, with the next gen consoles coming out, scalability can be ignored somewhat; within the context of the budget anyhow. In a perfect world, no one would have to worry about things like meeting deadlines, or profit margins.

Also worth noting; which brings us back to Posers original post, is the shortage of game artists capable of creating quality high poly assets. I might be wrong, but artists that tend to focus on this area are generally more interested in working in the film industry. I think the pay and relative job security is more attractive.

So yeah, Low poly is not a waste of time and creativity. It's what games are made with, and therefore a skill any prospective game artist should be proficient in. The ability to create quality high poly assets is undoubtably desirable, but is not yet required to get a job with.

Personally, I believe everyone serious about getting into games should learn to work with high poly. It'll give you an edge over artists who have focused entirely on the low. Be versatile guys, and willing to embrace new techniques and technologies. Employers love people who can perform outside of their job description, and it gives them another reason not to sack you when it's time to 'trim the fat' ;).

Poser, I'm interested to know how your engine handles stuff like light, shadow and sparkly bits. Pushing that number of polies is impressive to people in the know. But for average schmucks like me, it's the 'oooh ahhh' factor that wins things over. Plus, how complex are the meshes you have in there? If there is lighting, could the surface area of a mesh create a variable in performance?
I mean say if you have two spheres of 50k polies, one is round and perfectly smooth, while the other is basically a gigantic demon turd, with spiny bits, chunks of corn and david hasslehoff hanging off it.

Err, lets see some screenshots!!

I believe the reason for Rahnems question, is that levels are simply not built in one huge chunk. It's simply not feasible.
Level designers aren't going to receive all their assets at once for starters, and once everything is in, they'll need to be able to adjust things to get the balance right.

mcdrewski's picture

quote:
Show us a demo, its not really that hard is it??

Nope - http://www.fraps.com/

mcdrewski's picture

quote:Originally posted by poser

[snip] don't do separate models or skinning as these are slow and actually not even as flexible as working with materials. [snip] if you have a lamp post you want to knock down give it a separate material and rotate / animate that / then can even hide / show a broken / good lamp post material ? if you want to cull it just don?t show that material. [snip]

I might be misunderstanding, but are you saying that the most 'efficient' way to animate a building falling over would be to build 50 models per second of stop-motion animated collapse, each with it's own material(s), and then cycle through them one by one?

You either need one hell of a tools programmer to automate this process from the artist's animations, or you're going to have an army of artists out for your blood.

poser's picture

Hi Sorceror Bob,

the lights don't really make much of a difference - using 3 but if wanted to use lots it might. For lighting effects - i.e. something in lower light in a scene most of the time a trick is to use different normals not to use different lights as this is faster / but mainly because it is less hastle than setting up lots of lights. (again more free knowledge ? I?ve really got to stop giving this away but on the other hand I?m not doing it for the money)

Hi mcdrewski, no that's not what I'm saying - but I'm sorry I'm really not willing to explain the implementation details in full.

EVERYONE,
Yes you will get a demo but it was just going to be character animation with backgrounds next month to prove the poly rates now it seems everyone wants more than that?

Ok so what I'm going to do is give some simple character animation this week and then for next month everyone can tell me what else they want me to prove. I.e.
Scenery / other object manipulation.
Volumetric special effects.
Shader effects.
Character Logic.

The simple demo already has collision detection, network play and things like gravity, panning, movement (mouse driven), clipping (for characters only).

Ps. I asked for but did not receive anyone else's models so I could prove I wasn't cheating - so now you will have to take my word on the poly counts. If when you see the models you disbelieve the poly counts then you will need to send me one of your own.

If anyone wants to then tell me what video cards you are running so I can tell you what to expect. I?m thinking of allowing you to specify the number of models to show simultaneously so you can test it on varying cards.