Best polycount for lowpoly models in babylonjs

[royibernthal]

What's the best (though feasible) polycount for lowpoly models in bjs assuming I want to have 10 lowpoly meshes rendered at the same time on mobile with dynamic lighting? (no rigs/animations)

Is 600 per model reasonable or is it too much?

[Wingnut]

ANOTHER "what is too much" question?  *sigh*  Why?  Why are you doing these, Royi?

When you asked this 2 months ago, folks said things like... "It depends upon what is TOO LOW of FPS-rates, in Royi's opinion."

Is the scene going to use shadows?  Post-processing/effects?  What does "feasible" mean to Royi?

Have you done any testing/profiling on benchmark-testing scenes?

*sigh*  Sorry.  But truthfully, Royi... if YOU had to answer your own questions, HOW does someone know how to answer?  Put yourself in my shoes... or even better... put yourself in @fenomas's shoes... who is a God of benchmarks/profiling.  How does he answer?  Your questions are SO "wide open" and without fine details... that it's like asking "What should I do next?". 

I'm being a jerk here... but please... say that you do "see" what I am saying, and that your "What is too much" is very subjective and opinionated.  I PROMISE... I PROMISE... that if you spend less time asking these questions, and more time learning to use your profiler in your browser dev tools area... your world will improve and your questions will get answered... by you.

I love ya, man.  I hope your project flourishes, and the gods bury you in happiness.... really.  But, man, you ask some super-difficult questions that have very "liquid" or rubbery answers.  You won't know anything for certain, until you do profile testing.  Then turn-on your FOR-loops, and beat the crap out of your desktop and mobile browsers.  See how they respond.  Produce a playground or two, and ask others for FPS testing on various devices they have... record their numbers and thank them.  Soon, you will have a graphable "big picture".  Which devices run what, and how well.  There are 41 million mobile devices, now.  Which ones are you asking about "too much"?   

See?  It's an overwhelming question.  It can't be answered accurately.  I think if you do some precision web searches, you'll find that many others are comparing performance of mobile devices.  It is likely... that your 3D rendering results will follow the trends seen in those already-published perf reports... but that is just a prediction/opinion... worth little. 

What is too much reduction in FPS? 

[royibernthal]

Let me remind you that 2 months ago people WERE able to give me accurate replies given my info, if you'll take a closer look

http://www.html5gamedevs.com/topic/24956-meshes-memory-usage/

Sorry but I have to disagree on that one, I don't think the question is subjective. As far as I know "low poly" isn't too open to interpretations, there are acceptable known ranges for polycount, and once I specify I'm aiming for mobile it should give a clear enough picture. Correct me if I'm wrong.

 

You're right however, I did miss a few important details, again. I'll try to narrow it down more:

-60 FPS, I'm aiming for no FPS drops at all, but let's say 50FPS is still acceptable

-Scene uses lighting, let's say 1 light source

-No shadows cast on other objects, if that's what you mean

-Device - iphone 5 at minimum

-"Feasible" means - while a polycount of 10 might provide great performance, you can't really model anything with that few polygons, so the min polycount should be realistic, feasible.

 

Very basic scene all in all, let me know if I missed anything that can make this question more specific.

 

I'm familiar with the profiler and browser dev tools, but I don't have a relevant device to test on.

I could produce a PG or 2 and ask others for their results, but it'd be a very cumbersome and long process , imo.

I insist that this is a very simple question to answer, at least for someone like Deltakosh, unless he corrects me and says otherwise.

 

btw I did try to google different things, I didn't find any relevant results. I found only 2 bjs benchmarks, which didn't help me understand what I need to know, or maybe I don't know how to deduct what I need to know from them.

Where are these amazing performance tests that you keep mentioning?  Please link me to them if you find them.

[ozRocker]

Have a play with my site here www.punkoffice.com/webiverse

The avatars are around 7,000 polygons.  You can log in through as many browsers as you want to increase number of avatars.  Turn on frame rate so you can see how that impacts performance.  There are armatures and animations though so I'm not sure how useful this would be for you.

[Wingnut]

http://www.notebookcheck.net/Mobile-Graphics-Cards-Benchmark-List.844.0.html

Mobile devices perf tests, not BJS perf tests.  vertex/pixel shaders.  webGL performance.  That's what i was speaking-of.    SO much variance among SO many devices.

But, yeah, perhaps somebody will speak some words that you will buy-into.  I did my duty... flipping your pancake-like topic... after it laid on the floor of the forum with 0 replies for 10+ hours.  I find that if I roll them over, less moss grows on them. 

[royibernthal]

Thanks for flipping my pancake-like topic I guess 

Indeed there is so much variance, when I'm talking about performance you can always assume I'm talking about the weakest hardware, i.e how's the performance on the weakest iphone5 device manufactured by apple. (assuming they're not all manufactured with the same hardware)

 

I think this link doesn't deal with mobile, the title of the page says "Laptop" even though the url and browser title say "mobile", there's no mention of WebGL, and Laptop is mentioned a lot in the list. (I might be wrong though, I barely know anything when it comes to hardware)

Anyway, having perf tests like that for mobile webgl should indeed be able to answer my question, plus a few more other questions I might have in the future. With some luck I might be able to find something if I dig a little.

 

Still hoping for an answer here regardless though

[royibernthal]

300-400 is considered to be very low poly, and it can produce satisfactory models. So that's the answer I guess, for anyone else who might be interested.