Irrlicht Performance tests (Custom/Spintz implementation?)

per4manz · Post by **per4manz** » Mon Jan 30, 2006 9:07 pm

Has anyone done some comparative testing on the base 0.14 irrlicht versus for example your custom build (profiled/optimsed in vs2005 etc) or irrspintz?

what I am looking for is a set of benchmarks that take a standard demo that uses non exotic features but a scene of around 20,000 polys, some billboards + particles etc and then tests in the different implementations of irrlicht that are floating around.

Irrspintz mentions many speed-ups and there are plenty of "improvements" posted on this forum (some of which I have put into my custom dll) but I would really like to see a central "testing" area (on the wiki or forum) that can remove ambiguity and say "this improvement definitely works" or "this implementation is faster than standard by X%".

Failing that, some anecdotal evidence from anyone who has tried spintz/custom and what improvements they got. The worst areas in irrlicht that are causing bottle necks. And places for improving things like batching, reducing state and texture changes etc.

Thanks for any info - I am sure everyone will find it interesting.

Guest · Post by **Guest** » Mon Jan 30, 2006 10:27 pm

and I am sure you know how to use the search function.

sorry, couldnt resist

we have lots of threads that cover the bad things about irrlicht and a few threads how to speed it up a bit and how to improve it (with benchmarks and stuff).

see you!

Spintz · Post by **Spintz** » Tue Jan 31, 2006 12:23 am

that's a good idea. I'll try it out with the techdemo.

Eternl Knight · Post by **Eternl Knight** » Tue Jan 31, 2006 2:23 am

Um, dude - he was asking for benchmarks between the latest (0.14) Irrlicht & IrrSpintz engines. I did a search and cannot find such a comparison - please enlighten us with your anti-noob wisdom on where such a comparison can be found.

--EK

EDIT: that was to GFXStyler, not Spintz.

Pr3t3nd3r_guest · Post by **Pr3t3nd3r_guest** » Wed Feb 01, 2006 5:21 pm

irrlicht spintz 1.14a

but my questions is: are hardware buffers finish? There are 4 people on the forum 1 guest already made hardware vertex buffer , spintz is working at that, one more is working too at that ... of curse alone, and an other made hvb for his old engine ...
... and who knows how many are working to do that ...

... maybe if we work together ...

+ i want independent code for adding it to irrlicht 1.14 (there is something in irrlicht spintz ... but i don't know what need exactly to be modified in irr and how good it is)

Spintz · Post by **Spintz** » Thu Feb 02, 2006 2:50 am

no hardware buffers are not fully implemented in IrrSpintz. Mostly because the sceneNode/Meshes need to be changed to create vertexBuffers if you want to use them and manage them, meaning clearing them up when you're done with them. I'm still thinking the design of that out, i.e., if vertex buffers can be enabled/disabled per mesh/node and how to handle different types of buffers( specifically in DirectX )

per4manz · Post by **per4manz** » Thu Feb 02, 2006 10:07 am

GFXstyLER wrote:and I am sure you know how to use the search function.

sorry, couldnt resist we have lots of threads that cover the bad things about irrlicht and a few threads how to speed it up a bit and how to improve it (with benchmarks and stuff).

see you!

Gfx - I have been posting here virtually as long as you have so are no noob, I just choose not to log in most of the time. I know about search, I use it often. What I wanted was specifically 1 thread to compile all the test results in rather than 10 random threads with "post 13 on page 2" of one and "post 26 on page 5 of another" being only the useful parts - and even then they are not specific benchmarks. I suggest your compulsion to post "use search" is becomming more spammy and signal to noise increasing than most "newb posts" around here. In other words, if you don't know, please try and hold back from posting the obvious, there's a good chap.

Anyway, thanks to the others for taking it seriously. Hopefully we can all start to look at real numbers and maybe give Niko cohesive and collated feedback/data for his future performance tweaks. At the very least we can as individuals see what is and is not worth adding into our own custom built .dlls. (my custom dll has more windows compat fixes and enhancements than pure performance boosts - though I have added many of the tweaks posted in these forums).

rgds