blender 2.30 optimized (update 16 november)

Compiling, libraries, modules, coding guidelines and porting

Moderators: jesterKing, stiv

Post Reply
Antares
Posts: 0
Joined: Fri Nov 14, 2003 4:04 pm

blender 2.30 optimized (update 16 november)

Post by Antares »

*UPDATE*

i managed to get another speed upgrade :)
and the main advantage with this upgrade is, that it works with EVERY CPU now.

i compiled it with "automated CPU dispatching" which means the programm is checking with instruction sets the current CPU is supporting and then switching to the appropriate code.

this gave me another 10 sec on my test pic (mentioned below)
times dropped from 2':05'' to 1':54''

i also included a test on my athlon 1200 512 MB RAM FSB 133
there the rendering times droped from 3':18'' to 1':37''

c00l !!!
but stil the P4 got the better advantages from the compilation... of course :D... its and intel with SSE2 compiled with an intel compiler :D :D

main disadvantage:
blender is 5.3 MB now :(

*UPDATE*

i posted this in the elysiun forum and cause i am a lazy dumbass i just copied the text now :D
(elysiun=> http://www.elysiun.com/forum/viewtopic.php?t=17626)

i compiled the blender 2.30 source with the spezial intel compiler and tested the speed improvments.

my system is a 1700 P4 mobile with 512 MB DDR FSB 100 RAM.

first i rendered this image again:
http://www.silentthunder.de/pics/throne1152x864.jpg

with the common binary for windows this task took 5 min 3.44 sec

after recomilation with all the intel opimizing stuff it only took 2 min 5.25 sec.
it was really amazing and i could first belive it, thus i did a lot more testings and it was always the same result.
the intel speed opimized blender was always at least 100% faster.

this is the file i used:
http://www.silentthunder.de/files/blenderintel.zip
(just copy this file into the directory where your "Normal" blender is located)

it is somewhat larger than the normal binary of blender 2.30 but thats (of course) the cost for better speed.

WARNING !!!
if u dont have a P4 with SEE2 instruction set this wont get u the results i got from it. not even mentioning athlons which totally lack this instruction set. (except the newest athlon 64 CPUs)

so DONT complain if it doent work with u when u dont have the requirements. u MAY get better results to cause of general code optimization but thats not granted.

perhaps some of u can give me some feedback on how it works with u.
i didnt have the time yet to examine the overall speed of the blender while working with it.

my next CPU will DEFENITLY be a intel now (with a nice intel compiler rdy for getting the max out of the CPU :D)
Last edited by Antares on Sun Nov 16, 2003 2:21 pm, edited 2 times in total.

JWalton
Posts: 58
Joined: Sun Oct 13, 2002 7:39 pm
Contact:

Post by JWalton »

similar experiences with SGI mipspro compile for IRIX. we only distribute the mipspro
compiler generated version, the gcc one is slower.

leinad13
Posts: 192
Joined: Wed Oct 16, 2002 5:35 pm

Post by leinad13 »

Thats awesome, i tested it on my machine:
P4, GeForce 4 Ti 4200 128Mb DDR, 760 MB DDR Ram.

I made a scene using loads of particles, just as a test. It took 40 Seconds with Normal Windows Binary, and only took 15 Seconds with the Intel Optimized one! That is awesome, it should be put on the server with all the other releases.
-------------
Over to you boffins

L!13

Post Reply