[Robert Cervellione]

after a few posts on another thread

http://www.maxwellrender.com/forum/view ... =benchmark

i decided to use the skydome.mxs file and do a real test between x64 and 32. i got some shocking results: below are the render times, you will notice that win32 rendered FASTER than x64. to sum it up the bench was to converge to 15 samples, but the renderer did not stop at exactly 15 (maybe someone can tell me why) so i equate where 15 sample would of eded up in order to normalize the data:

win 32, -s:15 time= 40m 3.63s
win 64, -s:15 time= 42m 27.72s



this just tells me that a 32bit app runs the same no matter what windows, the only advantage would be for large sceens that need the extra ram

SYSTEM USED TO TEST
-dual xeon 3.6
-6gb ram
-dual nvidia 3300 on SLI
-10k sata HD's on RAID 0


Windows 32

0h 0m 3s] of [16h 39m 0s] , current sampling level : 0.00 /
[0h 0m 12s] of [16h 39m 0s] , current sampling level : 0.58 /
[0h 0m 25s] of [16h 39m 0s] , current sampling level : 2.67 |
[0h 0m 48s] of [16h 39m 0s] , current sampling level : 4.55 |
[0h 1m 30s] of [16h 39m 0s] , current sampling level : 6.34 -
[0h 2m 43s] of [16h 39m 0s] , current sampling level : 8.09 |
[0h 5m 10s] of [16h 39m 0s] , current sampling level : 9.82 |
[0h 10m 4s] of [16h 39m 0s] , current sampling level : 11.54 /
[0h 19m 52s] of [16h 39m 0s] , current sampling level : 13.26 -
[0h 29m 47s] of [16h 39m 0s] , current sampling level : 14.27 /
[0h 39m 42s] of [16h 39m 0s] , current sampling level : 14.98 |
[0h 49m 37s] of [16h 39m 0s] , current sampling level : 15.53

Render finished successfully

windows x64
|
[0h 0m 3s] of [16h 39m 0s] , current sampling level : 0.00 \
[0h 0m 12s] of [16h 39m 0s] , current sampling level : 0.58 /
[0h 0m 27s] of [16h 39m 0s] , current sampling level : 2.67 \
[0h 0m 51s] of [16h 39m 0s] , current sampling level : 4.55 |
[0h 1m 34s] of [16h 39m 0s] , current sampling level : 6.34 /
[0h 2m 52s] of [16h 39m 0s] , current sampling level : 8.09 /
[0h 5m 26s] of [16h 39m 0s] , current sampling level : 9.82 /
[0h 10m 37s] of [16h 39m 0s] , current sampling level : 11.54 -
[0h 20m 25s] of [16h 39m 0s] , current sampling level : 13.19 -
[0h 30m 14s] of [16h 39m 0s] , current sampling level : 14.17 -
[0h 40m 4s] of [16h 39m 0s] , current sampling level : 14.87 |
[0h 50m 1s] of [16h 39m 0s] , current sampling level : 15.41

Render finished successfully

[daros]

Dear Robert Cervellione. I think that to take advantage of 64 bits you must allocate in memory from 16 GB to several terabites of data. Most of 64 bits motherboards can't accept more as 16 GB of memory. To work with a 16 GB geometry data you will need on your motherboard about 12 dualcore opterons. That's an other big joke of Microsoft.

[andronikos916]

hi daros...

But how can you explain 20%+ render improvement under C4D and Win64?

cy,
Andron

[Robert Cervellione]

i can answer that. i test a 32bit version of maxwell under two different os's. C4D has a 32bit and 64bit version of the app. a app compiled for 64bit will run faster than a 32bit app. the fact that there is only a 20% increase in C4D means that somwhere the app is not fully 64bit. when maxwell releases a 64bit version you should see closer to 50% improvment in speed as long as you have 64bit hardware and os to run it on.MentalRay has released a 64bit cmd line render and you will see 50% or better speed improvement over the 32bit version. also win32 can only handle 2gb or ram anymore than that in you system and its useless. winx64 can handle up to 16gb and above. so i have 6gb of ram in my sytstem. when in win32 4gb of it is useless

[bri]

nearly correct. C4D is fully 64bit, but it is a very optimized renderer and the improvements came from disabling legacy code in the 64bit version. MRay is faster in 64bit because of new optimizations that make use of the new general purpose and sse registers (doubled) and better memory management. and every project that uses more than 3gb will see huge benefits from going to 64bit. MXWR coders said they don't see benefits from doing 64bit, but it is known to be very dependent of compilers and code optimization for it. I work in architectonic acoustics, and our progs are going 64bit for better speed (more registers) and MO' MEM