In the office we have a couple of dual xeon pcs, thats 8 cores on each pc. when i render an image locally all 8 cores are used.

today i started using rendercow with 3 pcs but i see in the task manager that only one core on each pc is used (so 3 cores instead of expected 24!) thats slower than single pc?!?

is this a limitation of rendercow or do i need to configure something?

thanks in advance
chris

Are you rendering a panorama? Panos are not multithreaded, only reason I can think of.
Hypervue is on a per machine basis, not CPU....

YES! it was a panorama ... thanks!

but why can the tiles of a panorama rendering be split up between pc's and not between cores?!? does this make sense?

My knowledge stops here, I'm afraid...

i am testing this now with a "normal" image. GlobalRenderosity 2400x1500 pixels tuned final settings ... took 13hours on single pc. now with 4 pc's in hypervue i see that the average cpu usage is between 10% - 50% per pc. and although in the task manager more than one core is used only one single tile is rendered in hypervue at a time per pc. does this include multiple cores?

so compared to 8cores with 100% on single pc that would result in render power of 4-20 cores (with 5 pc's = 40cores).

i am guessing the single tiles render same speed but the overhead between tiles must be slowing the process down.

its a little disappointing that 5times pc power only results in none-2.5  rendering speed improvement.

weird? does this get better when rendering animations?

or would i need much more rendercows so it makes more sense?

Radiosity and caustics preparation times are not multithreaded either. One tile is rendered per machine, this is normal, but CPU usage should be 100% per machine,when the preparation time is over....

hmm ... well i need the radiosity. the lighting is so much better. and tomorrow i will pop in the office and see if and how much faster the hypervue version of the rendering went ... so have a nice friday night and good weekend to all who read this! :)

Cheers. Have agreat weekend too.

quick update: it took under 3hours with 5 pc's in the end. started with 3 and added the other 2 later.

so i can forget all my theoretical calculations and be happy with much speed improvement :)

Cool!

rendercowupdate:

ran another rendering with 4pc's today but this time i unchecked the "force tile size to 64x64" in hypervue. instead vue chose 5x5 = 25tiles and now i see all pc working 100% on all 8 cores each. and that is with GR!

so i guess with 64x64 the tiles where so small and rendering completed so fast that prep time for new tiles was slowing everything down.

the reason why i checked 64x64 was, that i thought a multiple number of the 8 cores will be good, but i guess 64 was way to much :)

i will do another panorama test soon. i had one rendering going (apx 4000xsomething pixels) and on one 8core pc it was slaving away for 3 days! i want to see if i do the same thing with more pc's if the panoramic rendering does split up between pc's ...

I would be interested in how you make out in your time savings. So say on a single pc it took 13 hours to render and with 5 pc's with rendercow it took ....3.

I ask because I hooked up another pc to my network and rendered an animation using rendercow and two PC's. I dont know what it was for sure but i wasnt  very impressed with the time savings. I can hook upto 3 pc's in my home but wondering if its worth as it seemed to take just as much time with my one pc.....Yes it depends on cpu and memory ...both my machine are dual core 2.6 ghz w/2 gigs of ram each. I guess i was expecting more out of rendercow ...? From what i could see my local machine did 3 times more frames than the other pc.

just curious..

i'm afraid  have no other measured times than the above mentioned. i was so busy completing the project that i wasn't able to do tests.

from what i have experienced i believe the tiling of the render is the most optimizable part. but also the network speed will determine how fast the overhead is handled.

before the actual rendering of a tile/frame starts the geometry and textures are distributed to all rendercows and after the rendering is finished the calculated image is collected at the host. this is what i call the overhead and what "get's in the way" of the pure rendering time.

also i have no experience with animations. i dont know if the overhead time is spent another way (does it send all textures at the beginning or does it start sending texture only if they come into camera vue?)

basically i would expect that the longer the render time is the more benefit you would get from more pc's. if single tiles or frames are calculated "too" fast then the overhead might turn out counterproductive.

as soon as i have more comparisons i will post them here ...

good luck
chris