Page 1 of 1
about muti GPU render speed...
Posted: Mon Dec 27, 2010 8:27 pm
by hubman
i seen the film "Real-time rendering in Octane render with 2x GTX480 " on youtube,
i got question about render speed,
2 GTX480 video cards = double render speed、double video memory?
can anyone tell me?
sorry my english is very poor...
Re: about muti GPU render speed...
Posted: Mon Dec 27, 2010 8:40 pm
by BlueBread
No it will only use one graphic's card vram however speed is combined. Also, something to keep in mind is that if you have two different GPUs running in one system, Octane will use the GPU with lowest ram available. For example you're running a gtx 460 2GB and a 480 1.5 GB whilst running octane and using both you will be limited to 1.5 GB.
Re: about muti GPU render speed...
Posted: Mon Dec 27, 2010 9:04 pm
by hubman
BlueBread wrote:No it will only use one graphic's card vram however speed is combined. Also, something to keep in mind is that if you have two different GPUs running in one system, Octane will use the GPU with lowest ram available. For example you're running a gtx 460 2GB and a 480 1.5 GB whilst running octane and using both you will be limited to 1.5 GB.
i tested my GTX480 & GTX260 in multi GPU mode by demo version... same as you said,
but the 2 Different video card ran under multi GPU mode, the time result is slower 30.sec...
that you mean if i put same 2 video card, the render speed will double?
thanks you!
sorry my english is very poor...
Re: about muti GPU render speed...
Posted: Mon Dec 27, 2010 9:50 pm
by BlueBread
In this case i my self am confused

. You will have to wait for one of the forum moderators or someone with more knowledge to answer your question, sorry I am no help here.
Re: about muti GPU render speed...
Posted: Mon Jan 03, 2011 12:36 am
by radiance
This is due to the GTX260, being a non fermi card.
Combining a fast fermi card and a slower pre-fermi card usually makes the slow card slow down the faster card.
Nvidia has made this even worse in the new cuda 3.2 drivers and the problem now also manifests with geforce fermi cards.
(I guess it's nvidia trying to get everyone to buy quadro/tesla hardware)
We're currently rewriting the multi-gpu code to fix this.
Radiance