Iray and 2 x GeForce 970's SLI

rbel_295e7c1d4c

Is anyone running a 2 GeForce 970 SLI setup who can advise if they have experienced a significant performance improvement with Iray renders compared to just one 970?

Tobor

nVidia specifically recommends against using SLI for Iray rendering.

rbel_295e7c1d4c

nVidia specifically recommends against using SLI for Iray rendering

Interesting - I wonder if there is any significant performance improvement in running a pair of 970s without the SLI?

[email protected]

rbel_295e7c1d4c said:

nVidia specifically recommends against using SLI for Iray rendering

Interesting - I wonder if there is any significant performance improvement in running a pair of 970s without the SLI?

Not sure, but I would expect there to be an improvment.  The Iray engine will split the work up across both cards.  Doubt you'd see a 2.0x increase, but maybe a nice 1.5x.  There's a discussion of this here: http://www.daz3d.com/forums/discussion/54908/sli-and-iray and here:  http://www.daz3d.com/forums/discussion/54056/

The short answer seems to be that SLI is used to make two cards act like a single card, say for driving a video game.  To do this it pulls some tricks with the graphics processors and memory.  Iray doesn't need that since it's designed to make full use of two (or more) cards.  If you bring the SLI layer into it, you add more complexity which can lead to conflicts between the Iray and the SLI both trying to run the second card.

BTW, I just upgraded from a 750ti (hanging head in shame) to a 970.  Let me know how your setup runs.  I'm hoping to slide my old card in and use it along-side my 970 once I upgrade my mobo (I only have one PCI-E 16x right now - still hanging head in shame.)

Havos

Since the render has to fit on both card (ie for the GTX 970 it must be 4GB or less) then SLI does not give any advantages to iRay. Having said that, I would expect two cards to run close to double the performance of a single card for renders of 5-10 mins or more in length. For very short renders (ie below 5 mins) a fair amount of that is set up time, compiling shaders etc. so will not be much faster with two cards. I have a GTX 970, and I usually render GPU only as using CPU as well barely lowers the render time at all, and by not using CPU it makes it much easier to use my computer for other tasks whilst the render is running. Rendering CPU only takes roughly ten times the time of GPU only (and the PC is unusable in that period with CPU rendering).

In my experience at least, assuming the scene fits on all cards, the render time is roughly inversely propotional to the available CUDA cores, so the more you have, the better.

shokizm

I run 2 x gtx970s NOT in Sli - i get around 60-80% faster performance on renders upto 2048*2048 px - (I dont go bigger as i use gfx for gamedev work).
for example - 1024*1024 px - genisis2 female with translusent Resin shader (max 10,000 samples) down from 32secs-single card to 18secs-dual cards :)
it is a worthy investment - just dont run them in SLimode!