Does anyone know if Cycles multi GPU with NVidia runs Implicite or Explicite LDA?

Support Technical Support
#1

As some of you know, NVidia is dropping support for more than 2 cards in Implicite LDA. However, applications developed in explice can still control more than 2 cards because they are handled by the application. (DING DING… BTW, if you were scarred your new 1080 wont support more than 2 way SLI, it does. Lots of misinformation about this. It has to do with how the game or application is developed, NOT the cards themselves. There is however an application from nvidia you will need to unlock the third, or more cards in order for them to be functional in explicite).

So, my question for everyone here is: Does anyone know if cycles runs implicite or explicite for multi GPU rendering? Since you can chose which cards to render with I wuld think explicite but since im not sure id really love to ask before I buy 3 all at once vs waiting on the third.

#2

SLI (or LDA Implicit) is not recommended for cycles and will significantly effect in a bad way the render speeds.

The way cycles uses multigpu is all cuda based… so its not the normal sli bridges or sli mode that is usually recommended for games.

It should be totally fine.

#3

Ok cool. So just to confirm, blender should recognize the third even tho it has nothing directly to do with sli? Thank you so much for your response! Your help means a lot.

#4

Blender recognizes all CUDA capable cards in the system, no matter how many and no matter if SLI is used or not.
However, with any number of cards SLI should be disabled for GPU rendering, as it creates so much overhead that you will see a massive performance drop when compared to the very same cards without SLI.

#5

Gotcha. Thats good to know. If SLI doesnt even matter, does it even matter if I run them on a bridge? It doesnt sound like it would even matter that the 1080 doesnt directly support 3X+ way SLI out of box.

#6

No need for a bridge. Just plug the cards into the mainboard and you’re good to go.
Blender sees all CUDA cards as individual computing devices and addresses them parallely anyway.

#7

Awesome. Thank you so much for the help guys!

#8

Im not 100% certain about the bridge… whenever we have had dual cards we havent used a bridge… when we have used a single card dual chip (ie 590) we had to disable any sli bridging and what not in the nvidia settings.

I am pretty sure that bridging really only is suitable for games, where the cards need to communicate with each other directly… rather than communicating to the cpu (the cpu manages which tiles each gpu is rendering)

if you want good cycles performance, disregard any sli / bridges / gaming related tweaks.