The Inq explains Lucid's "Hydra"

[spacemonkey]

http://www.theinquirer.net/inquirer/news/1406316/multi-gpu-tech-lucid-graphics-giants

"...Hydra splits each frame into different objects - say the game character, weapon, floor, walls - to be distributed among the various GPUs."

How is this possible for (what I assume) arbitrary game engines?

 

[MDolenc]

You can send one draw call to one GPU and other draw call to the other GPU for example. You could also do some load balancing in case one GPU is getting cheaper draw calls. Of course you'll end up with two weird uncomplete back and z buffers that you'll have to join (check which of the two pixels is closer).
Kinda like split frame render, but without double vertex shader load. However you'd still have to solve the off screen rendering that games do nowadays. And you wouldn't be more effective then AFR especially in games that are written with multi GPU in mind.

 

[cho]

http://www.nvidia.com/content/nvisi...zation/NVISION08-Does_Your_Software_Scale.pdf