You might recall that Lucid had an MSI motherboard up and running at this last IDF. It featured the Lucid Hydra 200, a smart PCIe switch. The idea is simple, the Hydra 200 looks at PCIe traffic going to a GPU and parses the DX calls to one GPU or another intelligently based on several algorithms. The Hydra then recombines the images and sends them out. It all should be transparent to both the GPU cards and the user.
Last year at IDF, Lucid had a prototype working, and as you can see in the above demo, each GPU does indeed render a portion of the screen. One interesting thing I learned is that the gun is duplicated because they realized that most games render that first, so for this demo, they sent object 0 to both GPUs. Basically, the chip works, and there are a lot of interesting tricks you can pull with it.
Since then, the software running on the Lucid Hydra 200 has become much more sophisticated, and uses a more generic task division scheme. The newer one will do image division as above, time division, and other more proprietary schemes that Lucid isn't likely to talk about publicly. The software and firmware have come a long way since the 2008 debut.
Read the entire article in
SemiAccurate.
