View Single Post
  #8 (permalink)  
Old March 17, 2011, 01:36 PM
SKYMTL's Avatar
SKYMTL SKYMTL is offline
HardwareCanuck Review Editor
Join Date: Feb 2007
Location: Montreal
Posts: 13,641

I'll expand a bit on what Arin said.

Basically, dual GPU technologies don't allow for a fundamental "doubling up" of performance in terms of memory utilization. AFR or Alternate Frame Rendering allows GPUs to work together to render a 3D environment but each GPU will render SEPARATE frames. In a dual GPU config, this means GPU #1 will render frames 1, 3, 5, etc while GPU #2 will render frames 2, 4, 6, etc. this is also why sometimes you see "flickering" as a driver error makes one GPU skip certain frames. This means even though each GPU has say 1GB of onboard memory, the total accessible memory at any one time is NOT 2GB. Rather, each GPU will have access to its own memory so the same bandwidth constraints will be in place but rendering times should (in theory) be cut in half or thereabouts.

Naturally, load balancing and partial resource pooling at the driver level allow for increased overall memory performance in every scenario but usually the same bandwidth limitations will be in place.
Reply With Quote