1gig video pwnage
- Thread starter maxbb1
- Start date
- Status
One important thing to realize is- having 4 cards with 256mb of vram is NOT anything like having a 1 gig videocard. ALL the textures, geometry, vertex buffers and the like are required to be duplicated across all the cards, so it's indeed simply like having 4 cards with 256mb. The only thing that is not shared/cloned is each card's portion of the framebuffer- which is the smallest component.
Lastly, due to texture and geometry replication- this does have a big impact on bus traffic. As most games are texture and geometry heavy, this does take it's toll on performance by saturating the bus with all this data. It's extremely wasteful also since the same T&L operations are being performed on all cards GPU's as if they were singular/solo.
The only time multicard solutions have any positive impact is in the specific conditions of being heavily fillrate bound. In today's games, this means with extreme resolutions with antialiasing. For example, playing at 1280x1024 with 4xAA, you'd see a DECREASE in performance between 2x256 and 4x256 assuming semi-modern cards (like 6600GT's or greater). Too much overhead is introduced to replicate all texture/game data over the bus to 4 GPU's. You'd only start to see a return towards equalization and possible improvements as the resolution and AA samples go UP.. (say like 1600x1200, with 8xAA or 10x mixed mode antialiasing).
Lastly, due to texture and geometry replication- this does have a big impact on bus traffic. As most games are texture and geometry heavy, this does take it's toll on performance by saturating the bus with all this data. It's extremely wasteful also since the same T&L operations are being performed on all cards GPU's as if they were singular/solo.
The only time multicard solutions have any positive impact is in the specific conditions of being heavily fillrate bound. In today's games, this means with extreme resolutions with antialiasing. For example, playing at 1280x1024 with 4xAA, you'd see a DECREASE in performance between 2x256 and 4x256 assuming semi-modern cards (like 6600GT's or greater). Too much overhead is introduced to replicate all texture/game data over the bus to 4 GPU's. You'd only start to see a return towards equalization and possible improvements as the resolution and AA samples go UP.. (say like 1600x1200, with 8xAA or 10x mixed mode antialiasing).
- Status
Similar threads
