Just ran a few benches @4K with 1 and 4 cards @stock.
Max settings were used as it shows GPU scaling best even if playable fps are rubbish.
No overclocking was used as we are looking for scaling not playable fps.
CPUs and GPUs used
3930k @4.0ghz + 1 and 4 Titans.
4930k @4.0ghz + 1 and 4 290Xs
No 2 and 3 card runs tested as 4 card results will give a good idea what less cards are capable of. Also it is difficult for me to get to the SLI bridge on the NVidia cards.
Tomb Raider
NVidia
Single card
4 cards
Scaling = 3.94x
AMD
Single card
4 cards
Scaling = 3.83x
Sleeping Dogs
NVidia
Single card
4 cards
Scaling = 3.83x
AMD
Single card
4 cards
Scaling = 3.86x
Sniper Elite V2
NVidia
Single card
4 cards
Scaling = 3.81x
AMD
Single card
4 cards
Scaling = 3.93x
Thief
NVidia
Single card
4 cards
Scaling = 2.91x
AMD
Single card
4 cards
Scaling = 3.05x
As you can see from the games used it is very very close as to which is better and well within the margin of error to choose a winner.
Some things that could have a tiny effect on the result were -
Frame pacing was not used as I was after the scaling not the visual appearance.
The AMD cards were on a slightly more efficient CPU.
Max settings were used as it shows GPU scaling best even if playable fps are rubbish.
No overclocking was used as we are looking for scaling not playable fps.
CPUs and GPUs used
3930k @4.0ghz + 1 and 4 Titans.
4930k @4.0ghz + 1 and 4 290Xs
No 2 and 3 card runs tested as 4 card results will give a good idea what less cards are capable of. Also it is difficult for me to get to the SLI bridge on the NVidia cards.
Tomb Raider
NVidia
Single card
4 cards
Scaling = 3.94x
AMD
Single card
4 cards
Scaling = 3.83x
Sleeping Dogs
NVidia
Single card
4 cards
Scaling = 3.83x
AMD
Single card
4 cards
Scaling = 3.86x
Sniper Elite V2
NVidia
Single card
4 cards
Scaling = 3.81x
AMD
Single card
4 cards
Scaling = 3.93x
Thief
NVidia
Single card
4 cards
Scaling = 2.91x
AMD
Single card
4 cards
Scaling = 3.05x
As you can see from the games used it is very very close as to which is better and well within the margin of error to choose a winner.
Some things that could have a tiny effect on the result were -
Frame pacing was not used as I was after the scaling not the visual appearance.
The AMD cards were on a slightly more efficient CPU.