I was looking at this article because I am interested in the 5960X CPU but some of the benchmarks in it really shocked me. They were using a pair of GTX 980s at 1080p, 1440p and 2160p to check SLI scaling.
For 1080p and 1440p the scaling looked fine and what you would expect, 1080p was showing the effects of a CPU bottleneck and 1440p was getting very good scaling as there was no CPU bottleneck so no problems there.
What really stood out though was the poor scaling @2160p and the associated loss of performance, at this resolution there is absolutely no CPU bottleneck with 2 cards so why was the scaling bad?
They had done some runs with the Heaven 4 bench which was useful as it makes comparing cards on different PCs and CPUs easy as the CPU (providing it is a modern intel quad or better) does not really effect the result.
So what I decided to do was test some 290Xs on Heaven 4 using the same 2160p settings they used to compare scaling. For anyone who does not know Heaven 4 is NVidia's home ground when it comes to benching and their cards are very hard to beat on it.
Below is the settings and results the review site got for a pair of 980s
http://www.ocaholic.co.uk/modules/smartsection/item.php?itemid=1461&page=2
Here are my 290X scaling results using stock cards and my 4930k @4.0ghz. The same Heaven 4 settings as they used @2160p
1 card
2 cards
3 cards
4 cards
So for my 290Xs the scaling was -
1 Card = 377 Scaling = 100%
2 Cards = 748 Scaling = 198%
3 Cards = 1113 Scaling = 295%
4 Cards = 1446 Scaling = 383%
So what is going on here, not only is the scaling a lot better @4K on my 290Xs but my 2 stock cards managed to beat a pair of slightly overclocked GTX 980s on NVidia's home ground @4K.
Have NVidia missed the bus?
For 1080p and 1440p the scaling looked fine and what you would expect, 1080p was showing the effects of a CPU bottleneck and 1440p was getting very good scaling as there was no CPU bottleneck so no problems there.
What really stood out though was the poor scaling @2160p and the associated loss of performance, at this resolution there is absolutely no CPU bottleneck with 2 cards so why was the scaling bad?
They had done some runs with the Heaven 4 bench which was useful as it makes comparing cards on different PCs and CPUs easy as the CPU (providing it is a modern intel quad or better) does not really effect the result.
So what I decided to do was test some 290Xs on Heaven 4 using the same 2160p settings they used to compare scaling. For anyone who does not know Heaven 4 is NVidia's home ground when it comes to benching and their cards are very hard to beat on it.
Below is the settings and results the review site got for a pair of 980s
http://www.ocaholic.co.uk/modules/smartsection/item.php?itemid=1461&page=2
Here are my 290X scaling results using stock cards and my 4930k @4.0ghz. The same Heaven 4 settings as they used @2160p
1 card
2 cards
3 cards
4 cards
So for my 290Xs the scaling was -
1 Card = 377 Scaling = 100%
2 Cards = 748 Scaling = 198%
3 Cards = 1113 Scaling = 295%
4 Cards = 1446 Scaling = 383%
So what is going on here, not only is the scaling a lot better @4K on my 290Xs but my 2 stock cards managed to beat a pair of slightly overclocked GTX 980s on NVidia's home ground @4K.
Have NVidia missed the bus?