Originally Posted by mcnabbmc
Will a 1090t bottleneck two GTX 570s? Just wondering as I might SLI two 570s as the new 990fx chipset supports SLI.
I will give my analysis below. It is a little long and detailed:-
benchmarkextreme has tested a Phenom II x6 1100T with GTX570 (single, SLI, 3-way SLI) for 14 game benchmarks at 2560x1600. One of the conclusions of their study is:
A stock Phenom II X6 1100T is sufficient to run GTX 570 and GTX 570 SLI at full potential.
2) CPU consideration
OP's 1090T is more or less the same as an 1100T, so we can treat them the same for the time being.
The only question left is OP's 1600x900 (as in sig rig) vs guru3d's 2560x1600.
3) Resolution consideration
When the resolution goes down from 2560x1600 to 1600x1200 (close enough to 1600x900), framerate will increase because the GPUs have less data to calculate. Using guru3d's another review of GTX570 SLI over 7 game benchmarks at 1600x1200, I calculate the average % increase in fps as:- (55+49+37+47+100+25+48)/7= 37%
This fps % increase will happen only if the CPU can keep up its pace to communicate with the GPUs as not to bottleneck them. So, can a stock 1090T keep up with this 37% fps increase?
Generally speaking, besides lowering the resolution, increasing the processing power of a graphics subsystem can also increase the framerate. In other words, we can simulate the impact on a CPU due to a lowering of resolution from 2560x1600 to 1600x900 by increasing the graphics power by going from GTX570 2-way SLI to 3-way SLI on the same resolution because both changes can increase the fps load.
Let us return to benchmarkextreme's study as they have such data available:-
For 14 game benchmarks at 2560x1600, the average % increase in fps is:-
If 1100T at 3.3GHz, then (44+38+10+10+11+12+36+ 8+26+45+45+26+5+38)/14 = 25%
If 1100T at 4.2GHz, then (44+44+19+30+23+28+39+16+39+48+48+42+1+40)/14 = 33%
4) My conclusions
As the 33% figure is close enough to the 37% figure, an overclocked (4.2GHz) 1090T can handle GTX570 2-way SLI at 1600x900 with minimal or no bottleneck. This is generally speaking, of course, as the 33% is an averaged number of 14 game benchmarks.
To be precise, I go to each individual game benchmark.
For those 8 games with % (44, 44, 39, 39, 48, 48) larger than 37%, it can be concluded there is no bottleneck.
These games are:- Aliens vs Predator, Bottleforge, Lost Planet, Mafia 2, Metro 2033, Napoleon Total War, Necrovision, Stalker Call of Pripyat.
For the other 6 games (19%, 30%, 23%, 28%, 16%, 1%), there is various amount of bottlenecks.
These games are:-Cryostasis, Crysis Warhead, Far Cry 2, Hawx 2, Lost Planet 2, Resident Evil 5).
If the 1090T is not overclocked but remains at stock 3.3GHz, then only 5 games will exhibit no bottleneck while 9 will have various amount of bottlenecks.
Overall speaking, even when there is a bottleneck, its amount is not huge, especially when the 1090T is overclocked to 4.2GHz.
So, IMHO, using GTX570 2-way SLI for 1600x900 with a 1090T is certainly a good choice.