Odd Resolve Standard Candle Results

Hey Guys,

Just wondering if anyone else has seen this:

I'm running Resolve 9.1.6.009 on a 12 Core Mac Pro with 12GB of Ram, running 10.8.4

The fastest standard candle results I'm getting are a single GTX680, without a GUI card.

Here are my results - they were actually the opposite of what I was expecting.

GTX680 used as combined GPU & GUI :
04 Blur Nodes 24 FPS
08 Blur Nodes 24 FPS
16 Blur Nodes 14 FPS
32 Blur Nodes 07 FPS
64 Blur Nodes 04 FPS
01NR 2 Nodes 11 FPS
03NR 2 Nodes 04 FPS
06NR 2 Nodes 02 FPS

GTX680 used as GPU only
Quadro 4000 used a GUI, Y-Split power cable used, resolve set to NOT use GUI as GPU.

04 Blur Nodes 22 FPS
08 Blur Nodes 22 FPS
16 Blur Nodes 15 FPS
32 Blur Nodes 08 FPS
64 Blur Nodes 04 FPS
01NR 2 Nodes 11 FPS
03NR 2 Nodes 04 FPS
06NR 2 Nodes 02 FPS

So, here we see they adding a GUI card actually stopped the system from hitting 24 FPS at any level, as well as making the fans run louder, and using up a PCIe slot

Just for a test, I also tried setting resolve the use the GUI as a GPU - it actually made the entire system slower again.

Finally, I tried what should be the current "Holy Grail" configuration for a machine without an expansion chassis, using a GTX570 as a dedicated GPU card, and the Quadro 4000 as a dedicated GUI.

04 Blur Nodes 22 FPS
08 Blur Nodes 21 FPS
16 Blur Nodes 11 FPS
32 Blur Nodes 06 FPS
64 Blur Nodes 03 FPS
01NR 2 Nodes 12 FPS
03NR 2 Nodes 04 FPS
06NR 2 Nodes 02 FPS

So, basically, a single GTX680 is running faster then the combination of the GTX570 with a Quadro 4000.

This seems counter intuitive to everything I have heard previously, has anyone else seen this before?

Thanks,

Simon "Smash" Ashby

What PCI slot did you have the cards in?

GTX 680 was in Slot 1,
GTX 570 was in Slot 1,
Quadro 4000 was in slot 2.

GUI monitor was plugged into slot Quadro 4000.

Quadro 4000 was removed when testing the GTX680 by itself.