Originally Posted by tsm106
His overall loop flow would go up changing that equation wouldn't it?
No, or only by a tiny bit. His restriction in a CPU + 2x2 Quad GPU setup is the same as CPU + GPU. He's planning to use a MCP35X2 pump, so you wouldn't want to decrease the resistance of the loop any further or there would be no benefit. Also, the flow through the rads does not keep temps down on your GPUs, higher flow through the GPUs does.
People get so confused about series vs parallel so I may do a write-up later to help explain it all. The key is that you want to be in series unless going to parallel or partial parallel will increase your flow in the water blocks enough to warrant parallel. In the case of 4 GPUs, having 4 GPUs in series would create substantial restriction, so you try moving them into 2x parallel. The flow through the CPU block will be MUCH higher this way and the flow rate through each GPU block will be the same or slightly faster this way. Looking at 4x parallel, there would be very little if any CPU block flow difference and you've cut the flow to each GPU in half.
It is very much pump dependent, and you use your pump's PQ curve data to decide which setup is best. A pump does not have infinite pressure at 0gpm flow and does not have infinite flow at 0 pressure, so you want to figure out the best flow rate/pressure combination.
In this particular case, he has 6 radiators, a CPU block, and 4 GPUs. We'll look at 2gpm overall flow: 6xseries rad=2.4psi, raystorm cpu: 2psi, 2x2 series/parallel heatkiller gpus: 3psi; total=7.4psi; at 2gpm, the MCP35X has about 10.2psi, so you have room for the restrictions of tubing, elbows, etc and will still probably end up with a bit higher than 2gpm: likely about 2.3gpm, meaning the GPUs each have about 1.15 gpm through them.
Take a look at the MCP35X2 PQ curve:
MCP35X2 (Click to show)
At 3.5gpm, the Raystorm block is at around 4.9psi (the Raystorm Copper would be 6.5psi). That's higher than the pump can supply, and doesn't consider anything else in the system, so there is no way to get to the needed 4.6gpm to match the flow of the 2x2 GPU setup.
As a note, if you put all four heatkillers in series, you'd be able to run the loop somewhere around 1.3gpm, but obviously you'd have 1.3gpm vs 2.3 gpm to consider for your radiator and cpu block heat transfer.