I don't know how much this helps, but according to Overclockers.com, their overclocked 1080Ti system (with a 7700k@4.2) used 411W running Fire Strike Extreme, while their 290X system (with a 4770K) needed 488W for regular Fire Strike when OC'd. As such, heat output should be comparable or lower for your new cards, there will just be half as many of them. I don't know what CPU you are using (or its clocks), but it should be possible to estimate loop temps from calculating the reduction in heat output vs. the reduction in radiator cooling performance. Of course, changes in flow rate due to two less blocks, one less rad and one less pump throws quite a few wrenches into the gears of estimates like this, but I'll give it a try. Given that the rad you want to get rid of is 9*120mm, that's very roughly equal to the rest of your loop (at least all your rads are the same thickness, I didn't look up fin density and the like). However, it's only set up in push, while the rest is push/pull, so it's not responsible for quite 50% of the cooling. Let's call it 40%. That depends on fan speeds too, of course. If your four 290x's each consume ~350W, and you have a 200W CPU, that's a total heat output of ~1600W max (what PSU do you have in that rig? That's pretty bonkers). Now that's being reduced to, say, 900W, a 44% reduction. Given that Pascal cards are inherently more efficient than Hawaii cards (both idle and load), I'd say that sounds reasonable to me. That should match your reduction in cooling potential reasonably well too (if the change in flow doesn't mess too much with things). So my napkin math says you should be okay. But then again, napkin math isn't an exact science, and there's a whole heap of assumptions in my calculations. For example: does 4-way XFX stress all four GPUs fully? If I were you, I'd try it out. It's probably the only way to be sure. Put some quick disconnect fittings between your case and the external rad, and it should be easy enough to test out.