Block order experiment

[Scoob]

Hi all,

I re-plumbed my loop the other day as an experiment. You see, my loop has always gone Res > Pump > Rad > GPU’s (parallel flow) > CPU and back. Worked great and looked very neat.

I constantly read, and people tell me (“you’ve plumbed it WRONG!”) that I should ALWAYS go CPU first. This does make sense as of course the CPU puts out a lot more heat than the GPU’s. However, my view was that in a well-designed build, with good flow – Parallel flow on the GPU’s is crucial here – it really shouldn’t matter at all. Basically, there’s ample thermal capacity in the coolant, regardless of block order. Ok, we know the efficiency of the thermal transfer to the coolant is reduced the hotter the coolant becomes; however, as the coolant never really gets “hot” it’s just fine. Plus remember that the block before should never “heat” the following block, unless something has gone REALLY wrong lol.

Other people have said to me – and even built their loops this way – that I should be going rad > Block > rad > Block and so on, adding additional rads to the loop. Again this might help a little IF you’ve reduced the efficiency of the coolant by allowing it to get a bit too warm. However you would be reducing flow by adding more restrictions which might even compound the issue. You might choose to overcome this by adding an additional pump – but this can be untidy plus it will add its own heat to the loop potentially.

For me, I wanted to build a neat loop. I considered extra rads to be messy, plus there really wasn’t room in my CM 690 II. Originally I’d though I’d just be cooling the CPU and one GPU – so a couple of 240’s in the case would be fine – but after going SLI (initially with 2x GTX 570’s) there really wasn’t the cooling capacity and things would have been TIGHT inside that case. An external solution solved all my problems in one go – plus it seemed like a fun project

So, my loop has been running great since I re-built it last – I got 2x GTX 680’s a while back as one of my GTX 570’s stopped playing nice in SLI, though it was just fine on its own, go figure. Cooling my 2500k @ 4.8 (@1.45v – it’s not a lottery winner! Lol) and my two GTX 680’s at 1.2ghz core with 7ghz vRam just fine. Remember, my GPU’s are first served with coolant from the rad.

I was a little bored yesterday so decided to experiment a little and reverse the flow of my loop (properly re-connecting things so “in” on the CPU block was still in of course) and see if it improved my CPU temps at all. It didn’t take me long to do this, thanks to the QDC’s I used in my build (my radiator, res and pump are external to the PC case remember) and, after a quick leak test and air bleed (shake) it was all ready to go. Yes. I have faith in my work so don’t do over-long leak tests these days.

I started off with the usual OCCT AVX tests using 50% Ram (50% = 12gb for me) and left them running while I did stuff on another PC. Checked back and my CPU temps were the same as ever at about 40c over ambient (it’s a Supreme HF CPU block – fairly middle of the road) and my GPU’s (unloaded) were idling at their usual 5-6c over ambient. Note: that’s 40c over ambient when being hammered by OCCT. During normal gaming, on a fairly demanding title – say Skyrim which (modded) likes lots of CPU and GPU – I’d see about 25-30c over ambient usually. So, OCCT is a harder CPU test than pretty much any game.

Next I moved on to some GPU testing, I use the Valley Benchmark as it provides a good level of load. After testing and letting it loop for a while, I checked my temps. As they were before my plumbing change, they were hovering around the 15-16c over ambient peek.

Now, my loop is overkill – just how I like it – I have a 1080 (360x360) rad mounted externally with four 180mm silent (really, my laptop is far noisier) shrouded fans, a cylinder res and a D5 variable pump, and I think it’s this that gives it the thermal capacity that such plumbing changes make no difference. This was my plan! Incidentally, the loop with the res full takes almost exactly one litre of coolant.

Oh interestingly, when I was bleeding/leak testing and you could see the flow of the coolant thanks to the little bubbles, the flow on this loop is EPIC, I’d forgotten quite how good it was. Really, if I’d plumbed the GPU’s in series the flow would be hugely restricted compared to this. I’ve no doubt it would work fine, but the coolant would be sucking up more heat moving more slowly thus possibly I’d see increased temps due to the efficiency reduction. Not gonna test that though! This is why parallel flow is great, my loop is LESS restricted with two GPU’s than with one. Three GPU’s would be less restrictive still, but I think the CPU Block would then become the “slow” part in that case.

Not sure if this is of interest to anyone, just thought I’d share my playing around

Cheers,

Scoob.

 

[SparkleDJackson]

good read. i had always wondered if the flow rate would drop off on one or more paths if splitting over multi gpu setups in my mind the closet to the source would be a little more powerful as it has less distance to travel. like when honda introduced the twin muffler system to the crf 250 i think it was. the exhaust port on the what would be normal silencer had a smaller I.D. than the I.D. of this second silencer to balance the flow by adding slight resistance to the straight run. so i guess this concept has formed my thinking up till this point

 

[SPS]

Haven't heard from you in a while Scoob.

Where's the pics of this so-called overkill loop?

 

[andrewmchugh]

Only thing that would say about loop order is that its nice to have the liquid flowing upward. When I did my first loop I went down through the cpu and gpu and i had to bleed it upside down, which was a pain. Second time round I pumped out directly in to the GPU and it bled in no time at all.

 

[B NEGATIVE]

good read. i had always wondered if the flow rate would drop off on one or more paths if splitting over multi gpu setups in my mind the closet to the source would be a little more powerful as it has less distance to travel. like when honda introduced the twin muffler system to the crf 250 i think it was. the exhaust port on the what would be normal silencer had a smaller I.D. than the I.D. of this second silencer to balance the flow by adding slight resistance to the straight run. so i guess this concept has formed my thinking up till this point

Parallel=half flow rate thru the blocks.

And pointless.

Loop order makes no difference,much like dual looping stuff...also pointless.

The thing to look for are short tubing runs in loop design and res before pump,thats all.

OP,the CPU block is always the choke point in regards to restriction,thats the nature of impingement blocks. You flow needs to be tailored to that block (or in my case,blocks),its the most thermally sensitive part of the loop,it will hit the 'wall' before any GPU does.

 

[Scoob]

Hey all,

Oddly only just got a thread reply notification!

@ SPS - yes, I've not been around so much of late, various reasons, none of them interesting... As for pictures, well, I took some but that was with my crappy camera before it broke...which broke sometime after my GOOD mobile phone (the camera in it BETTER than my crappy camera!) died on me. Not wanted to spend out on another camera - yes, yes, I know they're like £50 for a half-way decent point and click, but I'm feeling tight of late lol.

In my loop, from testing I've done, the CPU block is far less restrictive than a single GPU block - the CPU block is marketed as "High Flow" so this is reasonable. Going to parallel flow with two GPU blocks allows the CPU block to flow at its maximum rate, so, the CPU block is likely the slowest point in the loop now. Parallel flow only = half the flow rate if the potential flow-rate happened to already perfectly match the potential restriction. With a high-flow pump like mine, parallel flow makes a HUGE difference. I.e. if I turned my pump down a few notches and ran my GPU's in serial, then the GPU blocks would not be restricting the flow overly. Once I turn the pump up however, those GPU Blocks in serial become a restriction. I use the larger 1/2" ID tubing throughout the loop - the exception being the smaller ID SLI connectors, but as the flow is parallel flow remains high.

My goal, well one of them, with this loop was to build something that flowed very well indeed. So, things are efficient and the pump isn't straining. It works well, and is still near-silent. With the coolant moving quickly through a large rad with slow 180mm fans fitted, it works great. We observed - on a friends loop that had a blockage - that a straining pump is a lot louder and runs quite hot. Both things bad.

My "flow testing", such as it is, is all done by eye. My view is if something is suficiently faster flowing to look faster flowing then it's significant. Obviously once the bubble disappear, you cannot see the flow at all.

Loop order does indeed make no measurable difference in my case - sorta the point of the exercise, well, proving it one way or another - though loop design will of course. So, short neat piping where possible, good blocks, nice flowing rads all backed up by a good pump. It all helps.

Re: bleeding the loop, I've never really had any issues at all. Currently the coolant goes in the top (CPU) then down to the lower GPU block before going up and out. Might sound odd, but it is the neatest way. Previously it went into the TOP GPU, out the lower GPU, before going back up to the CPU. Both variations bled out really well - just a couple of light shakes of the case released the air moments after first power up. I was tempted to drill some extra holes in the case to make the tubing even neater, but there wasn't really the space to do so.

I will try to get some updated pictures up, the only ones I have were during the initial design - which was compromised by my two (both EK) GTX 570 blocks NOT lining up with each other. This resulted in me having to bodge it...nasty.

Cheers,

Scoob.