101m4n

I neglected to update this as I've been quite busy with work. With the fans tuned to fairly quiet levels, I get about 35-40 idle in windows and 60 ish under full load, package power at full load is around 250 watts and it turbos to 4.25 on all core loads, which isn't bad at all! I'd tune it quieter, but as the GPU is open-air and not blower, I won't be doing that. At some point in the future I will probably replace the 1080 with something more recent and add it to the loop. Then I should be able to tune the fans to complete silence . The high idle temps seem to be because it doesn't clock down to idle frequencies when idle. This isn't an issue in linux, could be windows background nonsense keeping the frequency high, or it could be windows itself trying to control the frequency directly and making a hash of it. I don't know. I only use windows for vr and the occasional game that won't run acceptably in steamplay anyway so I don't really care. On the other hand, compiles go crazy fast now! They also use a lot of memory though. With the GNU linker I frequently get within a hairs breadth of paging on some of the projects I am working on at the moment. One interesting issue had is that steam appears to have some sort of shader caching mechanism, whereby when I start a vulkan game it will attempt to pre-compile the shaders in parallel, which will in turn use all the memory and swap and cause linux to start killing off processes in a desperate bid to keep the system afloat. So that was fun. Fortunately this feature can be turned off in steam settings. Otherwise it's been great! I hope to get many many years of use out of this system. Fingers crossed AMD puts the zen3 threadripper stuff on the same socket!

I mean like one of these:

Thanks! I'm afraid I was skimming and didn't notice the link. Welcome to the money-pit that is custom watercooling . In that case, the right angle fittings should be fine. Just make sure they aren't under tension when the tubes are installed and you'll be set . In the case of hardline, one thing I will suggest is to avoid tubes with more than one bend in them wherever you can. Like the one from your pump/res to the GPU in image 1, or the one from the GPU to the top radiator in image 2. In the case of image 1, If you rotate the fitting on the pump/res so that it goes straight up instead of off to the left, that bend will be a lot easier. Single right angle bends are easy, just bend and trim to match the fittings. Multiple bend runs are a nightmare, as it's very hard to control the distance between the two bends. You may be able to do this if you measure very carefully and use a jig of some sort to ensure a fixed bend radius, but if you are bending by eye it will be very difficult. I'm guessing you are planning on using spacers to connect the right angle fittings on the CPU to the top radiator? If so, that will save you a lot of time and effort. Another thing you can see in this picture is that we used a couple of rotary fittings to give ourselves some wiggle-room on the radiator -> CPU bend. So that's another tool you can use to make the tough bends easier. Lastly, wait for big navi (if you can). Even if it's not as good as nvidias offerings, if it's even a little competitive it could prompt nvidia to lower their prices a bit.

I'd use the first one as it keeps the slots below the GPU free for other expansion cards. I actually did a build with this exact layout a few years ago. It worked out pretty well (aesthetically speaking). Performance wise, it's probably better to do pump -> cpu -> gpu -> 420 -> 280 -> pump. IE: heat source -> exhaust radiator -> intake radiator. With your hardware though, these radiators should be plenty for your system regardless of the loop order . Also, are you planning on using hard or soft tubing? In either case I'd advise avoiding rotary fittings wherever possible, but especially so with flexible tubing. Reason being if anything puts tension on the fitting it will eventually cause it to leak as one side of the internal o-ring is compressed. Also if you are going with soft tubing you can definitely skip the right angle fittings on the CPU block. Just connect the compression fittings directly to the block As far as tube choices go, thicker tubes look nicer if you ask me. That being said, there is no appreciable performance loss with thinner tubes and thinner ones are much easier to work with. With thinner tubing for example, you shouldn't need the right angle fittings on the GPU block either. Lastly, your board actually has a 2 pin temperature sensor header. Which means you can get a g1/4 temperature sensor and set up your fans to ramp with coolant temperature rather than CPU temperature, which is very nice (especially with a GPU and CPU in the loop). Mine is a gradual ramp (20 to 40%) from 20 to 30 degrees, then an aggressive ramp from (40% to 100%) from 30 to 40. This would be noisy and annoying for a CPU cooler but it's great for a water cooling loop as the specific heat capacity of water is very high (it takes 4200 watts to raise the temperature of 1 litre of water by 1 degree in 1 second), and you don't really want your coolant getting that hot anyway. Now, I should probably get back to my work! Best of luck with the build

The evga card works. I have seen many reports of 10 series cards not posting in various motherboards and a disproportionate number of those issues seem to be with MSI cards. I pretty sure they just guffed up the bios on these cards somehow. Oh well, I suppose I'll just include a note and a return postage label with the card when I sell it. Here's the system now: Not without some more issues though (why not?). One of the things I wanted was fan control via coolant temperature. Fortunately the board supports this with a couple of headers for 10k thermistors. For reasons that defy all logic and sense though, certain fans can only be controlled by certain sensors. The sensor that suited my needs however, was sandwiched between the edge of a usb2 header and the bottom-most PCIe slot. I ended up having to solder some new leads on that would fit in the gap: Can't say I was thrilled about soldering this close to by far the most valuable computer I have ever owned, but hey ho. It had to be done. I'm sure the cardboard will protect it from the 250 centigrade molten metal . Time to get an OS installed :

Nope, same outcome. Edit* I found an evga card within tube distance in london on ebay for £230. I'm picking it up this evening.

Actually it was in the primary slot, but I had a second GPU (the rx550) in the other x16 slot with the display attached. I think I remember seeing something like that. I also tried a few different settings (like forcing pcie gen 3) etc without luck. I'll give this a go.

Hmm. I have tried several displays, a vga via an active vga to dvi adapter and a "full fat" display port to my displayport monitor. As for hdmi, I don't actually have an hdmi monitor or cable! I suppose I could go down the street and get one, then try plugging it into the TV... But the GPU would still be useless in normal use. I also don't think this will solve the problem. The system doesn't detect the card at all. In the bios, the PCIE slot registers as empty. I am currently hunting for other 1080s in the area that I can go pick up this evening. Then I can sell this card on ebay to recoup some of the cost. It's a shame, I really like this card. Too bad both customer service for MSI and gigabyte are a bag of dicks.

The system works now, but there are more issues. The first is that the fans have very little clearance on their front face and scrape against the metal on the radiator and case at higher RPMs. So I'm going to have to remount them with the spacers on the other side. I have done this for the front radiator, but the top radiator will be a little more difficult. I may have to drill the threads out of the nuts I was using as spacers. The other issue is the GPU. The machine simply refuses to post with it installed. I have tried other GPUs (an rx550, and a gtx680) and they work fine. I have tried the 1080 in other systems, and it works fine. I even tried it in an am2+ board from 2009 and it worked perfectly! I have measured the pcie power cables with a voltmeter to check that the card is getting power. It is. I have tried updating the motherboard bios. I tried the GPU with every past bios that has ever existed for it. No luck. This is actually a problem that I have had before with this card. Another board a couple years ago, an (also gigabyte) gaz68xp-ud3p also refused to post with the GPU installed. I'm beginning to think this is some sort of weird bios bug that only presents itself with MSI GPUs and gigabyte motherboards. It could also explain why there are so many MSI gaming X cards from this generation on ebay. And why they seem to be cheaper than the others. Whatever it is, it must be something pretty fundamental to have spanned 9 years of hardware. To top it all off, the post code I get with the card installed (0D) isn't even in the manual! Just fantastic.

Thanks! No I'm going for an all black build with ek zmt this time around, there are some pictures earlier in the thread. Did consider doing copper again, but I wanted something more practical (and lower effort ). Just as well, with the flexible tubes I didn't have to empty the loop to remove the motherboard! I'm also not watercooling the GPU, as it's an old card from another machine and I wouldn't mind a ray-tracing capable one at some point in the future. If I get a new GPU for it, then I will likely watercool that (which the soft tubes will also make easier). I'll be waiting for AMDs answer to rtx3000 before I make any decisions though. Incidentally, the copper build actually belonged to a friend, he moved out of the country at the beginning of the year and disassembled that system. I bought some of the components from him and they are actually being used here, namely; the pump-res, power supply and a few right angle fittings.

New board has arrived. This evening I'll get it installed, hopefully I can polish off the build without killing this one. Even at the cost of a new motherboard, I'm actually still £150 below retail price on the motherboard and processor. Fingers crossed the CPU isn't damaged.

That's what I was thinking as well. There could be a firmware aspect to this too though. I can imagine some microcontroller running off 5vsb somewhere doing pre-power-on checks, initialising devices on the board etc, then getting stuck at the dead one. Maybe that's just wishful thinking though. edit* it's also possible the dead chip is now a short and that's what's keeping the system from powering on.

Can't RMA. Gigabyte only does RMA through authorised retailers. It's not done by serial number like for example ASUS does. Also, as it's second hand I don't have proof of purchase. I have taken it out of the system now and found the borked chip: The serial number is 3947s934gb, made by nuvoton. There are several of them on the board, they appear just to be fan controllers. If that's the only damage, it might be repairable

Afraid not

Unfortunately I don't have one. I did attempt to measure conductivity with a plain multi-meter with probes set a fixed distance apart. The resistance is about double tap water, ish. I say ish because the measurement isn't very stable. It's certainly higher than industrially de-ionised water, but I wouldn't expect it to be as low anyway because the condenser was made of copper. It should be clean enough for my purposes though. Yup: This was towards the beginning as well, it was much worse than this at the end. We'll see when the gas bill comes in! I distilled about three and a half litres, it took about 8 hours total.