mATX 480SLI watercooled renderbox

So my build (sig) is shaping up to be mostly done by this Friday-ish(had to RMA motherboard and order a few more fittings as I ran into some snags getting things attached to the GPU blocks. The case is supposed to start being put together later this week, so probably 2 more weeks till its all done, but everything should be on the test bench by the end of the week (pics of course).

The idea is to have a luggable workstation as I'm studying to be an architect and move around a bit, baggage permitting I can take the system apart into 2 components (qmicra as a carry-on, external rad left at home/checked baggage) instead of trying to lug around my old 800D system which I just sold off.

I decided to go with the Koolance blocks vs full cover as I want to be able to upgrade later without having to buy new blocks, or even an adapter(MCW80). The rendering apps I use are starting to take advantage of CUDA, and maybe openCL, so I figure the i7 should be enough and sort of obsoletes cpu/socket changes... as long as the x16 pci lanes are enough I "should be okay" for a while.

Main loop:

EK-BAY dual-bay acetal spin resevoir (has 4 G1/4" plugs...)
Swiftech MCP655 w/EK-D5 X-TOP Version 2
Koolance 2.120 30FPI radiator w/2 20mm yate loons
EK-Supreme HF Gold (saved 10 bucks /shrug)
Koolance GPU-220 x2 (using 2 bitspower rotary T fittings and 2 L fittings to give me parallel flow through the blocks; also needed to use 45s on the blocks to get the T/L fittings attached otherwise they're too close, looks sort of like a V-engine block)

All of the above should will be mounted internally in the qMicra2. The Koolance was the thickest/densest rad I could get that would fit in the case. Figure it maaay have a chance at handling the 480's in sli with the cpu at at stock/undervolted in scenarios where I can't take the mora3 along.

Auxilary:
(same res as above, using the middle 2 plugs)
Swiftech MCP350 w/EK-DDC X-TOP Rev 2
Koolance VL4N quick disconnect fittings male+female x2 each
MO-RA3 LT 4x180mm w/classic grill
Silverstone AP181 x4 (running off a sunbeam 30w x6 fanbus)

The external rad can be mounted to my desk/hanging off the wall/standing on its own/etc... I think I will mount the pump to it somehow as well. Gotta work on some cabling to get power out to the fans + pump but shouldn't be too hard.

All running Tygon 3603 1/2" ID (3/4" OD) with Bitspower black sparkle compression fittings with some 45s thrown in where needed.

Should be fun... this is my 2nd watercooled build, first was in the above-mentioned 800D.

Will post some preliminary gear/etc pics tomorrow.

Epic... subbed

I love small, dual GPU builds

had issues getting the 24GB to POST... taking longer than anticipated to get everything tested before putting it under water:

Decided to go with full cover xspc razor blocks as the weight of the koolances plus all the fittings was starting to get worrisome. Mora3 and new blocks scheduled to arrive Tuesday.
Posted via [H] Mobile Device

Mora3 arrived a day late and I miscalculated the # of compression fittings I needed by 2. Waiting for last shipment from sidewinder of 5 more feet of tygon and 2 more bitspower fittings and everything will be here to hook up. The XSPC full cover blocks are mounted on the cpu/gpus and just waiting for the last bit of fittings to get everything hooked up.

It's not pretty as it's still sitting in the test bench, but looks like the qmicras should be shipping within a week or two (hoping...)

watanabe said:

Decided to go with full cover xspc razor blocks as the weight of the koolances plus all the fittings was starting to get worrisome. Mora3 and new blocks scheduled to arrive Tuesday.
Posted via [H] Mobile Device

Click to expand...

Good call. I really had no idea how you were planning to do the mounting and connecting of that thing at the same time. Sometimes simpler is just better.

Looking awesome, cant wait to see the finished product. A sub compact dual GPU monster. Turn the key and lets see.

Very nice setup!!! Should be one nice setup when you finish.

everything's setup and watercooled on the testbench case is supposed to be finished cutting out by end of the week, and then some assembly time.

can't get 24GB to post (although it all shows up in CPUID)... decided to go ahead and start to OC even though only getting 12/16GB ram.

so far:




also will be swapping out the 930 for a 950 (building a pc for friend, i bought this 930 right before the price drop so going to use my 930 in theirs and pay the diff for the 950)...

please let there be an update

moar pics plz

indokyne said:

please let there be an update

moar pics plz

Click to expand...

Having a real bitch of a time getting everything (or one thing) working.

1. All sorts of things to get 24GB recognized, RMA'd original motherboard, worked for a bit, installed wc'ing and now I'm back to 16GB (not really acceptable as I was up to 12GB usage just last night working on a simple project). I ordered another rampage gene iii for a friend's build and everything worked in there... swapped my stuff over (draining the entire loop/etc) and now it's showing 16GB again. Going crazy. I thought it might be due to overtightening the waterblock, but I've loosened it up quite a bit and still 16gb.

2. Pcdesignlab is taking a long time making the qmicra v2f: was supposed to be here by the end of August but still in the manufacturing stage. So everything's scattered across my desk/workbench and looks like crap.

Mora3 is a beast though, here are some pics (again with a 120mm fan in the front for size comparison:

Well, I bought another Rampage Gene III for a friend's build, tried out the 24GB in there, was working fine, drained loop/swapped everything over and was back to 16.

Of course when I put the 24gb in the friends build, it worked! Just swapped everything back and now I'm in windows, 24GB visible (for how long??) running at 200 bclk x 23 for 4.6ish GHz:



volts are high (1.5v) but temps are fine (40ish idle in OCCT) was getting 101/9c BSOD's at 1.4/1.45v, and also getting much less ram(8GB!) recognized at higher bclks so now I'm up for some stability testing.

came down to 1.45v at 4.2ghz (only using 21x multi) trying to lock in a stable OC. 1hr OCCT was no problem, just finished 30 minutes of furmark running on both GPUS @800MHz core/stock ram+ OCCT running in background, cpu was getting up to 80c, gpus went to 50 and climbed up to 60.

When the new board and 950 first arrived it posted at over 5ghz so hoping I can work my way back up.

Temps running furmark for 15 minutes + OCCT at same time... hard to make out but 60C on the GPU and 80s on the CPU:


1 more shot of the cards/blocks:



220BCLK x 23...



I can get into windows at 220bclk but seems the chip is hitting a wall around 4.4~4.6 GHz... I took the 240mm koolance radiator out of the loop for now as things were just too messy scattered all over the desk. Dunno if it'd help much/at all though as I turned down the fans on the mora3 to very low during the testing run and temps only raised about 5c... seems it has no problem handling the heat load.

edit:


seems the 950 wasn't really a great trade-up, was getting 4.6ghz with the 930 running a higher bclk/lower multi. Not supercomfortable with this vcore although temps are still fine...

Seems even just running it that high for 1 or 2 runs did something. I had dropped it back down to 1.475v ("safe" according to the geneIII BIOS) and was doing some stability runs in OCCT. Had it running for an hour and went to bed, woke up in the morning and the system had turned off.

Upon trying to start it back up the system would *immediately* power down (before fans/anything could spin up/bios post/etc)

Tried a few times and got a poof of acrid smoke and a small fire coming out of the 8 plug CPU plug. RMA'd the mobo, replacement has the same instant off problem (just tried once or twice) so I'm guessing its the CPU. So let that be a lesson to you all in your runs to 5GHz. Will see what the next 950 can do, of course not ever taking it over 1.5v, probably more like 1.45v even.

I doubt it was a Vcore issue, they are rated up to 1.55v according to Intel. How much Vtt/QPI did you give it? Considering you were overclocking with all of the DIMM slots used, I'm wondering if it was something else. *wink wink*

Edit:

What is up with the loop? Are you going res→D5→CPU→GPU→res and res→DDC→Mo-Ra→res?

80c on the CPU is VERY high on a water loop. Were those your temps at 4.5 ghz then?

QPI was abit high at around 1.425 iirc, but that's around where I've had to have it to get 6 sticks running in the past.

For the loop: Yes. I plan on housing the Mo-ra a bit away from the desk, and reading about how it's restrictive (tube design vs usual fins) with the CPU and SLI'd 480s I thought it best to keep it on its own loop, and just let the water normalize through mixing in the res. I've setup the in/outputs on the res to aid in this although it's probably not necessary:

(1)...(2)...(3)...(4)

1= output to D5->CPU+SLI
2= input (cooled water) from Mora
3= output to DDC->mora
4= input (hot water) from CPU+SLI

Idea is to keep things as turbulent as possible in the tank and mix up the flows. May have a little bit more noise running two pumps, but I figure that will be made up for from fan noise lost once I can mount the radiator a bit away from the desk.
edit: Also note the Mora loop has koolance VL4N qdc's in it, so I can unplug the mora and leave it at home if I need to lug the renderbox somewhere. The case has room for a 240mm rad so I'm hoping that I can setup a profile which uses stock/undervolting/etc to get me to the point where I can use this on location if/when necessary.


Right now I'm running at 221BCLK with a 17 multi, been stable for 30mins so far in OCCT. btw, temps are taken from the builtin monitoring there, which I've found is anywhere from 7-15 deg C higher than what I pickup in speedfan/other monitoring programs, and I figure it's best to go with those even if they are off, as a worst-case scenario. With volts at 1.475 (not necessary, but plan to scale up multi until I lose stability) I'm at 79-76-73-73 which seems to be as high as it's getting.

Well, after locking in an hour run of occt at 4ghz (167x24) at 1.375v/1.86pll/1.4qpi I could not get it to run stable at 4.2ghz (221x17, 221 was confirmed working fine w/lower multi) at 1.4v, 1.425v, 1.45 or even 1.475v.

Gave up and decided to go the other way: have been dropping volts back down. Right now I'm trying to lock in another 100% stable run at 167x24, this time at 1.275v/1.81pll/1.3qpi

Temps are much better, fans can be on the bare minimum for them to start spinning and its still peaking at 60c even on the hottest core. I'm guessing if I load up the GPU's with the CPU faster fan speed may help but for now it's nice to have the pumps be the loudest noise (along with some coil whine from the AX1200 when running linx+furmark at the same time, although redbeard is setting up a cross-rma to take care of that).

This setup will mostly be stressing the GPU's anyways, as I'm now able to do my renders on the GPUs instead of the CPU, so 4ghz vs 4.6~4.8 wouldn't really matter in my current usage scenario... would have been nice though.

Nice to wakeup to a still running PC:



Linx takes a long time to complete a pass with 24GB...