-   LaurV (
-   -   Custom water-cooling (

LaurV 2012-05-18 17:01

Custom water-cooling
1 Attachment(s)
C/M/2075/90 went down already on the market, since they announced K10, you can already get a M2090 for under $1K8, and they already announced $2K for a K10. This means we are looking at $2K5 for a K20.

I think I would skip kepler and wait for maxwell... GTX780 would be fine. Till then, I just made space in one of my boxes for another two fermis, by getting rid of the big air coolers... The new toy is (much!) cooler and (MUCH!) quieter.


TWO (soon 3) air coolers (complete, almost new, fully functional) for [URL=""]asus gtx580 dc2[/URL] (they are really "monsters", very powerful and efficient, but power-hungry and not really quiet, especially during the Thai summer in April, when I had 35 Celsius in the room), plus ONE [URL=""]cpu cooler[/URL] [URL=""]like this[/URL], are for sale for the best bidder.

LaurV 2012-05-19 09:14

[QUOTE=Karl M Johnson;299818]
Offtop: that water cooling solution, is that Zalman's ?[/QUOTE]
No, is kind of custom, the most of the parts were ordered from the web. As I said before, there is only one full-covering water block for asus gtx580 dc2 on the market, and that is [URL=""]EKFC580-DCII-EN[/URL] from EKWB. This fact dictated the rest of the shopping cart. CPU cooling is easy and you can buy everywhere. If you don't like or don't know searching around, then [URL=""]this sets[/URL] come with everything you need, the difference is the size of the radiator and the number of fans. I took the biggest (360 HF). The pump is really brilliant, if you put your hand on it then you can fill some vibration, but you can not hear any noise, no whistling, no grumbling... The radiator is so big (in fact thickness is the problem), there is no way to put it in any existent case (I have [URL=""]HAF932[/URL], which is huuuuuggeee), that is why you see the hoses running out of the photo. The plan was from the beginning to install a small cabinet on the other side of the wall (that is, outdoor, out of the house, shadow side) where the heat and the noises would go. Right now this is not done, because the designer was idiot (that is me), and the cabinet is too small. I did not expect that radiator to be about 7 cm thick, that would make about 11 cm thickness including fans and guides. The cabinet is done from aluminum profile, with special ventilation, but the space inside is only about 20x10x60 cm. As I said, I would need for the radiator and pump to fit in, about 14x11x(40+10(pump)). The "11" is the "odd" number here. I never liked 11, its mersenne is not prime either... :P
Right now the radiator and pump are still under the desk, I may put a new photo in few minutes.

Ok, the order contained also [URL=""]this mobo block toy[/URL], which is not yet installed, and you can not see it in the picture, but this time the idiot designer is on their side. I am not talking about compatibility. There is a known compatibility issue, as the block is done for maximus 4 extreme, and my board is maximus 4 extreme[U][B]-[/B][B]z[/B][/U]. They said (see the [URL=""]cooling configurator[/URL]) it is necessary to "bend" a capacitor to be able to mount the plate on the "z" mobo. This is not quite right, I had to desolder the capacitor and solder it differently, but I had also to mill out a part of the acrylic plate to step over a coil which was on the way, it seems as newer "z" mobos have a coil higher then older "z" mobos (asus replaced few components with new-technology stuff, like "super alloy", you can read on their page). This all was known when I bough the plate, and it was expected, I did all the measurements, and I am working in an electronic company. What was unexpected is the fixture system has no couterplate, and the original counterplates from the air cooling can not be used because they have pressed-in nuts fixed from the other side. In short, when you put the water block on the mobo and tighten the 6 screws on the edges, the PCB bends and the mosfets in the middle do not touch the block anymore. Imagine you put a plastic ruler 10cm long, 1cm wide and 0.5mm thick on the table. That is "the thermal pads". The table is "the water block", a massive metal (not bendable) piece. Then you put another ruler of 15cm OVER the first one, centered, in such a way that you have it few cm longer then the "thermal pads" in each direction. That is the PCB (the mainboard). Now you press the ends of the bigger ruler against the table. That are the screws. You will see that the "PCB" bends microscopically in the middle, no matter what washers or spacers you use for the screws, and this is accentuated by the fact that the thermal pads are "rubbery", elastic stuff, creating a pushing force. The mosfets in the middle will not stay in contact with the cooling block.

The original air-cooled stuff has counterplates (another rule placed on top of "our little experiment with two rulers", which is made of 2mm-thick steel, unbendable) but as I said, to use it I would need to drill out the pressed-in nuts and find some special (longer) screws (on the original the screws come from the other side). I did not want to destroy the original parts, so I will wait with the installation of the mobo water block till I can make the suitable counter-plates that would press straight on all mosfets and avoid bending the mobo, and also fit into the case :D

I will put some photos immediately I finish the things I am doing here around (I am writing to this post since 4 hours, something always interrupts me!)

p.s. and no, I would not get upset if a mod moves all this discussion in a new thread called "water cooling" or whatever...

LaurV 2012-05-19 11:28

1 Attachment(s)
It took me a while to find where the thread was moved :D

This is the radiator, which as you can see, it is using only one fan momentarily. This is a 3-wire fan, slow rotation, very silent, one of the new fans from the set (i.e. the fans coming with the CPU cooling kit linked in the previous post). The other fans I did not take out from their boxes yet, and I probably never will. The final mount will have different fans, with PWM (4-wires) as I like to "dictate" the rotation speed by myself, not only to measure it (edit: the mobo has plenty of 4-wire fan connectors). When I will start pushing the clock, I won't see this slow-fans too well...

As it is now (one fan, not really aerated under that box-like table) it can keep the temperature under 60 when CPU is maxed (prime 95) and one of the GPUs is 30-50%, with about 27-30 Celsius in the room. More than that 30% of the first GPU, or some small activity on the second GPU, and the temperature is raising (that is why my gpu272 output decreased in the last days, as I already explained). The consumption of all the stuff is much smaller then the heavy air coolers it substituted, but this means nothing yet, until the "real" fans will be in place and the radiator in the cooling box on the wall, outdoor. That time the consumption may go back to the original values, or even higher. This was just for the people who want to use "standard" set, and do not go very "custom". A simple kit with a smaller radiator, with only one fan, fitting into your case, would be enough to keep cool the most nasty CPU (no GPU!) for the (al)most nasty (over)clock. You do not need the big radiator and big pump if you want to cool only the stock CPU.

R.D. Silverman 2012-05-21 13:15

[QUOTE=LaurV;299833]p.s. and no, I would not get upset if a mod moves all this discussion in a new thread called "water cooling" or whatever...[/QUOTE]

Nice idea. Then we can all hang out by the water cooler.

pinhodecarlos 2012-05-21 14:43

As in an industry you should have two pumps in parallel in case one

LaurV 2012-05-23 05:22

[QUOTE=pinhodecarlos;299976]As in an industry you should have two pumps in parallel in case one[/QUOTE]
No worry, there are flow-meters, different sensors, pressure switches (we produce [URL=""]this stuff[/URL] in our factory, by the way, for Bosch and few other companies).

PageFault 2012-05-24 01:37

That is the ideal place to put it: outside of the house. Since you are in Thailand, you will get a much better saving - on your aircon, which has to transfer all that energy to the outside.

I was in Africa, ambients of 45 were common, in the house it was 55. Due to the high cost of power (1.25 $ per kW/h), we only AC the bedrooms. Just one box and a craptop severely impacted the AC's performance.

In Canada it is seasonal. I once heated my appartment all winter with computers alone. Only on the very coldest nights (-30 or less) did I turn on the electric heater.

The consumption of all the stuff is much smaller then the heavy air coolers it substituted, but this means nothing yet, until the "real" fans will be in place and the radiator in the cooling box on the wall, outdoor. [/QUOTE]

LaurV 2012-07-21 07:28

1 Attachment(s)
[edit by LaurV, 2012-07-23 GMT 16:28 - following 4 posts were moved from the GPU272 thread and the "yuppy" part of the posts refers to the gpu272 team being the first on top for the last 365 days, for P-1 and TF, that is why the discussion about ratings]

But.. that's for the last year. Lifetime is a bit different. [for example team's P-1 position in top is second! :P]
Anyhow, yuppy! :smile:

BTW, I put the team on the map, for ECM and ECMF: the total of 630 GHzDays are all mine, you see, I got this crazy idea to get back in TOP100 LifeTime for all categories, and that is why I pushed ECM, ECMF and P-1 during the last month... :razz: (well the reality is that I had available few "small" computers whose thermal behavior had to be tested).

kladner 2012-07-23 00:08

But.. that's for the last year. Lifetime is a bit different.
Anyhow, yuppy! :smile:.......[/QUOTE]

I am happy to be in the top 50 for TF, P-1, and Overall for the last 365 with just one second-tier box. Lifetime ATM, I am 43, 93, and 147 respectively.:grin:

EDIT: .....on PrimeNet.

EDIT2: @LarV: have you described the "small" computer in the photo? Just curious.

LaurV 2012-07-23 05:23

EDIT2: @LarV: have you described the "small" computer in the photo? Just curious.[/QUOTE]
That was just one "plate" from many different plates hanging here around. Its purpose is to study the heat transfer in heat pipes, and no other.

It consists from a 10mm aluminum plate of about 25cm x 50cm, assumed to digest the heat. The final system will be smaller, but packed in about the same volume\surface of metal, with cooling fins around. All the system "wastes" no more then 35-45Watts (no joke!) and the big aluminum plate does not get hotter then 50C. Mobo is a mini [URL=""]wade 8020[/URL] from Portwell. Later on, it will be our own, but now we use this for development and to study the thermal thingies (the last is the main goal, that is why the heat pipes, as I said). Passive cooling is a must, that is why all the fuss. We use heat pipes and special oils in between the blocks, for a good heat transfer, and some special measuring blocks, not shown in the photo (otherwise you would believe that all those heavy artillery are part of the computer).

The "vertical" position is... guess what! a "design error" :blush:. Logically, the heat pipes work better with the hot part down and the cool part up, as they have an internal evaporation-condensation cycle. When we designed the plate we did not consider that thing, and that is why the vise/jaws used to clip it to the cabinet's wall. The cabinet wall is wood, but, anyhow, we placed some layers of bubblebags in between, just to be sure the cabinet does not suck the heat influencing the measurements. :smile::smile:

[edit: there really IS a BIG difference in heat transfer when the system is horizontal and vertical, otherwise we won't go through this trouble!]

The CPU is an i7-620M, 4 cores (Hyper Threaded), TDP 35W (!!) , which sometimes (i.e. 4 workers of P95 doing P-1, environment temperature over 35C) gets very hot, around 85C-90C on top of the black plate (T_junction max is 105C) and we have to come down with it. Each mem stick in the picture is 4GB/1333MHz RAM for a total of 8GB from [URL=""]Transcend[/URL]. Power supply is a standard one (takes 24V input and splits it into all the voltages the board needs). What you see in the lower side is a "3.5 inch to 2x 2.5 inch" hot swap bay (chinese stuff, don't have a link, but is is close to [URL=""]this one[/URL], really nice thing!) which holds two [URL=""]Agility3 SSDs[/URL], 6Gbps each (won't be both in the final system, but it is wonderful/nice/amazing to see how fast they are together, even for a little system like this: they are much faster than the things I have at home in my "heaviest monster" 1.3 DP Teraflops machine).

[edit: correction, there is only one SSD connected for the board in the photo! This can be easily seen as only one SATA cable is connected]

Some of these "plates" are busy doing P-1 with 2-4 workers each. And they are VERY efficient on P-1. Much faster then I expected. Unfortunately I can't put them to DC, their "life" is quite limited, soon they are going to be dismantled, reshaped, hit in the head, kicked in the butt, etc.

edit: one supermod could move the last 3 posts (including my photo) to the "custom cooling" thread, maybe.

edit(CH): messages moved as requested. Although this thread was the only one with "custom cooling" in the title.

kladner 2012-07-23 13:28

Thanks for the description and explanation. You do some interesting work.

All times are UTC. The time now is 01:42.

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2020, Jelsoft Enterprises Ltd.