I like this idea. Are you concerned about heat at all? I mean, from the CPU? You *might* be able to make use of an older dell case (because I think they have decently-designed cases as far as airflow goes)--to have a passively-cooled CPU HSF, relying on ducting and airflow generated by a single exhaust fan? Getting rid of a fan might allow you to go with a more powerful component somewhere else (though it's not like a fan consumes that much power in the first place). Might cause other temps to increase though, with less overall case airflow...
You may be right with the dell cases, but let me show something:
Originally Posted by A thread of mine of a few months ago
I am working on building for some time now a case for a really tiny server. What I expect him to do:
- file server: lots of space, data security
- all around connectivity
- low consumption, low noise
- no wires (except plug) and small.
- 2x500GB HDDs (brand unknown for now) in RAID 1
- connectivity Wi-FI (n draft), Bluetooth, eventually remote control TT Mozart style
- sempron 2800+ CPU passive, AMD 690G chipset, fans piloted by a electronic card I made in function of temperature (max supported =2 fans)
- home made wooden box with a metal frame
It's not exactly in chronological order as I've began it 2 months ago working now and then, but it's surely the logical order in which should have been done.
The main problem is heat - I have chose everything to enter a 150-170W power envelope and a minuscule volume.
The shopping chart:
Bought a cheapo 480W PSU (10$) that should provide 200w without overheating, an ASUS M2N-VM 690G, and reused 2GB of Ballistix DDR667, a 8GB old HDD and a 2800 sempron I had laying around.
Any ATX PSU is far to big for what I intended so I simply took out the PSU PCB out of the iron case; you'd be surprised how much space is wasted.
After this I bend the radiators to occupy a smaller volume - this doesn't change anything in their cooling abilities.
I removed the 120mm fan as there wasn't any room for it so the PSU became passive and as a result dead silent.
After this diet, the PSU is just the size of two cigarette packs one on top of the other - and it works (tested and retested every step of the way)
Sometime later, after mounting everything in the small case I built (details below) I realized the wires for the ATX 24 pin connector were too long and filling literally every cubic inch of air available in the case. They lost some weight also, some 8cm from the 15cm in total(4" from say 8" for the imperial units) in total to be just as long as they have to barely reach the ATX plug in the mobo.
The principle of the case is in the picture below. It has to be just a bit taller than a PCI card, and sufficiently wide as to receive a 25x25cm ÂµATX board.
Since the server needed to be really small, silent and more than anything cost effective it seemed obvious from the get-go that I had to build it myself. I had no experience in welding, but fortunately I have friends, beer and jokes in my schedule, so a total of 3 guys were enough to cut some iron plates that were to form the skeleton, weld them together to have the frame and cut the wood plates that were to silence the beast inside.
And a lesson to others reading and living in a flat: always a keep good relationship with the neighbors, 'cause they can get really pissed when hearing an electric drill at 10 p.m. If you have invited them to you barbecue the other day, they'd be easier to accept these phonic aggressions that are inevitable when building something.
I'm a highly mathematical person, I like to calculate everything in the smallest of details, the angles, distances so that there is a perfect fit. It just so happens that from the start of the project I supposed the PSU rad and the CPU rad would be face-to-face and touching.
Obviously, the hours of thorough calculations were brought to nothing when we realized that we forgot to take into consideration the effective thickness of the wooden plates. They're only 4mm thick, but enough to make us miss the overall inside height of the case by 5mm (more or less). This in turns meant some more modifying of the PSU's radiators...
Cooling as it was originally thought
The idea was that a 80mm fan I had laying around to pull hot air from the area that contains the radiators of the PSU/CPU/NB . It's sorta like a tunnel where the fresh air enters thru somewhere (not exactly defined in a first approach), is being routed in one end of the box, traverses the radiators that are packed together and is being pulled outside by the 80mm silent fan, much like this picture below:
I figured I'll see where to make the intake hole for the air to get in as the case is in wood after all and holes can be made easily. The idea was the hole to be where the eventual hotspot is.
Well, having worked for two months I was burning to see all the stuff in the case and install a Win2000 SP3 copy I have since a while ago when my main rig wasn't powerful enough to handle the "enormous memory requirements of windows XP".
BTW, Vista a memory hog? When XP appeared, the 128MB of ram became highly insufficient. 256 minimum for XP. XP SP2 upped to 512 minimum with 1GB to be smooth. In total there was a 4x multiply in ram needs over XP's lifetime. Vista say it needs 4GB to be smooth, so why all the hustle? And what did XP bring, some more eye candy vs Win2000. Same old story...
Now in the assembly, as stated earlier, I quickly found out the ATX cables have to be put on a diet. I welded for over 3 hours, since I'm not very good. I didn't want to make any mistake, as it would fry the mobo 'cause that would have sucked.
All smooth, nothing to see Except the fact it was installed in open-air (aka no case) because I needed a PATA cable connected for the DVD-ROM; the server won't have any optical unit.
A small issue that wasn't so obvious at the beginning was the lack of drivers for the 690G chipset in Windows 2000. Some XP drivers worked, others didn't...
Now the fun part: cooling design. Remember I have only one fan that is pulling air. From where? Dunno, probably from the hinges of the wooden case. The only purpose of this test was to identify the major heat sources. Anyway, after 20mn of 100% CPU it went at 47Â°C and still rising. Shutdown and back to the drawing board.
There seemed obvious 2 hotspots:
The CPU was the hottest thing around, the PSU rads didn't seem to hot. In the lower floor where the HDDs are isolated (in this test it is actually only one HDD) it heated up a bit too.
Now there are ways to tackle this issue and I thought of two scenarios pictured below:
1. Put a 80 mm intake fan in the right, output in the left and venting holes to ventilate the hdd are
2. Put a 80 mm output fan in the right, seal on the left and in the HDD bay area make a hole below the HDDs for the air intake, and make a hole in the left, so that the warm air rises and further cools the array of radiators; here I suppose the HDDs will not overheat since I suppose I'm getting Caviar GPs:
What do you think of the build, the story, the server, and more importantly, how do you see the cooling system to be as efficient as possible?
Thanks for reading, if you got this far you deserve a beer or at least an OCN's rep.
[Eventually] The cooling solution chosen was:
I have tried a compromise to cool both the HDDs and the upper electronics part. This is a done deal, the mini-server works.
Observations on the version 1.0 of the case:
- From the picture above it's not easy to realize, but the ATX wires obstruct the airflow, they are perpendicular to the flow.
- The sempron 2800 is hot. Since it doesn't support C'nQ it's at 1600MHz and 1.35V all the time (~60W dissipation)
- The 480W modded PSU works fine. However the rads on the PSU are hot. This leads me to thing the efficiency of the 480W PSU @150W is horrendous.
- The HDDs are quite warm: 45-47Â°C. In my room there is around 25Â°C which doesn't help either, but the 20Â°C delta is quite a lot.
- These heat outputs have forced the choice of two fans to cool the thing. Despite this, the CPU is @57Â°C, mobo @43Â°C and HDDs@47Â°C
- The fans run @1000 rpm, but the siffle of the air is noticeable.
The version 2.0 of the case:
From this version I planned a few improvements:
- Remove the ATX PSU from the case. Power the server with a laptop adapter and a 12VDC/ATX converter. These things are limited to 120W thus this is the new power consumption limit of the server.
- The case is quite tall, I would like to reduce its height. Removing the PSU will help do just that.
- HDD's will be moved in the place of the PSU, thus teh whole lower compartiment will disappear.
- Replacing the Sempron 2800 that does not support C'nQ with a CPU that does support C'nQ, thus can be underclocked @800MHz. This alows me to fit the power envelope.
The idea is to replace the design of the V1.0 with the V2.0:
It's a so much simpler design that it will probably allow me to work in only one fan at significantly lower temperatures, or at least in two fans at lower temps
I should have probably begin with this in the thread, and then come down to each element's power consumption. The idea is that I want the design to be the one in the 2.0 picture which implies the usage of a laptop source + 12VDC/ATX adapter and this in turns adds the 110W limit.
The discussion was moved on another thread. Click here to see all the build log.
Edited by dragosmp - 6/28/08 at 2:16am