To give just a little background, this rig is my wifes machine and also serves as router/firewall/print server. It's on 24/7 folding for SPCR. It sits right next to my desk, and as I'm frequently working at night with a low ambient noise level, the very first priority was low noise (duh!). I don't mind somewhat high temperatures, as long as its stable - I'm not afraid of running in the sixties (degrees C, that is) for my XP. The rig needs to be reasonably dust resistant (or else I would need to vacuum our place more often ). And I didn't want to replace any of the major parts I had (HSF, PSU), as these are fairly recent for my standards (a system lasts about 5-6 years for me).
UPDATE: after some people had difficulties accessing the pictures here (hosted at imgspot), I added links to the same pictures at ImageShack (thanks for the hosting!). The linked pictures are identical and not higher res, so if you can see the hotlinked ones, there's no need to click on images.
- Antec Sonata
- Seasonic SuperTornado 300W RevA2 modded w/ Nexus 120 mm Real Silent Case Fan
- AMD Athlon XP 2500+, stock speed, undervolted to 1.45V nominally, 1.51V as reported by lmsensors
- Zalman CNPS 7000A AlCu, modded w/ Nexus 92 mm Real Silent Case Fan
- Arctic Silver 5 thermal compound
- Asus A7N8X-E Deluxe v2.0
- 2 x 512MB PC3200 Ram (don't know the brand or model, it just says "Platinum")
- MSI FX5200 (don't laugh, I don't need 3D power)
- Seagate Barracuda 7000.7 120G SATA
- Plextor PX708-A (DVD burner)
- Nexus 120 mm Real Silent Case fan
- Zalman Fanmate2
- AcoustiPack noise absorbtion foam, precut damping kit for Sonata
- BeQuiet! noise absorbtion foam, Universalkit Big Tower
- Adhesive window sealing foam tape
This picture gives you a rough overview of the system. I've made an attempt of highlighting the airflow pattern I tried to achieve:
I made the compulsory intake bezel mod as radical as possible without affecting looks too much. In addition I put acoustic foam around the front connectors to prevent dust from building up around there.
The second layer of the intake bezel has not been spared, either:
Nor the third (steel) layer behind the (stock) intake filter. Hopefully I'll get around to putting some rubber sleeves on that edge sometime (I've only cut myself once so far, maybe that's not enough):
To improve air intake from below, I raised the case slightly by putting it on halves of foam training golf balls. They were really easy to cut and the convenient stripes helped a lot to make identical halves. Plus they add a little color to the setup.
With an open bezel like that, a HD in the cage would be clearly audible from the front. Instead, I opted to put an intake muffler made of BeQuiet! damping foam (and noise barrier mass) in there. It really works well, helps hiding the cables to the HD and front panel, and very little noise escapes the intake area:
The most restrictive part of the intake is probably the air filter. But for the low flow I use, it's perfectly adequate.
The drive cage being occupied by the muffler, the harddrive needed to go inside the case. I put it on two stripes of 2 cm thick foam on the case floor, so air can pass below the drive. As I have only one single drive, this is easy and works well. The computer isn't being moved around, so I have no need to attach the drive, but I guess that could be done without affecting acoustics, too.
The video card is fairly low power and considering that I'm not running 3D-intensive apps the air coming in from the front is more than sufficient to cool the GPU well with the stock (fanless) heatsink.
Towards the top and front, I have completely sealed the case using acoustic foam and duct tape. The foam inside the door presses on the seals and on the horizontal bar below the PSU to prevent air from coming in anywhere except through the front filter. The optical drive is sealed up mostly to prevent air from leaking in around its sides. Even before I taped it (and running higher speed fans) I didn't noice dust at the front of the drive, as it has a sealed tray.
My biggest trouble was the PSU. This is a RevA2 Seasonic, and while I never had the erratic spin-up-on-load-drop, the fan controller definately ramped up way too early. The second problem were slightly high CPU temperatures, which I attributed to insufficient elimination of the heat that the HSF transports towards the M/B, as the case fan in its standard location and the PSU fan are both towards the left (door side) of the case. For those reasons, I didn't want the PSU to be above the HSF. The exact opposite issue came up, when Bluefront wanted the PSU exactly above the HSF, and his psu re-location worked for me too: I pushed the PSU back into the case, by 12 cm to be precise. Luckily, there was just enough space for the optical drive not to block the bottom PSU intake. The PSU rests on the metal bar and the optical drive, and on the right (M/B side) of the case there is acoustic foam wedged in place. It's not going anywhere. And since the PSU now gets cooler air, I felt more comfortable bypassing the internal fan controller (which had the Nexus I swapped in still spinning close to max). I put the PSU fan on a 7V diet and it has shut up now.
The space behind the PSU allowed me to relocate the case exhaust fan to the top (and more importantly, to the right of the case, where I want it). The remaining gap towards the bottom of the PSU exhaust duct, to the left of the case fan (above the fan in the picture) was closed by a piece of sheetmetal cut from an old PSU cover.
The placement of that exhaust fan alone has helped a lot with temperatures (I wouldn't be able to recall the numbers now, though). I swapped the stock Zalman fan on the HSF for a Nexus 92mm (hardmounted), and found that the comments of a Zalman at 5V being equivalent to a Nexus at 12V (in noise and cooling) are quite accurate.
To go still further, I built a cardboard (beware of security issues if you do that) shroud around the CPU:
It ducts the output side of the HSF to the exhaust fan. It installes inside the vacant memory slot and between the sound and USB backpanel connectors.
Thus, the CPU only sees relatively cool air coming in, and none of its exhaust can make it into the PSU. I connected both the 92mm CPU fan and the 120mm exhaust fan above to the same Zalman Fanmate. Due to the somewhat higher flow rating of the 120mm fan, the shroud is always at a slight negative pressure with respect to the rest of the case. I opened up small openings in the shroud next to the northbridge (which got less cooling as the CPU fan is now very low speed and airflow is restricted by the shroud), and next to the voltage regulators, to help with cooling of these components. And finally, note the paper ring in the intake of the shroud. It assures that air can not bypass the heatsink, and has a big impact on the cooling performance.
With acoustic foam in place, the shroud became a bit heavy. It still stays in place on its own, but I wanted to be sure it wouldn't go anywhere. Since all was much too tidy for my liking I added a little more ghetto look with an appropriately shortened chopstick (see overview picture above).
Most of the hot air is exhausted in the back through the opening the PSU has left. The inside of that tunnel is clad with acoustic foam. Both fans being perpendicular to that opening and the higher speed fan (PSU) being further inside the case both contribute to rather low noise levels.
To attenuate that little bit of remaining turbulence noise (mostly from the PSU exhaust grill, I believe), I built a muffler of Styrene, clad with acoustic foam. Yes I know that styrene can be glued with solvent and the duct tape looks ghetto, but it's just a temporary solution until I get around to buying some acetone. I would also need to paint it black, but I doubt that's going to happen anytime soon.
The PSU has vent openings to the front and side. I covered the side with the acoustic foam that wedges the PSU in place, but I didn't want to take any more chances and left the front vents open. And since the top two 5.25" bays are unuseable anyway due to the position of the PSU, I decided to put a front exhaust (and muffler) in there. Its inside (and the lower bay cover which is not shown in the pictures) is clad with (you guessed it) acoustic foam. That foam between PSU and optical drive is also handy to hide the cables running across the drive down into the case.
This side of the PSU has really low flow, and I believe the small openings on the side of the door in front of the drive bays are sufficient for the door not to impede it. This airflow also doubles as "cooling" for the optical drive.
The back panel
The last thing left to do was to ensure no air would get into the case through the back panel. That would have been pretty difficult to do from the inside. But since the panel is set back into the case its easy to do fromt the outside. The (2cm thick) foam without barrier mass I put here is not glued in place but just wedged, so I can adapt my new custom back panel as needed arises:
The box now runs with the CPU and exhaust Nexus fans on the lowest Fanmate setting (5.16V). The PSU fan is at 7V. That's rather low (theoretical) airflow of around 15cfm for the CPU and 22cfm for the PSU, in total the equivalent of a single Nexus 120 mm fan at 12V (but quieter than a single fan). Actual airflow must be lower, as the above is based on the fans rating in free air (anyone have a clue how to estimate how much lower it really is?). I wrote above that ambient noise here is low at night. But doing these listening tests has been difficult. I have to wait till 3-4am to have long enough periods without cars moving in the neighborhood. I have to turn off my main rig, and the CRT of this machine (it buzzes). I discovered that the al cheapo speakers on that system give out a very faint 50Hz mains humm even when switched off and unplugged from the computer, so that needs to be unplugged, too. And then, when I'm ready, I'd hear the fridge in the kitchen switch on...
Idling, it is inaudible to me from outside the case. Seeks are clearly audible when listening next to the box, and are most pronounced at the intake area. I believe I can hear seeks also when sitting at my desk, with my ears a little over 1 m from the intake area of the case. But on at least one occasion I failed to hear seek noises from my desk despite a low ambient. Very tough to tell. As I have never been much disturbed by seek noise (I guess thats purely psychological, but I percieve it to be less useless and therefore more acceptable), I'm very satisfied with the HD noise level.
are audible from close to the case. Sitting at the desk I believe them to be inaudible. I turned off all noise sources I had access to except the monitor, started a delayed power down of the box, switched off the monitor and waited. There was some HD activity followed by a faint click (in the PSU?) on power off. But there was no difference in constant noise level that would resemble fans. Same thing on the following power on. I only performed this test once. I'm not sure the rear exhaust muffler is actually needed at the speeds I run my fans currently.
As far as the number of fans is concerned, it can be reduced. The CPU can be cooled "passively" in this setup, with just the exhaust running. But the latter would have to run way faster for adequate CPU cooling and noise would go up. I also inadvertently ran with only the CPU fan once, as the current fanmate setting is below the starting voltage of the 120mm Nexus (the 92mm starts fine). CPU temperatures were unaffected, but the system was not Prime stable (errors after 20 min). I suspect this to be due to the voltage regulators or the northbridge being too hot (note to self: get a few of those small video memory heatsinks). This might be something to explore further, though.
If I press my ears against the case, there seem to be vibrations transfered to the case. Even though the PSU and the case fan only touch the case through thin foam layers, they are wedged in place so tightly, that they are hardmounted for all practical purposes. So is the CPU fan. The fact that I failed to slide acoustic foam behind the M/B tray is noticeable when listening on both sides of the case (noise is slightly louder fromt the right than from the left). So this is not the very lowest noise that can be achieve with this setup. But given that it's all glued in place now and that I have a hard time finding out if I can hear anything at all from the system when sitting at my desk waiting for the lowest ambient noise level possible here, there is no way that I will do anything about it.
These temperatures are obtained while running the prime torture test ("mprime -t on Linux"). They are measured with an analog thermal sensor (except for the CPU temperature which is from the sensor in the socket). An example picture when measuring CPU intake temp is here:
- Case intake: 20.7 C
- HDD: 36.5 C
- GPU: 42.0 C
- CPU intake: 34.0 C
- CPU socket: 60 C
- CPU exhaust: 40.3 C
- PSU intake: 32.2 C
- PSU rear exhaust: 38.8 C
- PSU front exhaust: 37.8 C
EDIT: added temperature of the GPU
UPDATE: I was worried about what temperatures the optical drive would reach during DVD burning, as it is only cooled on part of its top and by some rather hot, slow moving air. I'm glad to say that this doesn't seem to be a problem. I just finished burning a DVD at 4speed for 15 min (the disc I had to burn wasn't longer but I think temp was stable anyway), and the drive temperature (measured on top, the bottom had a similar temperature to the touch) went up by only 4 C, from 33.7 C to 37.6 C. Note that currently the maximum rear PSU exhaust temp is at 35.8 C, because the machine is folding, which appearently does not load memory as much as prime does (the memory is exclusively cooled by the PSU intake).
It's virtually inaudible, folding 24/7 . Better than a sleeping ant . It's 24h-Prime-torture-test-stable and 600-point-Gromacs-WU-stable. To maximally undervolt the CPU I had to give up overclocking (lost 10% of folding performance on that machine , not to mention the downtime it took to get it silent and tested), but overall I'm very satisfied. Big thanks to everyone here at SPCR who helped me find a new obsession (I won't talk about my previous ones ...). And maybe now that this is done and finished, I can start obsessing about work for a change... My main rig will have to wait, and even then, I'll just copy what I did here as components are virtually identical - that is, unless I come across some interesting new ideas to test until then
EDIT: typos, added some links, updated description of the HD "mounting", linked to images hosted at ImageShack