Why should someone try to build a PC without moving parts? Well, first of all, for the rest of this page, I will talk about a noiseless or completely silent PC, which is, in fact, not true, however, I got close enough. So back to the why.
I bought my last PC in December 2002. At that time, the dual processor Athlon MP boards became broadly availabe at a reasonable price with the introduction of the Tyan Tiger MP and afterwards with the introduction of even more MPX chipsets. I really liked that computer. I didn't make much use of the two processors, though and the biggest downside of this assembly was its incredible noise.
Soon after I bought it, I went through a lot of trouble (and money) to reduce the noise level. HDD enclosings, PSUs and processorcoolers claiming to be silent like a whisper and the like. This stuff really costs money and really is ineffective. Maybe I should go more often to discos and clubs to ruin my ears, but I don't and therefore the noise for concentrated working was almost unbearable.
After some time (just the middle of this year), things got worse. A new sound popped up and this was really annoying. The harddisk started to viabrate in a way, the whole case was sort of an amplifier for that, although the harddisk was in a no vibration frame. Basically that was it. I was tired of spending money for things that wouldn't work, because a noise reduction is no noise cancellation. The level of whispering in some companies is really cool. I call that crying. However, my decsision was to build a new computer, being incapable of annoying me with noise.
The decision to go for it was made externally. I was so annoyed, that I talked to a lot of people, annoying them with my babbeling about this new and ultra silent PC. Finally they told me to build it, show it to them and otherwise shut up. After so much talking about how I couldn't take the noise any longer, I somehow couldn't retreat and so the decision to build it was finally made.
I knew, the PC should be completely without moving parts, because every fan, efery harddisk makes noises that can become unbearable. Furthermore, if you actually are sensitive to noise, it doesn't matter, if you eliminate the source of the loudest noise, the second loudest noise will then take its place in annoying you. Most critical parts where of course the hard disk and the processor cooler.
For the PSU this was easy. I already had taken some glimpses at the page of Silentmaxx. Well, here I have to admit, this was rather stupid. Some of the parts for my old PC were from this page, for example some Verax coolers, and they were crap. Crap here is, very shortly after being outside the warranty they got really noisy. Stupidity comes to bad luck and while trying to oil the bearing, one of the blades borke...
In addition to that, their webpage is not viewable with my konqueror (which is configured to ignore Java and Java Script on regular pages). In principle this should be seen as a sign, that people there don't want me as a customer and the Verax issue should have been an additional warning, however, their PSU looked most promising.
For some time now, there are fanless PSUs out there. Someone pointed out, that fanless PSUs are probably the silent computing item getting the best manufacturer support in the least time. However, many fanless PSUs are not made for operation in a fanless PC. They require a case fan for proper operation. Only few have some sort of heat spreading material on the outside. One example is the Thermal Take one with a heat pipe going to the outside of the casing. This is of course a violation of some specifications and therefore, many manufacturers don't do that. The silentmaxx pro silence does.
Their cooling fins however are show only without a significant real function. Aluminium fins on the outside bolted to an aluminium heat transfer plate on the inside of the PSU. As most of you know, aluminium is not very good at conducting heat. The heat transfer point seems to be not very well done. Therefore, the assembly on the outside stays almost at room temperature during passive operation. The inside doesn't. The silentmaxx people put a 80 mm Pabst fan of the whisper type (stupid, stupid me) in it and said, this is a rescue fan. Should the temperature within the PSU ever go above 65 degrees celsius, than this fan would prevent the psu from overheating. Well, after 40 minutes of operation, the magic 65 degrees are reached and the fan makes an infernal sound that lasts for 20 minutes. Now the cooling fins spin into action, as the warm air is blown through them. Cunning design. Don't buy that PSU, if you want a silent PC! (Remark for those who know the German law, it was more than two weeks before I had all parts together and realized the stupid, stupid mistake, I made).
After some careful examinations and reading some reviews, I found my real PSU being from amacrox, the Calmer 350W. This design is missing the wannabe cooling fins on the backside, but five of the six sides of the PSU are not covered with solid metal but with a fine grid, allowing air flow in and out of the PSU. When I put this thing out of its box, I thought myself to be stupid, as the only side of the PSU not covered with a grid but a solid metal plate seemed to be the top (kids, you remeber things about warm air going up?). Well, I was not to stupid and also, the manufacturer of the casing was not making mistakes either. The simple fact is, the only non gridded side is the top. However, the PC has been running through for several hours without any problem, so for now I'm satisfied.
Not much choice here. Pentium M or an Athlon64/Sempron. I admit, I've completely lost count with the Semprons and no longer have any idea, which one is which. The big downside here is, that Semprons tend to achieve calculation power rather through high frequencies than big caches, something that doesn't go to well along in a completely passive setup. Therefore it was A64 or Pentium M. No hard decision here. As a regular reader of different media, I know what happened with the chip factory at Frankfurt/Oder (under Intels influence) and at the same time happens in Dresden with AMD. Some behaviour shouldn't go unnoticed. Technically both processors should behave almost identical, while the A64 is officially made for desktops, the Pentium M is not.
So, with the A64 things get complicated. There are the low power processors A64 mobile and the ultra low power processors called Turion. My first idea was to go for one of them, however, they do not have the built on heat spreader (they look more like the Athlon XP) and therefore, stock coolers do not fit. There are some suggestions on the web as how to modify the retention frames to fit on those processors. I didn't want to do that and therefore went for a normal A64. The choice then is to try to find a Winchester core A64 or a Venice. Winchester is much better in undervolting, that is, there are reports out there from Winchesters running at their full designated speed at .8 to .9 V, which is ~ .6V below their original power rating. As power consumption scales to the square of the voltage, you can reduce the power consumption up to 70% only by undervolting! The Venice core however, is somehow limited in its undervolting capabilites. I have found no reports about a Venice that could be driven below 1.1V, not because of stability issues, but because the processor would not allow a lower voltage setting. The best you can get out of a Venice is therefore a 45% power reduction by voltage.
Things seem to be easy here, Winchester and Venice have the same power rating and reviews indicate, they actually do consume similar amounts of power and then Winchesters can be undervolted much further. Unfortunately, Winchesters are no longer produced, so you have to find a dealer who is still stocking them (which is a hard task). A friend (hi Chris, thanks for your valuable input!) got his Winchester from ebay, however, I really dislike ebay. I prefer a shop, where I can go in build myself up to my full size, slash my fist on the table and say with a mean German accent "I want my money back". Didn't help me with my PSU, though.
So, there I was with my Venice 3200+. Side remark, it is almost completely irrelevant, which Athlon64 you choose, as long as its a socket 939. With Linux and Powernow! the thing will almost always be at 1.1V / 1000 MHz. Keep away from the Socket 754 A64, however, they are older core revisions and are specified at 1.3V / 800 MHz and are consuming more power due to their 130nm build process. The thermal design power for all socket 939 is 89W at full burn and 22W at 1000 MHz / 1.1V. In fact, there are strong indications in tests throughout the web, that these numbers are greatly exaggerated. It seems around 10W is more typical for the low power mode and 40-60W more likely for full burn.
It was really surprising to see, there is quite a number of passive processor coolers around there, today. Most of them use some form of heat pipe concept to transfer the heat to a massive number of aluminium blades, that dissipate the heat. However, you should look closely, what you are buying. In that domain, more is not always better. A passive cooler has to move the coolant (air for that matter) purely by convection. Air as any other coolant is resistant to move and therefore there should be as few obstacles as possible in the air path. That is, the heat dissipation blades must not stand together to closely! This will reduce the convection air flow significantly and therefore reduce cooling performance. Instead look for a cpu cooler with space between the blades. I chose the Scythe Ninja. From reports, I would have prefered the scythe NCO 2005, but I couldn't find a distributor stocking it.
Well, obviously, a normal hard disk is not silent and another solution has to be found. Solid state disks like the hyperdrive would be fine, but are unreasonable expensive for my purpose. Gigabyte seems to prepare a very cheap SSD called I-Ram, however, this is not yet available. SO I went for a 4GB flash disk. Flash disks today are reasonable fast (up to 20MB/s) and endurable. However, in contrast to magnet based hard disks, they have the disadvantage of suffering from writing to them. They can only take around 100.000 to 1.000.000 write/delete cycles (no name products are rated more like 10.000 write/delete cycles). So, using a flash card as a hard disk replacement is possible but you should be aware that reducing writing to the disk is good. (more to that in the OS-section). Big don't here: Never put a swap onto a flash or say goodbye real soon! I necessary, make a ramdrive and put the swap onto the ramdrive.
Modern flash tries to balance the write cycles, that is cells with the least usage are prefered when some write access to the disk happens. For that reason, it is wise to have as much space free on the device as possible. Let me explain... Assume you have a 4GB flash disk and 3.9 GB are filled with the OS and static, mainly nonchanging data. Assume now, you create a 50 MB file and change a lot and safe that a lot. Obviously only 100MB were free initially, so, all those write cycles are distributed among those 100MB worth of free cells. That is, if the cells endure 10.000 write cycles, you can save your file 20.000 times, before the cells wear out. Assume, you have 200MB free on the device, the file can be stored in four different locations, so you have 40.000 times, before your cells give up. Last assumption, your disk is only filled with 2GB, that gives you 40 positions for your 50MB file and therefore 400.000 write times. Still with me? Much free space is good.
A normal install of Linux takes up 2GB of space, I chose a 4GB flash disk from Kingston with guaranteed 100.000 cycles. For that disk not to become to full (see last paragraph) I put my home directory on a 1GB USB memory stick. Basically, that has the same limitatons (as its flash, too) but I have the advantage of being able to carry my home around to other locations. All my mail, my desktop settings in one place, whether I work at home, with my Laptop or at work. Its marvellous. For that matter, I made sure, my home is also around half the size of the stick.
A normal case won't do. Well, not in a passive setup. All normal cases need some form of active cooling to prevent overheating of the air inside. A solution is to provide the case with lots of hole to allow the air to flow through. In my case, I chose the Mercury Aero One. Unfortunately I have not found a link to the manufacturer, only links to pages of sellers. In principle, both sides and the top is not made of solid metal but a grid, through which air can flow out. After some time of operation, this can be felt when holding a hand over the case. The case is not very stable, that is, my cat has left the top being bent in from lying on it, however, its stable enough not to collapse.
The rest of the PC is described fast, an Asus A8V deluxe, as the VIA Chipset does not dissipate as much heat as the Nvidia and does not require active cooling. Much ram (currently 2GB, soon probably 4) to allow buffering of hard disk data and an Sapphire Ati 9600 passive cooled video card. I've put in a DVD RW drive (thus introducing some moving parts), however as the drive only spins up with an inserted disk, I don't consider that a problem.
Phew. Works. Quiet. Cool. Not enough? All right. I hate Windows, which is based on my feeling that its not natural, resource wasting and so on. Won't go into that. I've only installed Linux on the PC. With my notebook I noticed, with Linux and Windows idle, Windows uses much more processor cycles and heats up the processor much more. With Linux on that notebook, the fan is barely springing into life (mainly when a web page uses flash or so) while with Windows it only needs around 5 minutes of idle runtime to get the fan to the second highest setting.
For some time now, I've been going with Debian, I don't know the reason
to switch to that initially, I think it was something about RedHat not
storing a file, where I searched for it (yeah, I can be real nasty with
that one). Debian Sarge currently provides only Kernel 2.6.8 which does
not include dynamic CPU scaling on its own. The solution here is to use
a userspace tool. I used cpudyn. Unfortunately, there is some work to do
by hand to get it running. After installing it, you have to modify the
/etc/init.d/cpudyn script to contain the line
"echo userspace > /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor"
and you have to modify /etc/modules to contain both lines
powernow_k8
cpufreq_userspace
after that, dynamic scaling should work. Try a manual "modprobe
powernow_k8" beforehand. If it doesn't work, there are two possibilities,
one is Powernow is disabled in the bios two, Powernow is not suppoerted
by the bios for Venice cores. That was the case for me, so I had to
flash a new bios revision (the latest stable one from Asus did it for me).
My home is, as noted above on a memory stick. Keep in mind, doing such things makes you vulnerable against losses and theft. Backup agains losses and encrypt against thefts. I worked with cryptsetup which is a more elegant solution than the cryptoloop. I worked along the lines of this tutorial to create my crypto partition. The crypto partition is automatically mounted after login, so the .bash_profile on the main flashdisk mounts it onto the same directory as it is working on. The .bash_logout unmounts and removes the mapper. For that, you hav to first cd out of your home and "sync", otherwise, you will most likely not unmount but only get a device or resource busy report.
Then we have the issue of saving write access. I really don't want my hard disk to die sooner than necessary. In principle, apart from updates, there should be no writing at all to the main file system, except for logging. Logging is really nasty yet still necessary, should errors occur. I decided to go for an unconventional solution. On every bootup a ramdisk is created and all the data from the original /var/log is copied there and the ramdisk is mounted to /var/log. On shutdown, that data is synced back to the original /var/log. With that, loggings are not availabe should the PC ever crash. It hasn't, sofar and if it should on a regular base, I think, I would try to find another way. I tried laptop utils, by the way, however, I'm not sure, they work, because of some abscure error message I think indicating that my flash disk does not support spin down or something like that, therefore I didn't take a closer look to that.
In unrelated news... Ok, after the PC didn't make sounds (actually, it hums a bit) other sounds startetd to annoy me. The huming of the PSU for my lamp for example, the amplifier for my speakers and so on. Gee, this world is not perfekt. My cat turned out to be rather dangerous. He seems to like the warm air out of the grid and therefore declared the casing his new personal sleeping area which efficiently prevents warm air from flowing out. I have not yet done testing in a lying cat scenario. Without the cat, the PC ran through for around 5 hours without problems and after around 30 Minutes of 100% CPU usage the processor temperature (measured after reboot in the bios) was around 60 degrees. 10 below the Limit specified by AMD.
So while I enjoy the almost completely missing noises, you are hereby asked to ask any questions, you might still have to me. Thanks go to the unnumbered websites where I got my inspirations and especially Chris, with whom I had a very vivid and fruitful discussion by mail.
back