Terry Gray
My quest for quiet personal computers started in the mid 1980s, when my totally silent Macintosh Plus surrendered to the lure of the shiny new --but horribly noisy-- Macintosh SE. Perhaps those early efforts to make the Mac SEs quieter by replacing their fans with quieter ones set the stage for my later conviction that personal desktop computers should not have fans, even if that meant they couldn't have "burning edge" processors. I'm not opposed to fast computers, but I am very opposed to turning my office and living room into noisy computer facilities.
My initial efforts to build silent, or at least fan-free, Intel-based PCs began during the summer of 2000, culminating in the successful deployment of the Pentium III computers described in the first chapter of this story. Both were housed in conventional tower cases and are still used today at home and in my office. Beginning in the Fall of 2001, I embarked on a new chapter in this adventure: determining whether it would be possible to create a fan-free (and thus --hopefully-- very quiet) computer that was also very small. Small is beautiful, but quiet is crucial. Together, small plus quiet, would be sublime!
Highly-integrated motherboards designed for compact systems don't seem to have much in the way of BIOS controls for underclocking or undervolting. That may be because the number of people in the known universe interested in reducing processor performance by underclocking can probably be counted on one hand, and while these are the same kind of voltage and frequency controls that overclockers would want, few overclockers would choose to buy a small highly-integrated mobo. Instead, they would tend to buy large motherboards with lots of PCI slots, for use in good-sized tower enclosures with lots of fans in them. Thus, there is no perceived demand for this functionality in small-form-factor motherboards, which are not targeted at the "enthusiast" market.
So... the underclocking/undervolting techniques I used to build a fan-free system 18 months ago would not work on the latest crop of super-small highly-integrated mobos. Enter the low power VIA C3 processor family, which promises to obviate the need for underclocking/undervolting in order to run without a fan. I had been reading about the VIA C3 since the 800MHz "EZRA" part was announced. Meanwhile, I had started a crusade in the office to get some quiet computers into our conference rooms. In due course I wondered whether the C3 might provide the basis for a suitable small/quiet conference room computer. The Transmeta chip used less power, but it was also slower, and more to the point, required a special motherboard. In contrast the C3 used the same motherboards as the Intel Pentium III and Celeron processors... more or less.
But first comes the C3 labelling/marketing disaster. A search of the usual online computer retailers revealed that sources for the C3 were rare in the USA, but TigerDirect.com carried them. I ordered some of the 800MHz processors, which were clearly identified on the website as .13 micron parts. After spending a lot of time testing the CPUs in various configurations, I was disappointed in the results I was getting. In due course, I learned that there were actually two different versions of the 800MHz part, and the ones I had ended up with were not the newer .13 micron parts that offered the lowest heat emission. Even better: you can't tell the difference by looking at the markings on the top of the chip; the only difference visible from the outside is the core voltage printed on the bottom of the chip. And no one seemed to know when the .13 parts would be available in the USA. Sigh...
Fortunately a VIA employee monitoring the Yahoo silent-pc egroup took pity on me (Thanks Tim!) and arranged for me to receive a .13 micron part, and he was also instrumental in getting me connected with other VIA folks to straighten out the TigerDirect problem.
What follows is a brief summary of several experiments I conducted with the VIA C3, with mixed results... none of which represent failures of the C3 idea in principle. While the C3 is not the speediest processor around, by today's standards, it is perfectly adequate for basic web surfing, personal productivity computing, and even basic streaming media tasks. The challenge is finding a suitable system enclosure; one that is oriented toward convection cooling. As far as I can tell, they don't exist.
Experiment #1: HP e-pc. When HP introduced the e-pc in 2000, I was impressed by the fact that the specs mentioned acoustics, still a rarity in computer system specifications. I ordered a 533MHz Celeron model and used it as my primary office computer for some time. It was certainly quieter than most computers, but it wasn't silent, and being on the desktop just a couple of feet from my ears, the noise emitted was quite noticeable. The e-pc has a single temperature controlled fan, and I thought that the fan might be superfluous if I plugged in a C3. Alas, the e-pc motherboard and/or BIOS is not compatible with the C3... at least it didn't work for me. (As it happens, this experiment led me to realize that most of the noise from the e-pc was actually coming from the hard drive, not the CPU fan. Replacing the drive made a significant improvement in noise level. Note that this was an early e-pc with a mere 533MHz CPU. Newer versions with faster processors may have higher fan noise levels. This original e-pc is now serving well as one of our conference room computers.)
Experiment #2: SaintSong MiniPC. There is a really tiny PC being advertised by several web retailers. It is marketed under several different names, but it appears to be based on the work of a Taiwanese concern named SaintSong. I couldn't do better than this for size, so I figured if a C3 would allow me to ditch whatever fan was in the unit, I'd be in fat/quiet city. In contrast to the HP e-pc, retrofitting a C3 into the miniPC worked fine, but the goal of a small silent system was to go unfulfilled. The MiniPC actually has two fans: one is a relatively quiet temperature controlled fan over the heatsink, and the other is 1" high-speed, high-noise case fan that is... a disaster. I tried lots of variations, but concluded that in the tight confines of this little machine, some forced air cooling really was necessary, and a 1" fan is a guaranteed ear-burner, since it has to spin fast in order to move any useful amount of air. Oh well... it would have been nice.
Experiment #3: a small (11" x 12" x 3") desktop case. I searched the web for hours looking for interesting small-form-factor cases. One that looked pretty good was actually bundled with a motherboard based on a VIA chipset, so I figured there would be no problem with C3 compatibility. And that was true, but I subsequently learned that motherboards can make a big difference in how hot a CPU runs. The bundled motherboard was made by FreeTech, and it was very similar in design to the Shuttle FV24. But using the C3 internal temp sensor, as reported thru the VIA chipset, the CPU ran substantially hotter in the Freetech board, as compared to the Shuttle board. The voltage monitors did not show any significant difference between the two. I never got a response from Freetech tech support when I inquired about why this might be.
The small size of the sealed enclosure was antithetical to convection cooling, even if the lower-temp Shuttle board could have been used. (It wouldn't fit without some case mods I didn't wish to pursue.) Moreover, the low height of the case severely limited the size of heatsink and/or heatsink-fan combo that could be fitted. I tried enlarging the side vent holes, and drilling new vent holes in the top of the case. Even with the .13 micron 800MHz C3, I was not comfortable with the temperatures I was seeing without some amount of fan cooling. I was able to use carefully selected fans that were undervolted to keep the system relatively quiet, but not quiet enough to declare this experiment a success.
Experiment #4: a slightly bigger case. I found a case that was slightly larger than the previous one, and tried the same approach of drilling holes to make it more amendable to convection cooling, but in the end I came to the same conclusion: a fan would still be needed to keep the system sufficiently cool.
Experiment #5: an even bigger case. Still small by normal standards, this was the first one of the series that was large enough to hold normal size CD-ROM and floppy drives. It was also the only one tall enough to allow a Zalman flower heatsink to be used. I did a lot of comparisons with different heatsinks during this project, and I concluded that for convection cooling, the large surface area of the Zalman was the winner. Even so, and even with holes drilled in the case directly above the heatsink, it ran hotter than I'd like until mounting an undevolted case fan between adhesive foam strips, oriented to move air from the CPU side of the unit toward the power supply side. Close, but no cigar.
Experiment #6: my "IKEA" case. Well, heck: it seems that nobody makes a computer case oriented toward convection cooling. But how hard can that be? Surely I could find a suitable box, intended for some other use, that might serve as a prototype. Indeed. I tried many options, and was taking a fancy to a Martha Stewart plastic cake pan discovered at a K-Mart, when a trip to IKEA revealed a small two-drawer wooden case that looked interesting. Sure enough, pull out one drawer, and you see a box that could hold a Shuttle FV-24 motherboard, a 3.5" hard drive, and have room left for a small convection-cooled power supply. Cutout suitable holes for the ATX backplate and power connector, and presto: a computer case.
OK, so I still need to find some kind of mesh top before I send it out for UL approval :), but it demonstrates that the goal of small and silent can be achieved. I also found an end-table-on-rollers at IKEA that could hold a complete system: monitor on top; printer and speakers on the bottom shelf, and my "IKEA drawer computer" sitting on the shelf in between. This left about 4" of clearance above the drawer. In this configuration, an external thermometer probe stuck into the Zalman heatsink reported idling temps around 35C, which was several degrees higher than when the drawer computer was sitting on the kitchen table with no obstructions above. I consider this experiment to be a success, although primarily as a confirmation that a manufacturer could build a convection-cooled system based on a VIA C3 if any of them were motivated to do so.
Experiment #7: Abit SE6. This isn't actually a C3 experiment, but in the course of building the various C3 systems, I became enamoured with the Zalman Flower heatsink, which seemed to do an exceptional job in convection cooled scenarios. I wondered if I could run my Pentium III 667MHz processor at full speed (though still undervolted), with an improved heatsink. I had used a very good Alpha heatsink previously. After running the system in this mode for awhile, I began to experience instability, so I went back to underclocking the CPU to 500MHz. It turns out that about the same time I replaced the Alpha heatsink with a Zalman, I also swapped one of the three PC133 128MB memory modules in that system for a 256MB stick. At the time it seemed like an innocuous thing to do, but in retrospect it may not have been, since the "8x" type of SDRAM in the larger module --I later discovered by accident-- was allegedly incompatible with the Intel 815 chipset. Whether this was the cause of the instability, or the lower core voltage was, I cannot say --although I would have expected the problems to show up sooner if the core voltage had been too low.
Experiment #8: Abit SE6. As a postscript to the experiments above, I recently added a TV tuner card to my trusty underclocked Pentium III system. After watching for awhile, I was seeing CPU usage at 100% and CPU temps in the mid-50C range, which seemed excessive to me. I later discovered that I was not using the latest Intel driver for the i815 integrated video, and fixing that dropped CPU usage to around 50% and the CPU temp down to the 46-51C range, depending on time of day and ambient temperature. However, I didn't learn that until later, so I embarked on what proved to be a multi-day misadventure of trying to use a VIA C3 in the Abit SE6 motherboard. It didn't work very well, but I'm not entirely sure whether there was an incompatibility between the mobo and chip, or whether trying the C3 triggered problems with a PC133 memory stick that was not entirely compatible with the i815 chipset on the Abit mobo. I'll spare the gory details, but think corrupted files on disk, complete reinstallation of the system on a spare Hitachi 2.5" hard drive just back from repair, followed by discovery that the "repaired" Hitachi drive was also faulty. Add problems with Windows XP recognizing external firewire drives, and you have a fun-filled weekend suitable for any masochist. Three days later, I had things back together, including the latest video driver. Perhaps one day I'll have time to complete this experiment by trying the TV tuner card in a C3 system...
Conclusions
This is one of a series of articles describing my silent computing adventures. This link will take you to the beginning of the story.
