This article demonstraits the potential of a silent, no moving parts workstation. Several of the topics discussed here are extensions of those discussed in a previous article avaliable here.

Hardware:
Hardware was selected for the demonstration system based on several factors. Every device chosen is capable of running indefinitely in a ambient (~24c) environment with no forced air cooling. This is important because without any cooling fans, the only air movement in the system will be due to convection. Devices were selected with solid Windows and Linux support. Also, all chosen components are reasonably priced. For example, Bitmicro is producing solid state IDE flash drives in the tens of gigs capacity; unfortunately, their multi-thousand dollar cost makes them outrageous for most applications.

Processor:
A VIA C3 933a processor was again used for this demonstration. The C3 is a socket 370 processor designed for low power and low heat applications. A more in-depth description of the processor is available in this article or directly from VIA's website. The C3 was selected for this application because, due to it's low heat output, it can be run with a simple passive cooler such as the Coolermaster DP5GN01. Also, the C3's minimal power draw allows the use of a lower output power supply.

Power supply:
An ATX power supply capable of running without a fan is needed to power the system. Although there are now several manufacturers producing PSU units with low speed, quiet, and/or temperature controlled fans, fanless desktop units are extremely rare. Fortunately, 1U models designed for rack mount systems exist which are rated for no-air-flow operation. These units are able to operate without forced air cooling by strapping their heat producing components to their heavy outer frame. The powersupply is then meant to be mounted in thermal contact to case, thereby providing a huge passive cooler. For this article a TK300PO-60 TKPower ATX powersupply was chosen. This is a 1U open frame powersupply rated to 300w, or 150W with no air movement. This PSU conforms to the electrical ATX specification, however is physically the wrong shape to mount in a conventional desktop case. It should be noted that the open frame design could potentially be dangerous if some care is not exercised.

Mass Storage:
Probably the largest impediment to a silent, no-moving-parts design is data storage. Conventional mediums such a hard drives, CD-ROMs, floppy drives, tapes, and Jaz or ZIP drives all use some form of spinning platter. Compounding this, a 7200rpm harddrive can be the loudest component in a system, so eliminating it is crucial. Noise in this area can generally be improved by using a slower spinning 5400rpm drive, or even a 4200rpm laptop drive. Unfortunately, as the platter spins slower, the latency is increased, slowing the whole system down.

CompactFlash, made popular by it's role in digital photography, provides an IDE compatible interface. Using a simple passive adapter (passthrough wiring with no logic) a CompactFlash card can be used as a IDE device. The adapter used for this review is based on a schematic provided by Sandisk. Unfortunately, CompactFlash only provides a PIO (programmed-IO, ie - not DMA) interface. The lack of DMA support makes access very CPU intensive (apparent in the HDTach scores below). This also appears to have a second, less obvious consequence. CompactFlash on a IDE interface is recognized by both Promise and Highpoint IDE raid controllers, but is not considered a valid device. Unfortunately, this prevents the formation of a simple Windows RAID, as shown in the picture below. (Note: We are still looking to confirm the cause of this; at the time of writing both raid controllers simply would not not recognize CompactFlash as a RAID-worthy device). CompactFlash can also be used as a removeable media to replace the traditional floppy drive using a USB adapter similar to the Sandisk SDDR-31. This adapter has solid driver support for Windows and Linux; however, it lacks the boot support of a conventional floppy.

The relative speeds of CompactFlash (CF) over both USB and IDE was compared to that of three more conventional hard drives. A Quantum 20GB AT Fireball 7200rpm drive, a 9GB older Quantum 5400rpm drive and a 20GB IBM Travelstar 40GN 4200rpm laptop drive were compared using Sisoft Sandra filesystem mark for general performance and HDTach for more detailed results. All tests were conducted under Windows 2000 with most efficient transfer mode (DMA if possible) selected. For Sisoft Sandra, a NTFS file system was used. Sisoft Sandra conducts it's tests within a file, making it susceptible to effecencies of the files system. HDTach makes more low-level calls to directly access the drive, thereby bypassing the filesystem..

Due to it's solid state construction CompactFlash never has to wait for a platter to rotate or heads to move, and consequently has much faster access times than a conventional hard drive. Unfortunately, it can only sustain about 15% of the data throughput of a traditional drive. The limitations of the USB interface are apparent as the IDE connections can sustain almost twice the data throughput with a dramatically faster access time. However, due to the reliance on a polled (PIO) interface, the IDE adapter consumes more than ten times more processor time than any of the other listed interfaces. It is interesting to examine the fluctuations in access speed on a conventional hard drive, versus the constant rate of a solidstate device as is shown in the following HDTach Results.

When using CompactFlash it is important to understand the degree of variance between different implementations. The CompactFlash specification is very open, leading to dramatic differences in performance between manufacturers. Also, many 'brands' are simply relabelled cards produced by one of the major manufacturers. For example, the Nikon piece shown below shows up through the IDE adapter as a Sandisk product. This relabeling is even more prominent in 'OEM' compactflash. The two 256MB pieces shown below were purchased at the same time from a supplier. Through the IDE adapter they post differently, use a different transfer level and perform very differently when benchmarked.

The 37.4ms access time of the 64MB Sandisk card was tested repeatedly and confirmed on a second card. The degree of variation here, especially between two reputably equal OEM cards, makes it hard to predict how a piece will perform before it is purchased. This is likely to make reputation and brand loyalty very important to consumers.

Other Hardware:
The system was tested using a ECS P6VXM2 micro-ATX motherboard which uses a VIA Apollo Pro 133A northbridge and a VIA 686B southbridge. The system was equiped with two 256MB Crucial PC133 cas2 ECC SDRAM sticks. Video was provided by a Radeon VE. By virtue of being based on the classic mobile Radeon core, this card provides multimonitor support and DVI output while producing very little heat. Ethernet was supplied by a 3Com 3C905TX 10/100 ethernet card.

Running the Demonstration System:
Originally, the test system was intended to run on a RAID-0 (striped) 2x256MB CompactFlash drive, effectively doubling the throughput of a single card, and giving Windows (or Linux) more room to stretch its legs. Unfortunately, as the IDE interfaced CompactFlash was not recognized by any avaliable IDE RAID controller, the system was limited to a 256MB system drive. Windows 98 was therefore selected as the test operatings system due to its minimal install size. The windows installer recognized, partitioned, formated and installed onto the CompactFlash drive without issue. Booting windows, and similar drive intensive operations were noticeably slower than on the equivilant hard drive based systems; however, once booted the 512MB of system RAM minimized disk access to the point where the apparent difference in performance was noticeable but acceptable. The USB CompactFlash interface functioned perfectly once the supplied drivers were installed (later versions of Windows and Linux have functional drivers built in). The only disadvantage of a USB CF card compared to a conventional floppy is the need to manually mount, and unmount the drive when physically removing a disk.

Thermal Conditions
The lack of forced air circulated was apparent in the fairly high system temperatures. In order to allow for the option of better convection cooling the system was run both with case open and case closed. All tests and measurements were taking with ambient air temperature about 23c. The VIA C3 idled in windows in the high 30s in an open case, and in the low 40s with the case closed. Under full load the C3 pushed into the mid 40s idle and into the high 60s under load. While these temperatures were potentially higher than ideal, the system appeared completely stable during long term Prime 95 stability testing. The TKPower PSU was stable, although very hot (temperatures varied depending on where they were taken) when run stand-alone. When in thermal contact with the case it ran much more confortably. Given that it uses the outer case as it's heatsink, its temperature was much less affected by the open or closed case operation.

Conclusion:
The test system was designed to demonstrate the potential of running a personal computer with no moving parts. Unfortunately, given technological and fiscal constraints, the resulting machine was only really practical as a proof of concept due mainly to it's small hardrive. The only real detriment to a silent/solid state system at this time is the difficulty and cost associated with mass storage. Assuming a larger drive could be found, the demonstrated system would be acceptable as a basic workstation. Despite some technical and financial hurdles, solid state desktop computing is possible, and will likely become fully practical in the near future.

Impressions:

Jon: jon@minervatech.net
It is important to keep in mind the narrow focus of this article. The attempt was simply to prove whether it is feasible or not to run a computer with absolutely no moving parts. Running with only CompactFlash for storage is restrictive at best. In my opinion, RAID would have eliminated most of the disadvantages of CompactFlash, had it worked. Two of the three biggest limitations of CompactFlash are throughput and storage size (the third is cost), which is exactly what RAID is meant to enhance. I would be very happy using the silent computer as my "work" machine (web design, web browsing, word processing, etc.). The fact that it makes no noise at is something you must experience for yourself.

Unfortunately, the cost of CompactFlash (close to $1 CAN per MB) is annoying and hard to justify. I find myself wanting to just use the 20GB laptop drive sitting around and put up with the noise. Also, the range of speeds and "luck-of-the-draw" buying style (although Lexar seemed a good bet) keeps me from feeling good about the technology at this point. The good news is that the CompactFlash specification prevents it from getting slower, but does not restrict going faster. This means progress can only go one way.

Dave: dave@minervatech.net
If it were not for the cost associated with CompactFlash I'd be running one of these systems right now. Fortunately, with a one gig card from Sandisk down to as little as $1000 CAN and falling steadily, I think it will be feasible in 6 months to a year. I am pleased at how simply the system went together. The C3 worked fine with the simple cooler, the CF-IDE drive worked the first time, same with the PSU... I think this level of 'plug and play' operation shows strong promise for the future.

It was tough writing the conclusion for this article. The system worked just fine. Noteably, it made no noise and nothing moved. However, the feeling of turning on a computer and thinking "DAMN! The spring in that switch is noisy..." because that small click is ALL that you hear is hard to capture in pictures, benchmarks or text.

**UPDATE**: Regarding virtual memory, it was disabled for all tests. Windows likes to "fiddle" with its virtual memory file enough that it would in fact destroy a CompactFlash card in short order. The other reason virtual memory was disabled was there was no room left on the CF card! When the RAID didn't work, we were left with a single 256MB C: drive to fit Windows 98SE in, and there was little room to work with. It's funny to think that the computer has double the main system RAM (512MB) than it does storage space :-)

Feedback:
Please feel welcome to discuss, comment on or discuss this article using the posting system below. Please we advised that this is a MODERATED board and that posts deemed 'improper' will be removed without warning.