Most of the drop in prices was due to 2 GHz Pentium 4s coming out, along with good competition from AMD in the form of very cost-effective and fast Athlon CPUs. AMD is probably still a good way to go, especially with Double Data Rate (DDR) SDRAM, which costs less and ends up with about the same memory bandwidth as Rambus. But note that the Intel/DDR solutions may not be as fast as Intel/Rambus (i.e., Intel 850 based motherboards are probably faster than 845). I'm amazed how well free markets equalized the cost/performance of Intel/Rambus vs. AMD/DDR solutions. The market equally well drove down their costs due to mutual competition. As usual, the consumer benefits.
One of my main concerns was for noise reduction. Both Athlon and Pentium 4 are problematic for noise in that they consume so much power that they need big heatsinks and sometimes big volume fans. Most of the aftermarket fans cater to overclocking gamers, and some ended up with insanely noisy 7000 RPM fans. The good news is that they also ended up with some big, efficient heatsinks and the fans can therefore be run slower and quieter while still attaining good cooling performance, especially when not overclocked.
Given my lack of interest in overclocking, the Intel D850MV desktop motherboard seemed a reasonable choice. (This 478 socketed Intel Pentium 4 motherboard does not permit overclocking, but 423 socket version may.) The version I got has integrated LAN and Audio, and it is generally high quality. The Intel Ethernet chipset is pretty well regarded, and the Audio chip is from Analog Devices. The board also features dual ATA/100 IDE controllers, 4-6 USB ports, AGP4, and other contemporary features. This is the first time I didn't get an ASUS motherboard, and I regret it somewhat in that the Intel does not have the ability to power itself up at a certain time of day.
I got a 1.8 GHz Pentium 4 to go with it since that was where the marginal cost started to ramp up. Also got 512 MB of Rambus (RIMM) ECC (9-bit, error correcting) memory, though annoyingly Windows 98SE seems to use up all available memory. Even with memory tweaking programs like CacheMan, Windows memory management seems very inefficient, using up whatever free space is available. For example, where it would seem that there should be twice the resources available, 512 MB seems to gain only marginally more usable resources than the 256 MB I had in the Pentium III. This is silly and a sign of poor engineering.
The good news is that the Pentium 4 1.8 with Rambus runs games about twice as fast as my old Pentium III 450, Geforce 2 machine. The new machine benefits from a ASUS V8200 Pure Geforce 3 Video card. The new performance is remarkably fast, if not quite the jump I hoped for. The Geforce3 Ti500 chips at 240 MHz core clock are faster than the 200 MHz core in my V8200, but not quite as fast as the 250 MHz Geforce 3 in the XBox. In terms of frame rates, the differences between these is not immense. The V8200 has 64 MB of DDR SDRAM vs 32 MB in my old V6800 Pure. Hopefully the current games can make use of extra texture memory and other features.
But the main story here is noise reduction, so on to that.
Case fans seemed to become more common as CPU clock rates crept up to 1 GigaHertz. The very nicely built all-aluminum Cooler Master ATC-200 case had 4 case fans: two in front, one in back below the power supply, and one on top. The included case fans were reasonably quiet Y.S. Tech FD1281253S-1N, rated by the manufacturer at 22.5 dBA and 28 CFM at 12 Volts. But even with these fairly quiet fans, 4 of them at full speed probably would have been noticeable. Surfing around, I found a dealer with quieter Papst fans and I ended up giving two a try to replace the rear and top fans. At the time I was planning to leave the front fans unused with the option of activating them later if needed. The Papst 8412NGL is rated at 12 dBA and 19 CFM. I had no way to confirm the measurement, but at full power they were extremely quiet. These are quieter than any other fans I've ever heard, though they don't flow as much air as the others either. Installed in the case, these two fans are barely audible and then only with an ear directly next to them.
The same dealer, NW Custom Computers was a reseller of SilentPC components from the Netherlands, which includes a hard drive enclosure called SilentDrive along with quiet power supplies. Power supply manufacturers such as Enermax have tried to market power supplies as quieter, but the Silent PC ones provide independent laboratory sound measurements in standardized rooms, which are probably a lot more credible than other manufacture claims. Some users on the Internet also commented the Enermax quiet power supplies weren't much quieter than standard ones. I decided to give the Silent PC power supply and Silent Drive a try.
Silent Drive is a plastic box with a dual foam lining and Aluminum chassis mounts and heat sinks. I could probably engineer something similar, but it would take more time and money, so I chose to buy the existing solution. The Silent Drive works well except that I could not get the back to close. Based on a temperature probe, the heat sinking seems to work well on my 5400 RPM drive. Jon at NW indicated Silent Drive was probably not safe to use on 7200 RPM drives due to heat concerns. Silent Drive does very significantly reduce hard drive noises.
The Silent PC 300 Watt power supply is quiet, but could perhaps be more so. A good feature is that it now includes connectors for AMD and Pentium 4 computers. Jon shipped very promptly and was helpful and friendly on the phone.
Web sites describing quiet PCs mentioned using automotive type heat and noise barriers to quiet PCs. I purchased a 4 x 6 foot roll of this felt and aluminum foil type material at Pep Boys for around $30, and also picked up a spray can of heavy duty contact cement, sometimes labeled upholstery glue, to attach it. Some other sites suggested using Dynamat, which was developed for the car industry to dampen out body panel noise. I happened to have a roll of Dynamat Super around for an audio project, so I used that too. Dynamat itself also markets kits for PCs, though the sheets are a bit small.
The Dynamat and felt decrease noise in different ways. Dynamat is highly viscous, like chewing gum, but thin and pretty rigid. The Super has adhesive on one side and Aluminum foil on the other side of the viscous material in order to perform "constrained layer damping." This is where the acoustical impedance of the aluminum foil and the surface the Dynamat Super is attached to is much greater than the sticky material in between. This impedance mismatch acts as a one way trap where sound gets trapped and dampened in the sticky middle of sandwich. The sound energy then gets converted into heat and dissipated. The amount of acoustic energy is small so the heat is miniscule, but that's how it works.
I decided to put Dynamat on the inside left and right chassis walls, inside the top cover and on the back of the motherboard mounting plate. I put it on the back of the motherboard plate to dampen the chip cooler fan vibrations at their nearest point to the chassis. There was some Dynamat left over which I put near the hard drive mounts. To apply the Dynamat I cleaned the surfaces with Methanol Alcohol, heated the chassis and Dynamat with a heat gun (but only after the alcohol was long gone), and applied the scissors-cut sheets gradually with a roller. Cleaning, heating and rolling make the Dynamat stick much better. Not using them, there's more chance of it detaching and shorting out the internal components.
Felt or fiberglass insulation also dampens sound, but does it by providing a high friction path for the air molecules excited by the sound, courtesy of the fine but dense fibers. The automotive thermal shielding material has an aluminum foil on one side to reflect radiated engine and exhaust heat, but since heat is not really an issue here I put the felt side towards the inside and glued the foil side down. Putting the non-metallic side of the felt outward (toward the internal PC components) should also discourage the Aluminum of the Dynamat or felt from shorting electrical components if they come unglued.
To apply the felt, I cleaned the surfaces and sprayed glue on both the felt and the surface. I needed masking tape and newspaper in some places to keep the glue where it belongs. Let the glue dry to a tack, meaning not wet, but still sticky, then press together firmly. Once it sticks it won't come off, so line up the pieces carefully. I put the felt on the left and right walls, bottom of the chassis and a few other places. In the case of the walls, the felt went over the Dynamat which is perhaps excessive but probably made things quieter. I left plenty of clearance for airflow where any chassis parts may have come near the felt, so there's less felt coverage than Dynamat on the side walls. Remember that the main mechanism for cooling is still airflow so it's important to keep it flowing around all components. Adding more felt would probably make it quieter, but may obstruct airflow too much and raise the heat inside.
Note that the felt material is also used to insulate heating ducts, so you may be able to find it at hardware or heating and cooling shops.
I tried the new Molex Socket 478 from NW Custom Computers. The Molex is a pretty elegant folded radial fin aluminum design originally sold for use with AMD CPUs I believe. While it was quieter than the stock fan, it was still pretty noisy. More importantly even at full fan speed the Molex did not cool as well as the stock Intel fan. When playing games and running CPU-intensive financial analysis programs, my CPU frequently hit the 59 degree Celsius alarm on the Intel Active Monitor program, resulting in CPU slowing, and eventual CPU halting. The stock heat sink may be noisy, but it never let that happen. I had seen other references that the Molex was for 1.7 GHz and slower Pentium 4s, and that may be a more appropriate use.
Searching further I found an Australian review of several 478 coolers and decided to give the GlacialTech Igloo 4300 a try. I bought one from from Phil Huddy, and the GlacialTech cools very well, better than the stock Intel and much better than the Molex. Peak CPU temps are now well below 50 Celsius under all conditions. It can even get down the 30s at times, which is remarkable. Phil mentions that he also carries other great Glacial coolers for socket A/370 (AMD, Celeron, etc.).
One problem with the GlacialTech 4300 which I have brought to Phil's attention, and he the U.S. distributor (and he hopefully the Taiwan manufacturer) is that it does not clear the white pushpins that lock the heatsink bracket to the D850MV motherboard. I got around this by replacing the pushpins with toothpicks shimmed with tape to a similar diameter. Note that Phil no longer seems to sell the Glacial Tech coolers, but they are available from the U.S. distributor ABS PC.
Presumably they have not run into this much because 478 Pentium 4s are usually sold boxed with the stock heat sink, because the 478 can't be overclocked so it doesn't appeal to the usual heatsink aftermarket, or because other heatsink board retention mechanisms are used more often. Or maybe people just don't use the white pushpins, at the risk of having the heatsink suddenly detach itself from the board and CPU. Another possibility is that this product is relatively new to the market and hasn't had all the bugs worked out. With the modification above, the 4300 works very well for me. A 4200 model with a slower and quieter fan is also available. If you're not going to modify the fan speed this may be a better choice for you if your Pentium 4 is 2.0 GHz or lower.
For the Geforce3 I tried running a ThermalTake Blue Orb screwed on top of the stock heat sink, but found the VGA chipset and RAM both ran cooler when I removed the stock heatsink and clipped the Blue Orb directly to the board. (But with the stock RAM heatsinks still in place!) The Blue Orb, like the GlacialTech ran significantly cooler temperatures under all conditions than the stock heatsink fan assembly. Be sure to use heatsink grease on all replacement heatsinks. For good measure I added grease to the RAM heatsinks also.
As a baseline, here are some temperatures reported by the onboard monitoring circuits for the stock ASUS V8200 cooler and Molex 478 cooler:
Molex 478, stock V8200, 12 Volts cpu/mobo gpu (video) system cpu chipset ram full 44 52 57 45 cpu cool 34 35 57 45 gpu cool 44 52 49 39 cpu & gpu cool 34 35 49 39The cooling modes are controlled by ASUS SmartDoctor program (and others). When the CPU and GPU are idle, they can be set in a truly idle state where they consume about 20 and 7 Watts less power each, a huge difference. Without this setting, by default they run in a loop that keeps them hot even when "idle". Perhaps it's to reduce thermal cycling. Temperatures are in Celsius and the ambient temperature was 21-23 Celsius.
Running at full speed on a solid 12 V, both the Blue Orb and GlacialTech were much noisier than the stock units, so the next step was to reduce their speed. I worked up a toggle switch panel to power these fans and also the front 2 case fans from either 5 or 12 Volts. I carefully checked temperatures at both voltages and also confirmed all fans would start at 5 V. Some 12 Volt fans won't start turning when run from 5 Volts. The toggle switch for the CPU and VGA fans is On-On so it will either be at 5 or 12 Volts, but never off. The case fan switch is center off (On-Off-On) so it can be switched to 5 or 12 Volts or off. At 5 Volts the front case fans are nearly inaudible.
Here are the temperatures with the GlacialTech Igloo 4300 Pentium 4 cooler and the ThermalTake Blue Orb screwed on top of the stock Geforce cooler, both running slow on 5 Volts:
GlacialTech 4300, Blue Orb ontop stock, 5 Volts low speed cpu/mobo gpu (video) system cpu chipset ram full 38 43 59 40 cpu cool 30 30 59 40 gpu cool 38 43 49 35 cpu & gpu cool 30 30 49 35As you can see the Glacial is much cooler than the Molex under the more difficult condition of a slower fan. Note how the GPU RAM is much cooler with the Blue Orb, even when slowing the fan, but the GPU itself is hotter. Please take the absolute temperatures with a grain of salt since the sensors are not calibrated. But the relative differences are probably large enough to be significant.
Removing the stock V8200 GPU heatsink and attaching the Blue Orb directly to the GPU lowered the temperatures of both the GPU and GPU RAM 2 to 5 degrees further, again with the fan slow:
Glacial, Blue Orb directly on GPU chip, 5 Volts low speed cpu/mobo gpu (video) system cpu chipset ram full 38 43 54 36 cpu cool 30 30 54 36 gpu cool 38 43 45 33 cpu & gpu cool 30 30 45 33
The new coolers are so effective that they can probably run at 5 Volts even under full demand. But for added safety and peace of mind I crank them up to 12 Volts when playing games. Usually the games are so noisy that they mask the computer noise. But it's nice to be able to turn them down for quiet work. In "Game Mode" at the 12 Volt setting the Glacial and Blue Orb run even cooler than at the 5 Volt settings above:
GlacialTech 4300, Blue Orb on GPU, 12v high speed fans full 33 39 47 34 cpu cool 28 30 47 34 gpu cool 33 39 39 31 cpu & gpu cool 28 30 39 31All of the above temperatures were with the front case fans off. The rear and top fans pull in lots of air even then. On a hot day the front case fans may help cool things down further by sending yet more air through the case. I noticed everything was much cooler when the case side panels were off for maintenance. But with the case buttoned up on a cool day the front case fans didn't seem to make much difference.
Another way to slow down the fans is through software like SpeedFan, which can vary fan speeds continuously in response to temperature changes. Programs like these work in conjunction with chips on some motherboards to vary fan speed by using Pulse Width Modulation (PWM) to change their effective DC voltage. A downside is they use some CPU and memory. SpeedFan is a slick free program that uses about a Megabyte of memory and could be coded more efficiently. Interestingly it doesn't use much more resources than the ASUS SmartDoctor or Intel Active Monitor programs that watch the Graphics and CPU temperature and fan speed, but can't change the fan speed.
Here are some rough percentages of resources used by these programs and MBProbe on my system, as reported by Resource Meter differences. MBProbe has a tiny footprint, but it only reads values. It apparently can't change the fan speed.
Intel Active Monitor SmartDoctor MBProbe SpeedFan system 1-4 1-4 0-3 4-6 user 1-4 1-4 0-3 4-6 gdi 10 2 0 8-9
Note that unlike the Arctic Cooling Pentium 4 cooler, their VGA cooler is not thermally-controlled, instead it has a high and low speed switch on the back panel. Thermal control would be slicker, but perhaps not sufficiently reliable.