I thought I would take a few minutes to chat about FAH, my hardware setup, and some benchmarks I recorded. To forewarn, this is mostly a lot of rambling detail and not very interesting.
First, I’ll let a small skeleton out of the closet: Mourn = me
Now that I have made a small impression in SF’s FAH scores I figured “why keep it concealed any longer?” The problem is that I have only one machine, my home PC. I have been able to let it crunch away, unattended, for long periods recently. If I start needing it for anything intensive, though, my FAH output will be reduced or halted. I am considering building a second machine once I get recalled to work. (If ever) Then I could dedicate one for crunching and one for everything else.
Hardware: I am running a Palomino-based Athlon XP 1900+ on an Epox 8K3A+ motherboard. The Athlon is multiplier-unlocked.
It is well known that Athlon XP’s use quite a bit of power and radiate a substantial amount of heat. To make it worse, the following variables all add even more heat: Running a program (FAH) that causes a 100% CPU load nonstop, running a CPU at higher than default speeds, adding additional voltage to a CPU to access those higher than default speeds.
Personally, I hate hot-running CPU’s. (Not enough to go back to Intel CPU’s, though) I also hate noisy PC’s. After the above combination started making more heat than my HSF could handle I starting thinking about a solution. I didn’t want to go to water-cooling or a peltier. Air-cooling is as complicated as I intend to commit to. The first heatsink that came to mind was the massive Swiftech MCX462. This awesome heatsink blows away the competition but requires bolting it through the motherboard, which is more of a pain in the ass than I wanted. Instead I used the smaller Swiftech MCXC370, which allows socket clipping and is easily removable.
Next, I started thinking about a fan. Since I like to assemble rheostat-controlled fan harnesses I figured I would use one that could be run very fast and be dialed down to a quieter level if desired. I chose a Delta FFB0612SHE which is 60 X 60 X 38mm, spins at 7000 rpm (in free air, about 7500 rpm when mounted to the heatsink) and makes 50.5 decibels of noise running at full speed. The fan’s specs say that it draws .55 amps, but that is when it is when it is already running, not initially. I took my DMM, put it in series with the fan, and set it to record the maximum current draw. I then powered up the fan 10 times and earned a highest recorded starting current of 1.44 amps! That translates to 17.28 watts at 12 volts. Safe to say I wouldn’t be powering this beast from a motherboard header. I am to understand most motherboard fan headers are rated for about 3 watts or so.
The next step was to pick out a good rheostat and decide what values would be best for it. From tinkering with DC fans, I have found that the more powerful ones of a given size need less resistance induced to slow them down a given amount of RPM as opposed to slower fans. Experimenting with some potentiometers I already had showed that it only took 8.3 ohms of resistance to slow this fan from 7000 rpm down to 5500 rpm. With that in consideration, I obtained a 50 ohm, wirewound Ohmite rheostat rated for 25 watts. This is bigger than necessary, but I had the room to mount it and would rather overbuild than underbuild such a setup.
The fan wiring was routed like this: The RPM sense lead was terminated with a motherboard fan header connector and plugged into the motherboard’s CPU fan header. (This motherboard will not power up if it doesn’t detect fan RPM on the header) The ground lead was connected to an extra ground wire from the PSU by means of a quick-disconnect terminal. The 12-volt positive lead was soldered to the rheostat and another wire went from the rheostat’s other leg to a 12-volt positive lead from the PSU by means of another quick-disconnect terminal.
I obtained a nice 1 ¾” wide aluminum knob that I put on the rheostat shaft which sticks though the back of the case just below the exhaust fan.
The resulting setup has worked pretty well. If I want to overclock fairly high and run FAH, I can crank the fan up to maximum and leave the room. If I want to use the PC and not suffer deafness or insanity from the fan noise I can set the CPU speed and voltage down closer to default in the bios and dial the fan down to 4500 rpm or so, which is nice and quiet. The bios for this motherboard has the additional ability to power down the PC if the CPU temperature reaches a preset level. I have that parameter set at 60 degrees Celsius, which is the lowest setting. This is a nice safeguard against roasting a CPU if the cooling fan decides to die.
Benchmarks and Overclocking: A Palomino-based Athlon XP 1900+ runs at 12(multiplier) X 133(front side bus) for 1600Mhz @ 1.75 volts at its default setting.
In trying to see how fast I could get mine up to without suffering instability or overheating I experimented for several days using different parameters while monitoring the heat output with Motherboard Probe the whole time. I discovered that the ambient temperature was my limiting factor. During the week I was experimenting we had some 80 and even 90 degree (F) days. During that time I could only crank the CPU up so far without nearing the 60 degree Celsius redline that I had imposed on it.
After the weather cooled down a bit I had better luck. The highest I managed to push the CPU and keep it stable and cool was 10.5 (multiplier) X 172.84 (front side bus) for 1814.82Mhz @ 1.95 volts. At this speed, the fastest I was able to complete a single frame of ProteinA was in 6 minutes, 54 seconds. I had FAH running in a 20Mb ramdrive for this test. This speed and voltage was a bit too much for continuous use, I decided. I ended up using a slightly more conservative 10.5 X 167.73 for 1761.13 Mhz @ 1.85 volts for regular nonstop FAH use. At this speed the fastest ProteinA frames I have recorded were done in 7 minutes, 6 seconds. This was also done running in a ramdrive.
Running FAH from my hard drive instead of a ramdrive results in only a slight increase in frame completion time. I actually do run FAH from my hard drive normally. With a ramdrive you will lose all your work if the PC loses power. Running FAH from the hard drive lets you start right back at the frame you stopped at if the power is disrupted, which I have had happen.
Pictures I apologize for the horrible quality of these pictures. I don’t own a digital camera or even a decent film camera. These were all taken with a ten-year-old Canon Z-115 pocket camera. The images were obtained from the shoddy CD that was included with the developing.
[Mod: Picture links removed at the request of Groltz, who wants Handruin to be able to use his server space for other things. Pictures detailed the inside of Groltz's PC, a very nice overclocked machine with a clean wiring job. -- Mercutio]
Anyway, for those of you still awake, I hope to keep contributing to SF’s FAH team score. I think we have the capability of reaching the top 35 rank, eventually, at our current weekly output rate.
---Steve