Logo for "Nota Bene / non-blog" goes here. NB main page
Along with navigation links.
3 June 2007
I was setting up new computer for myself, which I called Cassini because I was working on the project when the Cassini spacecraft was arriving at Saturn. Previously Iíve had trouble with things overheating (or problems I blamed on that), and being too loud with all the fans going.
I decided that Cassini was going to have proper attention to this designed in from the beginning, and would run quietly as well as never getting too hot.
I looked into heat sinks for the CPU, power supplies, and cases. The first is the subject of this memoir. I searched online for everything I could find on CPU heat sinks, and what little real information was available was all bad: the best-performing traditional heat sinks are still not good enough, and those that do have good enough cooling capacity are non-standard, being too massive (like a kilogram!) to mount normally. This taught me that the fan on a heat sink has outlived its proper domain of use. Just as plain heat sinks gave way to fan-topped heat sinks when they stopped being useful enough, something needs to replace this now.
I bought an A-BIT IC7-G motherboard with an Intel ďPrescottĒ P4 processor with hyperthreading. In order to get the 3-year warranty on the CPU, I had to buy Intelís cooler with it. So I started by just installing that to see how well it performed.
First of all, it is not quiet. My reading on various coolers rated the Intel stock fan as being near the loudest tested, and certainly not the best performing. Iím pretty sure I used premium thermal compound rather than the pad it came with.
I discovered that under sustained CPU activity (e.g. ray tracing) the temperature increased steadily for 30 seconds and just kissed the red line, at which point it leveled off and stayed there. I thought that was very lucky that it wasnít a little higher! But that made me suspicious. I took the access panel off the case, and put a large room fan up against it, on high. Benchmarking showed that the computer, when running for several minutes, was faster than with the case closed. The system was automatically throttling the CPU when it got hot, without complaining about it. The CPU would indeed not overheat, but the fan was not adequate to run at full load for more than half a minute.
The room temperature is over 80 degree F in the summer months, and can get a few degrees warmer as I turn down the air conditioner when not home. So I was pushing the limits of what it wanted for the incoming air (85 degrees) in order to cool properly, as well as it did work.
As noted 27 July 2004
Posted at The Pub
I reported earlier that my new setup was about 4 times faster at MPEG encoding. Well, I got the motherboard monitoring software updated and the new one works
The monitoring software is all useless flash. Fancy background, numbers that look like LEDís, non-standard controls. Guess why it didn't work right? Meanwhile, the real stuff it does is not a useful as it could/should be!! Pu-leeze, why not just make performance counters? Or how about a tiny thermomiter that sits in the system tray?
…and I saw that the CPU temp. reached 85 degrees C and leveled off, dipping to 84.5 and then back up again. Lucky, since 85 is the safe limit!
Lucky…? No, not a coincedence at all. The motherboard will cut the CPU clock in half when it reaches 85 degrees. So, it was running at the red line, more slowly.
Now I bought the Intel box with the heat sink and fan bundled, with the 3 -year warranty. If I bought a 3rd party cooler and the bare CPU, Intel would not warranty it. The fancy stuff is mostly looks and cool shapes anyway, right?
Nope. The boxed cooler is mediocre. The putty used to make good contact seemed suspicious since it was a thick layer and was not protected like having to peel off a backing first. Replacing that with a carefully-applied layer of the good stuff should help right there.
What bugs me is that the sheet in the CPU boxed set stated that the ambient case temperature should not exceed 38. Well, it didnít. I'm running at 36. So the CPU should not overheat, right?
Why do they sell something which doesn't work?
I was initially worried that I didnít know how much over the line it was. Today with some experiments, it seems that I'm just at it. When I repeated the same program, it just got to 83.5. Same case temperature. Room temperature might be a little different; I'm not sure. Humitity is different. Perhaps in other parts of the sales territory 27 is not considered a normal room temperature? Perhaps it works more reliably with only one HD, a smaller power supply, or different cables? Itís really marginal in meeting the requirements.
The reason Iím setting up a new one is because the old one ran too hot for too long. So getting it to run well and cool is part of the goal of this new machine.
Rather than taking half measures, Iím going with liquid cooling. Itís quieter than air fans for aggressive cooling, and they work so well. I read some recient product reviews on Tomís, and I think I saw some like that at Frys. I didnít take a close look at them at the time, but noticed that they had become mainstream. I must say, a meter-high blue anodized heat sink / resovoir is ďcoolĒ looking; nice and modern in style and form following function, too! But I got something more traditional in appearance, just a box that sits over the tower with ordinary-looking fan ports and a thermomiter readout. But itís a complete kit, is quieter than the Intel fan, and is among the best performers. The best performer is not quiet. The silent ones (passive tower only) donít perform well, or are high-priced aílaícarte items. Lots of choices out there, but I would only buy something I found a trustworty review of that included measured efficiency and loudness.
Gee, and it seems like only yesterday yíall were teasing me about my Pentium Pro and its reputation of being so power hungry, that I could put a pizza warming tray in a drive bay.
3 June 2007
I realized that liquid cooling was going mainstream when I saw three different brands on display at the local Fryís. There was beginning to be more to choose from than only DIY solutions.
I went with a kit by Koolance, because I didnít want to have to worry about it. They said that all the parts were engineered to go together, and all the fiddling—such as whether the pipe was the optimal size and whether there would be a galvanic reaction between any parts—was already solved.
It is a high-end brand, and quite pricey. But I figured it would outlast one computer, as well as extend the life and usefulness of the components. I got an Exos kit, which is no longer made but looks similar to their current offerings as I write this. It looks like Exos 2 but with 3 smaller fans instead of two larger ones, or like the Exos LT stretched one fan longer. I think the performance is between these two units, rated at 700 watts of heat removal.
The CPU heat block is very slim and light. It is a copper labyrinth on the bottom, and clear plastic on top. The copper only has to be thick enough to give it structure, and in fact thinner works better because the real cooling is done with water. So it is not heavy. The bottom is polished smooth and gold plated! Installing it on the motherboard was remarkably easy. The standard heat sinks have always been a pain, Intel worse than AMD, taking quite a while to get latched in place properly. This one was so easy—it went into the fixture without any force required, and then a thumb screw tightened it. The screw has a built-in torque limiter, so you just turn it until it starts to click.
The HD cooler is a metal frame with an expanding bag inside it. Two drives (or one drive and a dummy plate) screw to opposite sides, and the bag conforms to the circuit board. I thought that was perfect for a RAID, and as it used the vertical mounting holes in the drive, I was able to attach bay wings to the side holes of one of the drives and insert that into a full-height 5 inch bay. Now, if you ever thought getting the disk drives plugged in was difficult with the drive in place, imagine how much worse it would be with compression fittings! And how will I ever get it out again without leaking inside the case? I began to realize that keeping the tubing length as short as possible was not such a good idea.
Later, I decided to replace the cooler that came with the northbridge/southbridge chip, which was a small heat sink with a fan on it. Koolance had a selection of different configurations and mounting hardware, and it was quite easy to replace. It just requires careful choice of a cooling block with hoses coming off in the right directions, so nothing is in the way.
At that time, I didn't like the sound that the fan on the Matrox graphics card was making, so I replaced that too. I used another generic cooling block like the one I used for the motherboard, but with different hose connections. They have special units for some graphics cards, and I guess some are not so simple to operate on as this was.
What can I say? The cooling capability is awesome. Rather than barely working well enough like any regular heat sink with a fan, this shows no heat build up at all and is barely touching the systemís capacity. The temperature of the CPU cooling block was only 2 degrees higher than the temperature of the inside of the case. The temperature of the coolant at the radiator would be about 5 degrees F hither than the room air temperature, and the fans were still slow enough to be inaudible. It is so quiet that I can hear gurgling over the sound of the motors and fans. I was very pleased with it.
As noted 2 August 2004
Posted at The Pub
Today I installed the new liquid cooling system on Cassini, replacing the stupid heat sink/fan that Intel pushes on us.
First impressions: The thing is so well made, itís wonderful! Attaching it to the CPU was much nicer than Intelís, both in being easier and in being confidant that itís correct. With Intelís, a plastic part went flying across the room. With this, there is a screw that has a torque-limiter in the thumbscrew so it tightens down exactly the right amount with no fuss, and applies perfectly even pressure to the top of the die.
The liquid-filled heat sink plate is small since it doesnít need a huge mass or huge fins. It has spikes that the fluid circulates around, instead. The business end is polished flat and electroplated with gold to a mirror finish. It came with real high-quality thermal contact grease, with specifications listed on the ampule.
The external unit has redundant pumps, that are so quiet that you canít even tell they are on, once the air is filtered out. There are also 3 80mm fans that are rated at 12V but normally run at 5, and the whole assembly when operating is quieter than the Intel fan alone. The power supply fan is the only noise!
OK, itís quiet. The other good news is that it works. Before, I noted that after idling for a while the CPU temp was 63 or 64. Now, itís 51. More importantly, when running flat out for a while, it plateaued at 59.5, a far cry from the worrisome 85 it was before.
A number of years ago, it was unusual that I thought to probe the temperature of the computer, and contemplated putting cheap sensors in perminantly. Today, itís standard practice, and the BIOS monitors 3 different probes. Two or three years ago, an article in Tomís on how to build a liquid cooling system said how hard it was to obtain the plate to mount on the die. Now, they are at Frys. So, I predict that something more effective than a fan and alluminium heat sink will become mainstream on fast systems. An effective fan+sink weighs a kilogram and is made from copper, and interfacing it to the die takes more care than is casually performed.
3 June 2007
It was quite difficult to remove stuff and reconfigure the cooling circuit, when I added more cooling blocks. It didnít have a drain, and the connectors to the external unit close up when unplugged.
I ended up getting coolant onto the motherboard, and I am so glad I double-checked everything because I had some wicked up into the AGP slot. I had a terrible time trying to get it out of there. Perhaps a special tool, the thickness of the card to fit into the slot but with absorbent cloth on the sides, would be developed to keep in our toolboxes once it becomes more wide spread.
I also realized that this was the first genuinely new computer problem Iíve had in many years. There are connector and cable problems, and the connectors change but the problems are essentially the same. There is bad RAM, and the modules change but the issue is the same. Likewise with everything else you might have to fix or fiddle with. But radiator fluid and the capillary action of edge connectors? That is a new one in principle. I asked my wife to help, and she remarked, ďWhy do you have such weird things in your computer?Ē
And why do they put blue dye in the coolant? Its only purpose is to stain the carpet wherever it drips. I suppose it is for finding spills and leaks, like when I could see it against the gold edge pins. But why not use something that washes off? Actually, I like the idea of the UV reactive liquid that is clear otherwise. The glow would be easy to see when making sure nothing got on a circuit board, and any stain would be invisible without such a light.
In December I put together Pluto, which features an AMD CPU. Although faster, it does not run as hot as the Intel, and I did not need anything fancier than the fan it came with.
Until spring, that is. Once the room temperature got into the 80ís, perhaps much higher in this hot room on the south wall, I started having trouble. I believe the trouble was not with the CPU overheating, but with the motherboard. The heat sink on the chip was very low profile, as it has to fit under the PC-express card. They made it twice as wide as the chip, but with no fan. And it simply is not good enough.
So I installed the trusty Exos, cooling the motherboard chip, the CPU, and 4 hard drives. I needed some mounting hardware for the CPU cooling block which Koolance readily sold me, and I used one of the little blocks I had before for the motherboard. They sell a new one just for that exact purpose, but the hose size is different and I was able to get a generic one to work.
I did one thing differently, though. I bought connectors that click together and come apart when a flange is pressed. These were hard to find for a reasonable price! The web sites that have liquid cooling parts donít cary anything quite like that, or are ridiculously expensive. After searching a few times I found some from a manufacturer. Perhaps I should start reselling them.
On each cooling block I put a pair of hoses as ďleadsĒ with the quick connectors (one of each gender) on the ends. The leads were long enough that I could hold it away from the inside of the case to connect together or disconnect. So now I have lots of extra hose inside the case, but the extra resistance will not be nearly enough to put a dent in the cooling capacity.
I also looked at power supplies and cases, and I have another machine named ETC that I want to take a different approach with.