WARNING: Ethyl Glycol can kill. This I find out after I filled my loop.
How I Watercooled My PC
I have a 3.2Ghz Pentium 4E with a Prescott core. This type of processor has been dubbed "press hot" as they are one of the hottest running CPU’s on the market. In fact, they run so hotly that Intel quickly redesigned and released a new version "Cedar Mill" that runs much cooler.
Who needs a heater when you have a Prescott?
Typical idle temperatures for a processor are around 30-40°C, and up to 50-55°C when under load. My Prescott with the Intel heatsink and fan runs at about 54°C when IDLE and rises to 69°C under load! When the processor reaches 90°C its time to start worrying as we are now approaching meltdown. The processor will automatically shut down at this temperature to prevent damage.
For any electrical component, temperature is enemy number one. There is an unwritten rule that a 10°C increase in temperature will half the lifespan of a processor. For example, a processor should last 10 years running constantly at 45°C. When running at 55°C constantly, the life span will be expected to only be 5 years.
Previous Cooling Attempts
Soon after I installed this processor I purchased a Zalman flower cooler and several more fans to help with the temperatures, bringing them down to 50°C idle and 65°C load. This cooler is made of pure copper and has a much larger surface area than the Intel heatsink. It also features a much larger diameter fan. The problem with more and larger fans is that they create a lot of noise. I also tried lapping the heatsink to see if that would help. It didn't do much.
The larger the surface area on the heatsink, the more heat can be transferred to the air. Larger fans can move more air through the heat sink and are measured in cubic feet/minute (cfm).
Another (hot) year later, the extra fans and new heatsink were still not really dealing with the issue, just moving hot air around. I decided that it was still running far too hot (not to mention that the heat produced heats the room up). I could get even more fans (which create more noise), a different heat sink (which may not help much), constantly run the air conditioning on full power or I could try something a bit more extreme...
After much careful consideration, I decided to water cool my system. This decision was made after carefully considering many other options, such as a bigger heatsink and fan, ducting to direct cold air onto the processor, extractor fans and getting a blower fan for the Cooler Master Stacker case. This would also greatly increase the noise levels of the computer (which is already loud). I was after a solution that would offer the greatest cooling for the least noise, and water-cooling is the most efficient method of heat dispersal and is nearly silent in operation.
I have never used water cooling before, so I decided to buy all the parts in kit form, rather than purchasing all the parts separate (and finding that I missed something). The kit I chose was the Swiftech H2O Apex Ultra+, which has won many awards and has a good reputation within the overclocking community as being a high performing, top quality kit. It was ordered late Friday afternoon from Overclockers.co.uk and it arrived early the following Monday.
This kit contains all the parts required to cool the processor, chipset and the graphics card, as well as all the hardware to install the kit on most computers platforms.
The Swiftech kit is supplied with the following items:
- Apogee CPU Water-block (3/8" and 1/2" barbs)
- MCW60 VGA kit
- MCW30 Chipset water-block
- MCR220 Radiator with fans
- MCB-120 R2 "Radbox"
- MCP655 pump
- MC14 Ramsinks
- MCRES Micro Reservoir
- Hydrx coolant
- Various noise reduction accessories
- AM2 bracket
The kit contains all the hardware that you could need, plus lots of extras. There is a different bag of bolts for each different fitting (e.g. LGA, Socket 478, ATI and nVidia) as well as an assortment of mounting brackets. The water blocks had a very heavy solid quality feel to them far from the light flexible feeling of cheapness. The tubing was thick PCV and resisted kinking, however, it was difficult to bend to shape (more on that later). The instructions are only diagram format (like a flat pack!), but easy to understand.
The first thing that I had to do is to remove the motherboard and replace the standard heat sink mount with the one provided in the kit. I took out every component apart from the PSU, as I knew that I would have to drill some holes in my case.
I didn't have to do anything to the Intel water block as it already had the correct bracket installed, so I gave it a test fit on the mounting screws. The chipset block had the AMD bracket installed, but it didn’t interfere with anything and I could still use the Intel clips, so I left that as well.
Finally, the GPU block was set up for a four-screw attachment, so I had to remove that bracket; there was no need to install a different one as the bracket converted two screws into four. I have also fitted VGA Ram sinks onto the chips to aid heat dispersion.
With all the components prepared with appropriate mountings and the water blocks with the correct brackets, they were set aside with the rest of the components ready for installation later.
I needed to mount the reservoir higher than the pump (which is at the lowest point in the loop) but the brackets supplied did not line up with the holes on my case, so I had to drill a new hole for it. With that done I reinstalled the motherboard in the case and fitted the reservoir and radiator, then installed the graphics card. I fitted the water blocks on the components so that I can measure out the length of pipe needed.
Measure Twice, Cut Once
With the entire pipe cut to the right lengths, it was time to attach them to the blocks. You need to warm the pipe with boiling water as this will make the pipe more flexible and expand over the barbs, they can then be secured with the supplied clips. Once these hoses are on, they are not coming off so make sure they are correct! My original plan was to assemble the kit outside the case and test for leaks in the bathtub, however, the piping proved too inflexible and incredibly difficult to remove from a barb so I only had one option – install in the case without testing. I did, however, remove all electrical components and I used power from a remote power supply while testing.
The route that the water takes in my "loop" is:
Reservoir > Pump > CPU > Chipset > GPU > Radiator > Reservoir
The first thing to do is to fill the reservoir with coolant. Water is still the best thermal transfer fluid for the temperature ranges involved in home PC liquid cooling. The kit recommends distilled water and Swiftech's own green HydrX additive, I have however used deionized water and antifreeze. De-ionised water is considered non-conductive, as the ions have all been removed, however, the longer it runs in the system the more conductive it becomes. Adding antifreeze has the added benefits of making the water less viscous (easier to pump around the system), acts as anti-corrosion, antibacterial and anti-fungal agent and its designed to run through car radiators, so a heater matrix core should be fine.
Ethylene Glycol itself is not toxic, but it is metabolised in cats and dogs into several extremely toxic chemicals that are responsible for potentially lethal effects. Cats and dogs will readily drink antifreeze as it smells sweet, and it may only take a teaspoon full to be fatal.
Using good common sense can avoid any incidents, don't leave antifreeze lying around unattended. When using antifreeze in an open environment (i.e. mixing in a jug) keep all pets out the room, or safely shut in another room. A water-cooling loop is a closed environment as long as there are no leaks.
I am now looking for a new coolant...
For more information please read Ethylene glycol toxicosis in cats.
As the reservoir is filled up the liquid will pass down to the pump and eventually you can see it coming out the top of the pump towards your components. When the level of the reservoir stops going down, gravity has done its part and you must now use the pump.
Never let the pump run dry, terminal damage will result in seconds. The pump will empty a reservoir of this size in less than a second!
I filled the radiator up as much as possible and set the pump to its slowest speed and blipped the power on/off. To do this I used a paper clip in the power cable to the motherboard. By bridging two pins you can turn the power supply on when you remove the paper clip the power goes off.
Refill the reservoir and repeat as many times as is necessary to fill the loop. I then ran for a few minutes checking for leaks, while moving the water blocks around to displace any trapped air bubbles. After filling the reservoir again, I left the system running for a few hours, before turning up the speed on the pump. Testing lasted 6 hours during which I gradually increased the pump speed. There were no observable leaks in the loop, so I syphoned off excess water in the reservoir and tightened the cap down.
It was then time to install the water blocks on the components, using the supplied Artic Ceramique compound. The components were all installed, as they would be when the build is complete. The loop was then tested again for a further 6 hours at the highest pump speed (50PSI or 3.5 bars) before finally adding all the other components and finishing off the build.
My last attempt at wiring the case was a little messy, so this time I was determined to have a cable tidy session. The CoolerMaster Stacker isn’t a cramped case, so there is plenty of nooks and crannies to hide cables in.
Even though I had thoroughly tested the loop for leaks, I was still nervous about turning the system on living for the first time. If there were any leaks now, it will go bang and wreck every component in the system. Luckily nothing went bang, and it powered on into the BIOS just as I planned it would.
At first, I had to double check it was all working, as it was so quiet! I am used to a case with so many fans running at high speeds that I had to visually check that they were running, which they were.
According to the BIOS temperature reading, the processor (CPU) is now running at 43 °C, where it was previously reading a temperature of 55°C. I left it on this screen for about half an hour to check that the temperatures were looking good, and I set a shutdown temperature of 70°C just to be safe. I then proceeded to boot into Windows.
I can now hear my hard drives working! All the fans on the old set-up were so noisy I couldn't even hear the hard drive heads clicking away.
As soon as Windows had loaded, I checked the temperature with Motherboard Monitor 5 and was delighted to see a temperature of only 43°C. Usually,, this is 50°C, maybe 49 on a cold day. More impressive was the graphics card temperature (GPU) was running at 32°C, down from about 45 when idle.
So far I am very impressed with the sound levels and the cooling performance!
- Windows XP Professional SP2
- DirectX DirectX 9.0c (4.09.00.0904)
- ATI Radeon X800 XT Platinum (R420)
- ABit IC7-MAX3 (BIOS 6.00 PG)
- Catalyst Driver 8.205-060104a
- 3DMark 2005 Free (1.2.0)
- AtiTool 0.0.23
- SuperPi 1.2
In order to fairly compare my old air-cooling solution with my new water-cooled system, the results must be scientifically obtained. Before I started this project, I formatted my system drive and did a clean install of Windows XP. Hardware drivers and software were loaded and this is the benchmark platform. For more details of driver and software versions, please see the box opposite. The same software and drivers were used in all tests. After each test, the computer was restarted. After Windows had finished loading, three instances of CLI.EXE were terminated using Task Manager. This is just the Ati control panel background task. All other tasks were left running. No applications were running apart from Motherboard Monitor 5 (MBM5), AtiTool and the test program.
MBM5 was set to take readings at 10-second intervals, and write the data to a log file. AtiTool was used to pass GPU temperature readings to an MBM5 sensor so that it can be logged in the same way.
3D Mark 2005 was used to heat test components, as this program will stress CPU and GPU to their limits.
The base results that I will be using to compare my results are taken using my old set-up consisting of the Zalman flower cooler, 4x 120mm fans, 3x 80mm fans and an exhaust blower. Fan speeds are set using Zalman multi-fan controller to be as they would if I were gaming.
Testing consisted of letting the machine idle for 15 minutes and averaging the temperature over this time.
This constitutes the IDLE temperature.
MAX is the highest temperature that the software recorded, similarly MIN is the lowest temperature recorded.
The LOAD temperature is an average of the temperatures taken while the system is under load. MBM5 is set to log CPU usage in the log file.
The first program run was Ati Ruby: Double Cross demo. This demo stresses the X800 GPU by utilising all available pipelines and uses the most complex shaders to create a graphically impressive animated sequence. The system was then allowed to cool down for a little, before being benchmarked by 3D Mark 2005.
Again, the system was allowed to cool down before the CPU was stress tested using the SuperPI program. This calculates the value of Pi up to 32 million digits and the complex maths involved will stress the CPU. The values given in brackets are the number of digits to calculate, 64K = 64,000 and 1M = 1,000,000. The system was then left to return to idle temperatures to conclude temperature testing.
Baseline Results are taken at 200mhz Front Side Bus (FSB) as standard for the processor, GPU is also set to standard speeds.
|Baseline Results CPU||Baseline Results GPU|
The base test produced a 3D Mark score of 5863 points, broken down as follows:
|Game Type 1||25.6 fps|
|Game Type 2||17.4 fps|
|Game Type 3||28.9 fps|
|CPU Test 1||2.1 fps|
|CPU Test 2||3.2 fps|
But with nothing to compare this to, these numbers are meaningless, so without further ado, here are the test results, which were carried out in exactly the same way with the water-cooling setup.
|Water Cooled CPU||Water Cooled GPU|
Air-cooled is coloured Orange, while water-cooled is coloured in Blue.
Immediately we can see that the water-cooling has lowered every temperature measured, with the biggest difference being the GPU temperature – reduced by a massive 34°C , that’s half the temperature it was running!
3D Mark 2005 produced a score of 6159 – 296 points higher than air-cooling. This test was done using exactly the same settings, the same programs running. The only difference is the water-cooling.
|Test||Air Cooled||Water Cooled||Increase|
|Game Type 1||25.6 fps||27.1 fps||+5.8%|
|Game Type 2||17.4 fps||18.4 fps||+5.7%|
|Game Type 3||28.9 fps||30.0 fps||+3.8%|
|CPU Test 1||2.1 fps||3.2 fps||+52.3%|
|CPU Test 2||2.0 fps||3.2 fps||+60%|
In the following graph, you can see the huge difference that water-cooling has made to the temperatures, the top graph shows the GPU, while the bottom graph shows the CPU.
I have highlighted the areas that represent the tests, the first being the ATI Ruby demo and the second 3D Mark 2005. You can even see the different tests being performed by the dips in the GPU temperature.
The first thing we can see is that as well as the temperatures are much lower, the temperature differential is also a lot less, and there are nowhere near as many or as high peaks.
From this, I can conclude that water-cooling has made a huge difference in temperatures of the system, in many cases the hottest the system gets under water-cooling is far less than the system used to get while cold.
The last test that was done was the SuperPI calculations. This program calculates Pi to a huge degree of accuracy and really gets the CPU working. Here are the results from the Air Cooled vs. Water Cooled tests.
|SuperPi Test||Air Cooled||Water Cooled|
|64K||00m 01s||00m 01s|
|128K||00m 03s||00m 03s|
|256K||00m 08s||00m 07s|
|512K||00m 20s||00m 18s|
|1M||00m 42s||00m 40s|
|2M||01m 35s||01m 36s|
|4M||03m 36s||03m 18s|
|8M||07m 36s||07m 22s|
|16M||20m 09s||16m 21s|
Again, these were all taken using the same clock speeds; nothing has been changed apart from the cooling method. On the last result, water-cooling has reduced the time taken for the calculation by 4 minutes!
I can only surmise that the reduced temperatures increase the efficiency of the processor. In the 3DMark tests, we also saw that the cooler temperatures resulted in higher performance. I wonder what will happen when I start overclocking the system?
Overclocking is the process of taking the performance of a component beyond its design parameters. The process takes advantage of the fact that most components are produced on the same production line and are identical to higher spec parts apart from some settings programmed in. For example, an Intel Pentium P100 is designed to run out the box at 100mhz. The Pentium P120 is exactly the same but programmed to run out the box with a different setting. Overclocking allows you to run the slower processor at 120mhz, so you gain extra performance, yet haven't paid any extra for the higher spec part. Extreme overclockers take the processor speed far above anything currently on the market; some overclocked computers are currently running at 4.7Ghz! Overclocking is most commonly carried out by increasing the Front Side Bus (FSB) from within the computers BIOS. Each type CPU has its own Multiplier setting which when multiplied with the FSB gives you the core processor speed. The higher the overclock on the chip, the more heat is produced, so a good cooling system is vital to overclockers.
Overclocking will invalidate all warranties for the product(s)
Processor on Air-Cooling
The highest stable overclock I could achieve on my old setup was an FSB of 220mhz, which when multiplied by my CPU Multiplier of 16 gives a core speed of 3520mhz or 3.5Ghz. The standard FSB for this processor is 200mhz x 16 = 3200mhz.
I say stable overclock because the FSB will go higher, if I set it to 223 the system will not boot up, at 222 it crashes often, at 221 it crashes sometimes and at 220 I haven't had any crashes, so 220 is the highest stable overclock, and with temperatures knocking on the door of 70°C it was too hot to be comfortable with.
I ran the same tests as above with the processor overclocked and the following temperatures were observed:
|CPU Overclocked to 3520mhz|
3D Mark Score was also up, this time to 6102 points.
Graphics Card on Air-Cooling
The other component that can be overclocked is the graphics card. These run on core speeds, mine standard is 520mhz and the memory runs at 560mhz. Both these values can be increased for a boost in performance. Again the trade-off is higher temperatures. The highest my card would go is 525mhz core and 564mhz for the memory, which isn’t much of an overclock, but looking at the temperatures it's about to blow up anyway. If the clock speeds are increased further the screen gets corrupted and the computer crashes.
With everything running as fast and as hot as possible the final figures can be obtained which will give the highest performance from my old air-cooled system.
|GPU Overclocked to 525/564mhz (core/mem)|
This system is getting very hot now, and I wouldn’t run this all the time – it's just too hot. 3D Mark 2005 produced a score of 6179 with the following results, which are only slightly better than the stock card.
|Game Type 1||27.4 fps|
|Game Type 2||18.4 fps|
|Game Type 3||29.9 fps|
|CPU Test 1||2.4 fps|
|CPU Test 2||3.7 fps|
So how does the system perform with the water-cooling setup? The same tests are done with the same CPU and GPU overclock speeds.
|CPU Overclocked to 3520mhz Water Cooled||GPU Overclocked to 525/564mhz (core/mem) Water Cooled|
3D Mark 2005 produced a score of 6348 with the following results, which is a considerable improvement on the air-cooled setup.
|Game Type 1||28.1 fps|
|Game Type 2||18.9 fps|
|Game Type 3||30.8 fps|
|CPU Test 1||2.3 fps|
|CPU Test 2||3.5 fps|
Strangely the CPU tests are slightly worse with the water-cooling, but the GPU has improved by a few fps. With the Front Side Bus overclocked to 220mhz, SuperPi was again run, with the following times:
|SuperPi Test||Standard Air Cooled||Overclocked Water Cooled|
|64K||00m 01s||00m 01s|
|128K||00m 03s||00m 03s|
|256K||00m 08s||00m 07s|
|512K||00m 20s||00m 16s|
|1M||00m 42s||00m 37s|
|2M||01m 35s||01m 26s|
|4M||03m 36s||03m 02s|
|8M||07m 36s||06m 42s|
|16M||20m 09s||14m 44s|
How High can I go?
Water-cooling has shown that by lowering the temperatures you can increase performance. We also know that overclocking has huge performance gains as well, so combining them should make for a lightning fast machine.
When overclocking, it is fairly difficult to damage any component by increasing the FSB or clock speeds. The computer will crash or restart before any damage is done, however, it is important to increase the clock speeds GRADUALLY in order to find the limits. It is also important to increase one value only. It's no good increasing the FSB and the graphics card because if it crashes what caused the crash?
I have been experimenting with different values, gradually increasing the speeds until I reached the highest values possible. After this point, the system starts becoming unstable. These values have been achieved by only increasing the Front Side Bus and the graphics processor clock speeds, higher values can be had by increasing the voltages, adjusting the timings etc...
Here is a sample of the data I collected finding my highest overclock.
The final figure is 6851 on 3D Mark 2005 achieved with a Front Side Bus speed overclocked from 200mhz to 240mhz (3200mhz to 3840mhz core) and the graphics card overclocked from 520mhz/560mhz to 573mhz/585mhz.
|3d Mark Test||Stock||Water Cooled and Overclocked||Increase|
|Game Type 1||25.6 fps||30.3 fps||+18.3%|
|Game Type 2||17.4 fps||20.5 fps||+17.8%|
|Game Type 3||28.9 fps||33.1 fps||+14.5%|
|CPU Test 1||2.1 fps||2.5 fps||+19%|
|CPU Test 2||3.2 fps||3.8 fps||+18%|
|SuperPi Test||Standard Air Cooled||Overclocked Water Cooled|
|64K||00m 01s||00m 01s|
|128K||00m 03s||00m 03s|
|256K||00m 08s||00m 06s|
|512K||00m 20s||00m 15s|
|1M||00m 42s||00m 33s|
|2M||01m 35s||01m 19s|
|4M||03m 36s||02m 59s|
|8M||07m 36s||06m 08s|
|16M||20m 09s||13m 21s|
I think that the main thing we can deduce from this is that water-cooling is far more efficient at cooling today’s components than air-cooling fitted as standard or aftermarket upgrades. We can also conclude that the lower the temperature of the component, the more efficient (faster) it performs.
I was really impressed with the quality and performance of the Swiftech kit, and after nearly a week I’m still amazed at how quiet my system has become.
Although the kit took me nearly four days to install, I was taking my time and triple checking everything. It was still a big step forward in terms of PC modification, but I think the end result was worth the effort.
Last updated on: Friday 8th September 2017