New CPU, New Watercooling System – My Upgrade Story

Detailed odyssey to a new watercooling system – Matthew Kaminski (aka Mops)

Two years ago, I built a water-cooling system for my mighty Athlon 1 GHz (AXIA stepping). I was very successful in overclocking it to 1.5 GHz with this setup. Over those two years, it served me extremely well, without a single failure! It required minimal maintenance, which basically came down to removing dust from the radiator fins every few months. Adding a fan filter could solve that, but I never got around to doing it.

Nevertheless, I consider this setup a full success. The problem was that it didn’t look very nice, or to be honest, that box looked really ghetto. Still, as I prefer functionality to looks, I continued to use it until that day in which I got my Barton 2500+.

I installed the CPU, raised the clock speed and core voltage and, after few minutes, the chip was running at 2.3 GHz at over 60ºC while idling!!! That setup just couldn’t handle that CPU at that speed. The water was hot, the homemade CPU block was inadequate (back then when I was putting together that water cooling system, luxuries like a commercial waterblock or watercooling radiators weren’t out yet) and the little 400 gph pump failed to provide decent flow.

I decided I have to do something about it. Then the idea of new watercooling system, utilizing the latest in watercooling design and construction, came about.

Another story was my Radeon 9700.

After volt-modding, this puppy ran pretty hot, even with majorly oversized heat sink. A long time ago, in the ages of the Celeron 300A and the days where CPU’s were putting out about 30-40 watts or thermal power, I used to run a water cooled peltier on my CPU. That was early year 2000. Once the clock speeds increased, so did the wattage that had to be taken away from the CPU’s.

Peltiers are pretty inefficient, and all though they still can be used to cool CPU’s nowadays, they do require a rather extreme approach while the benefits are not convincing (OK, maybe the temps are 20ºC less that you can get with water, but then how much further can you push the CPU in terms of MHz??? Another 50, maybe 100 if you are lucky), considering the resources needed to run them – excessive usage of electrical power, high risk of failure, as well you have to consider the fact that some measures of preventing condensation have to be made, which generally will void the warranty of your motherboard and CPU, etc, etc.

Overclocking for me is a way to get most for my hard earned cash and peltiers on a CPU just don’t fit that theory. I’m not saying that cooling today’s CPUs by peltier is not doable, I’m just saying that in my view, drawbacks of such a cooling technique outweigh the benefits. I still like Peltiers, but for the CPU, it’s out of the question.

However, a Radeon 9700 GPU core, which generates about 30 watts of heat…hmmm, and the idea in my head formed:

As I’m a frequent LANer, the requirements for my new box are:

• Fully portable, in one piece (i.e., no external watercooling components or power supplies)
• CPU cooled by water
• GPU cooled by water/peltier combo
• Does not have to look flash, but must be tidy
• Must have decent cooling performance
• Must be reliable
• Must be reasonably quiet (not silent – some degree of noise is tolerable)
• Must be reasonably cheap to build
• Easy access to all computer parts inside the case at any time (e.g. if I want to quickly throw in another CPU to see how good it overclocks)

Let’s start with particular components…{mospagebreak}

Matthew Kaminski (aka Mops) New Zealand

OK, we are dealing here with about 200 watts of heat, with about 100 watts from the CPU itself and around the same from the peltier + pump heat.

A search of forums revealed the best candidate – aka “Big Arse” by Cathar @ Overclockers.au forums. And here’s the issue that has been on my mind for a long time: You can install fans on the radiator in a blow or suck configuration (i.e. fans blowing onto radiator or sucking air through it). Both solutions have the same drawback, which is that only one side of the airflow is utilized to do actual work.

The is an airflow in front of the fans as well as behind them, so my idea was to sandwich two 120mm fans between two big-arse radiators. As more turbulent (read blowing air) airflow is better for cooling, hot water entering the radiator setup on the blowing-side of the fans radiator, to take most of the heat out of the water, which then enters radiator on the sucking side of the fans to further reduce the temps.

Now, there were few concerns about this setup on several forums I posted the idea. Many people think that air passing through radiator on the sucking side of the fans will warm up, hence decreasing the efficiency of cooling the other radiator. Yes, I admit, that is the case.

However, how much hotter is the air will be after passing the first radiator??? No one was able to answer that question. Neither can I, but it is a major factor in this argument. If the air is 10ºC hotter, than indeed, I’m out of line here and my setup is useless.

Here’s how it looks inside:

According to Cathar’s experiments, creating two separate “chambers” for each fan will improve performance of the radiator setup, due to the fact that the fans do not have to work against each other. Please note separator sheet between the fans. I do realize a little overkill of silicone.

However, mind that we are dealing with computer water-cooling system here, which is a very special cooling system. Special because the differences in temperature across all components of this system are minimal. What I mean by that is water that enters the radiator is usually less than 1 degree hotter than the water that leaves the radiator. Hence the air which enters the radiator (into which flows water already cooled by second radiator) will heat up only marginally, at least according to my theoretical understanding of cooling system.

I do not have any scientific data to back up my claims; however, after I built and ran my system, I measured that the water temp of the whole system is about 1 ºC higher that the ambient temp (which translates to C/W of about 0.005, assuming 200 watt load). Please note as well that I do not have any sophisticated measuring equipment to take readings, I’m using only cheap TT Hardcano integrated thermal probes, so all the results are “guesstimated”.

These are two 87′ Toyota Camry heater cores I got for NZ$9 each,

with fan shrouds on both ends and two 120mm fans (generic brand) sandwiched between them. Note marked airflow and waterflow directions. Here is the radiator without fittings soldered yet:

I used foam around the radiator core to prevent air leakage:

There are two waterblocks in the system:

The CPU waterblock is a LCC IceBlock. I think the company does not exist any more, so I can’t provide a link to them. Its waterblock is designed with cooling peltiers in mind, but it does a decent job at cooling my Barton CPU. However an upgrade of this block is scheduled to as soon as Cathar’s Cascade waterblocks will be available.

The waterblock for the Radeon is a homemade, multi-hole, cross-drilled waterblock. It does a very good job at cooling my peltier, which runs at 12 volts. The peltier itself is a TeDist Drift 0.8 – 172 watts @ 24 volts, 69 ºC max delta T and it runs of the same power supply as the rest of the components in the computer.

{mospagebreak}

Matthew Kaminski (aka Mops) New Zealand

The rest is pretty standard. I managed to acquire about 1 meter of 13mm Tygon tubing, which was exactly how much I needed. The waterpump is an Eheim 1048, which is just enough to push the water through the whole system. Probably more flow would be beneficial, but not absolutely necessary. Note the 1-liter reservoir mounted on the intake of the pump, constructed as described in a great guide HERE. That’s about it.

Which case to get was a major problem, in my case (no pun intended). Standard size mid-tower was out of the question – it was too small. Maybe it would be possible to fit all of that gear inside it, but it all would be cramped, and then any maintenance, like replacing computer components, would be a pain in the @#$%^&*.

Another factor is the way I decided to mount my radiator setup. I wanted air blowing from one side of the case to the other side; that is, sucking cold air from outside and dumping hot air outside the case, on the other side. That way the air passing through the radiators does not affect the airflow inside the case. Besides, the unusual physical dimensions of my rad setup does require an unusual placement in the case.

I needed something a bit wider and a bit longer than a standard mid-tower. 3R Case Systems came to the rescue with their LandRover.

Perfect match – although it did require some massive modifications in order to fit in my radiator setup on the front of it. I found that Dremel rotary tool was a bit, ummmm, underpowered for this kind of cutting job, so I just used a small grinder instead. Pretty straight forward, isn’t it? So here it is once I was done with it:

You can compare it to the one from the link above. Here are pics of the case with water-cooling gear and motherboard fitted:

Obviously, some sort of condensation prevention will be needed, as the GPU will be cooled to below ambient, and possibly below freezing. The waterblock was covered in neoprene and there is a slice of neoprene on the back of the GPU as well. Between the card itself and the neoprene sheets, there’s a generous amount of silicone dielectric grease.

Once I bolted everything together, I removed the excess grease. A good guide on how to do that can be found HERE in the articles section. Soon I intend to perform GPU and memory vmod for my Radeon. Have look at the pics of the waterblock, coldplate and the completed thing. Please note that screw nuts are integrated into the coldplate.
{mospagebreak}

Matthew Kaminski (aka Mops) New Zealand

Once I had all the parts, putting them together was easy, although time consuming – it took me about two days.

A particularly time-consuming part of this project was the GPU/peltier waterblock: How hard can it be to drill 21 holes in a block of copper??? Well, it was harder than I thought – even considering that I had access to some very decent equipment, such as a milling machine, a big drill press and a few engineering experts. There are two 3.5mm drill bits stuck in that waterblock, which I couldn’t remove – I hope they don’t decrease performance too much.

Once everything was together, I pressed the power button and waited for the POST while holding my breath. I was particularly concerned for my Radeon, as it underwent quite few modifications from the cooling perspective and it wasn’t tested afterwards. A second or two later, I heard very familiar “beep”, the screen turned on and I went directly to BIOS.

CPU temps were at stock speed and voltage were about 26ºC. I waited there for few minutes, watching temps, and they were stable. I booted into Windows, ran some benchmarks and played some games. After 30 minutes, I looked inside the case, only to find some major condensation on the bolts holding the GPU waterblock/peltier/coldplate/radeon combo. Those bolts:

That was fixed by adding some petroleum jelly (aka Vaseline : ) and rubber caps onto those bolts.

The last thing to do was to cut out some fan and radiator holes in the side panels of the case. As well, I replaced the standard IDE cables by MSI green rounded cables, which I was able to get very cheaply thanks to haytona @ Overclockers.co.nz forums.

After running this system for about two weeks now, I find it very reliable. I’m very happy with my new system, although even now I can see some room for improvements.

I’m a bit unhappy with my CPU temps. Obviously, the choice of waterblock wasn’t very good, but it was very economical; moreover, the relatively small coolant flow rate contributes to high temps. A Cascade waterblock will replace the current waterblock, once they are available again (the backorder is few months long). CPU temp hovers around 40ºC at idle and goes up few degree under load.

The temps on the back of the GPU are hovering between 1ºC and 12ºC, depending upon room temp and GPU load. I can monitor this temp by special external temp probe of my ASUS A7N8X deluxe. I can get those temps down even further by using an external power supply for the peltier that gives more juice, particularly I have built several peltiers, power supplied from an old AT computer power supplied as described there.

Matthew Kaminski (aka Mops) New Zealand

System Specs

• Asus A7N8X deluxe rev 2.0 motherboard (1007 bios) – nForce2 ultra 400 @ 215 MHz FSB
• Barton 2500+ CPU @ 2370 MHz @ 1.85V, multiplayer locked 9
• Two sticks of TwinMos 256 MB DDR400 RAM, @ 215 MHz 8-2-2-2 dual channel
• PowerColor Radeon 9700 non pro with Infinion 3.3ns RAM, stock speed 270/275 core/mem clocked to 450/310, VGPU = 1.65V, VMEM = 3.2V
• Hyena (generic) 500w power supply
• Two Seagate ‘cuda 7’s 120 GB each
• Lite-ON CDRW/DVD combo drive

Shortly after completing this project, I was offered to swap my pump (Eheim 1048) and my reservoir for an Eheim 1250 pump. I was happy with the deal, and the next day I built a very similar radiator to the original one – in fact, they are the almost the same, just the return water fitting is slightly lower (the original reservoir had it a bit too high due to measurement miscalculation). As well, the bigger pump’s intake fitting was fitted to accommodate bigger pump intake (3/4″).

My idea was that if I increased the flow, then CPU temps will drop. Hence, I decided to replace the “L” shaped hose joiners with a custom bent copper hose. Now there’s only one “L” shaped hose joiner, just on the output side of the pump.

During the whole procedure, a terrible thing happened: One of the wires connected to the peltier broke off in the worst imaginable place – exactly where the wire goes inside the peltier. Luckily, I had an 80 watt peltier lying around, so after two hours of disassembling the whole Radeon construction, then assembling it again, my upgraded rig was ready to go.

Now, I have to admit, even due to the larger pump and improved tubing, the flow didn’t increase by much. CPU temps decreased by about 5ºC under load, and about 2ºC at idle. However, the water temp went up by another 1¸C – it must be due to the fact that I have a different peltier. This one is rated at 80 watts @ 16v. It’s running at 12v off my rig’s power supply and it’s most probably generating more heat than the previous one. I’ll most probably remove the Radeon vmods, as the gains are pretty pathetic: 20 MHz on the core and 10 MHz on the memory. The performance improvement is not worth the risk of burning it.

The obvious bottleneck of my cooling system now is the waterblock. I’ll get a Cascade as soon as they will be back in stock (might wait till February 2004 for that). Maybe I’ll post an update once I have it in place. If Santa will being me drill press for Xmas, then I might build my own Cascade waterblock.

Another possible upgrade would be a chipset waterblock. While testing my board with unlocked CPU I found out that it will do stable 225 MHz FSB, while I’m only running at 215, and run benchmarks at 218, so I don’t think that’s absolutely necessary just yet. As well fan filter for the radiator is absolutely necessary, otherwise the radiators will be covered in dust very soon.

Special Thanks (in no particular order) to

• Auckland’s Howick College tech department, for assistance with the drill-press, advice and other miscellaneous tools
• Tom, my flat mate, for silicon, advice, and bearing with me when I was cutting metal and grinding at 1 AM
• Martinez, for his help and commitment in lapping waterblocks

Matthew Kaminski (aka Mops) New Zealand