Some more FrankenResults
Some more FrankenResults
I spent four weeks working to overcome the previous limitations, achieving 4.7ghz. I moved from air cooling to water cooling and constructed a custom loop from scratch, except for the CPU block. This week’s testing pushed me further with the water loop, reaching 4.8ghz stable enough for demanding use. This milestone followed a significant adjustment—reducing all bus frequencies in the BIOS to 800MHz, including memory, hypertransport link, and north bridge. This change kept the processor isolated, aiming to improve its clock multiplier performance. The temperatures remained close to the maximum tolerable for a high-performance build.
Despite this, 4.9ghz still struggled during the torture test even with a low system bus speed; the heat was clearly the limiting factor once again. However, progress continued. I plan further refinements to the loop, starting with adding copper fins to my Jardiniere and enhancing the water cooling. I also created a secondary reservoir for an additional pump, ensuring it operates safely within its 35°C limit. The new pump delivers 8psi or 2-6m head, compared to the first’s 2m head. I’ll relocate it into a sealed reservoir connected via glue to the loop, allowing it to pull the CPU block and return flow effectively.
My goal is to boost the return feed’s efficiency—either it will function properly, or the pump will naturally increase pressure and improve flow. I’ll test this by using a small plastic container as a temporary housing for the pump, adding fittings and glue. If I run it during startup, it should draw more water from the return line, pushing the CPU block and enhancing flow beyond 1 liter per minute. I’m hopeful this will help, though I can’t be certain yet.
I removed the system and tested it with just the radiator, short tubing, and two pumps in series. The flow was 4 litres per minute with one pump; when the other was turned on, it increased to around 9 litres per minute.
From a previous conversation here, I began noticing several issues in my loop. Initially, I suspected the CPU block was the problem, but it turned out to be some faulty couplings.
The fittings in the plumber’s kit narrow the water channel at the coupler without any clear reason—there’s just a small opening and a wider fitting. If a 10mm coupling is used, the channel becomes 8mm; with a 15mm-10mm coupler, it’s still only 8mm. I’ve decided to tackle the issue in a few ways.
First, I’ll drill out the coupler using a 10mm or 9mm bit. This should widen the channel by about 20%. Second, I’ll remove the compression fitting entirely. Instead of compressing it, I can connect the 15mm hose directly to the 10mm side of the coupler. The screw threads will secure it tightly. The 15mm tube has a 12mm ID, but the screw threads add another 2mm. It still works, so I can proceed. The tube won’t come off easily—it needs to be rotated. I’ll use hose clamps for this.
I plan to replace three of these fittings around the loop to better understand the CPU block. I’m also thinking about taking it apart and drilling a larger opening at the top for a bigger aperture. It’s just perspex with an ‘O’ ring and screws to keep it watertight. For that, I’d need a few bigger fittings.
I might also switch to 3/4 inch tubing, as some recent docs suggest it doesn’t significantly affect pressure drop. Why? Because my pumps are weak. They work fine on the radiator alone, but the tubing, fittings, and CPU block—plus gravity—might be reducing pressure and lowering flow.
Each adjustment should help improve the flow and overall cooling performance of the loop. I hope it doesn’t require buying too many parts if it works out; just a few fittings and some extra tubing should suffice. Using larger tubing could let me install bigger couplers, which would reduce pressure loss once drilled.
The plumber won’t be available until tomorrow. I’ll need to revisit the design and try again. There’s still room for improvement before investing in a new CPU block or a high-pressure pump.
Just a quick note on some physics. Liquids absorb heat much more slowly than air, so they require time to take in heat from a heat exchanger (like the CPU pickup and radiator). If the fluid flows too quickly, it can cause overheating.
At a flow rate of 1 GPM, 264W is needed to increase water temperature by one degree. You might be curious about why rivers don’t freeze as quickly in winter—flow plays a key role. If you wish to explore the calculations and additional information, visit the provided link.
One GPM corresponds to roughly 3.8 LPM, while lower flow rates around 0.5 GPM or 1.8 LPM are often ideal. Generally, higher flow improves the temperature difference and enhances cooling performance.
For context, closed-loop coolers usually operate near 1.0 LPM.
Some insights from the discussion: This topic led me to consider how fast flow affects overheating. After adjustments, the water temperature fell from 35°C to 31°C at full capacity; however, the CPU still reached up to 55°C on the package and around 75°C in the core.
@4.9ghz that’s the situation; during stress testing, I successfully passed a 4.8ghz test on p95 for two hours—sufficient for the AMD overclock club. At 4.9ghz, performance dropped after 15 minutes.
Testing from the sink showed a flow rate of about 4 LPM on the return path, while the second pump increased significantly—consistent with the known maximum flow rates of the pumps at 2m and 6m heads.
With the CPU block installed, the return flow decreased sharply to 1.5 LPM. This reduction likely stems from increased vertical pumping needed to reach the radiator, along with narrowing tubing (from 12mm to 8mm) and coupler channels (from 10mm to 8mm).
But handling the edge devices really boosts performance. By annealing the base of the copper jardiniere I’m using as a reservoir, removing the dirt and oxidation, the water temperature fell by 4 degrees. I was operating at 35°C, which was near the pump’s maximum temperature; it won’t go any higher. Now it’s around 31°C for normal operation and maybe up to 33°C under stress. All this indicates there are still many adjustments I can make.
I was just checking the WC thread, and they reported a flow of 1.5 gallons per minute with 650 watts of dissipation, depending on the fans and radiators. They used that setup. My loop isn’t reaching even half of that.
I calculated the volume of my radiator at 1,102,725 mm³, with a delta time of 7 to 9 seconds. My operating TDP is around 209 watts plus 45 watts for the two pumps, totaling about 254 watts. With a flow rate of roughly 1.2 liters per minute, that’s about 0.26 US gallons per minute. But these numbers are just estimates. Another radiator with a surface area of 1,190,000 mm² dissipates only 235 watts at a delta time of 10 seconds.
On my radiator, water passes through five 12mm copper pipes before returning to the reservoir. I have a box of aluminium fins left from my Arctic A11; I could fit them between the fins to greatly increase the surface area. That’s something I can try now.
Overall, the calculations are still uncertain because they’re based on estimates, and adjusting variables for my specific setup is tricky. In short, it looks close, but I’m not sure how accurate the numbers are.
Certain tasks need to be approached through trial and error, yet numerous kits are available to offer a foundation for building or adjusting according to the particulates in your system.
It requires more than just a basic kit. Acquiring comprehensive knowledge is essential before making purchases; then enhancing your system through overclocking on a water loop truly becomes a mysterious skill. Additionally, I am currently experimenting. My Jardiniere is intended to function as both a storage tank and a cooling device. Tomorrow I plan to purchase solder and flux; subsequently, I will attach a heatsink to the base and copper fins surrounding its perimeter. My aim is to refine their edges for a more refined appearance, though if that fails I’ll solder them flat. It’s quite challenging to solder a small section of conductive material. You must warm the entire assembly before applying heat, or alternatively cool it in a bucket and focus only on the targeted area. The process is far from straightforward. When soldering the base, excessive heat causes the fins to detach; when soldering the fins, the base detaches. Ideally, there should be a stable equilibrium between the stove and a bucket of water to ensure functionality. As long as I avoid soldering them too densely, or if I join two or three at once, the outcome should remain manageable. This will eventually evolve into the first Victorian-style radiating reservoir adorned with decorative elements.
That's why I suggested starting with a kit, not purchasing it outright but merely to understand each component and its features. Many of these items could be replaced with alternatives. For example, someone used a car heater for the radiator and another person used a main radiator from a small car. A motorcycle-style radiator could also be suitable. Some are opting for high-capacity fish tank pumps, and adding an active chiller from a water fountain into the circuit or through a heat exchanger is another option. This approach could help the system perform better in cooler room or ambient conditions.
I've completed the motorcycle transmission cooler and the aquarium pump, which can achieve 1 g/pm without pressure loss; however, in practice the return flow reduces to just a trickle of 0.26g/pm.
I considered purchasing a water chiller, but I'm uncertain about finding something compact enough to fit on a desk, affordable, and compatible with the existing water system without complications.
My refrigerator is quite large—essentially a box—so drilling holes for pipes would be necessary, or I'd need to buy components that might not be easily adaptable.
Where exactly can one locate such items? Typically, these products are sold in retail stores with significant price increases and aren't easily modified.
A heat exchanger? What does that mean? I assumed it was similar to a radiator, but they seem different. Where can I find one? What are the prices, dimensions, and compatibility details? I'm not sure how to source them without spending a lot.
The motorcycle transmission cooler measures 31.5x18.5x1.9cm, offering a surface area of approximately 110 million mm with 11 fins per inch across six tubes and five bends for water circulation.
Another off-the-shelf option is a 120mm radius model with an area of 1,190,000 at a delta t of 10cm, dissipating 235 watts.
But it has fewer turns and only eight fpi. Do these features matter? I've read about them but don't know much about their availability or cost.
The turns might be important—though I've heard mixed things.
Concerning condensation, I've mentioned it several times; I assure you I don't want the water temperature to drop below ambient. Maintaining a temperature of around 20°C is sufficient.
I've searched for parts, but nothing has provided clear guidance or affordable options that meet all my requirements without major compromises.
Currently, I'm exploring ways to boost the surface area of my jardiniere, adding more fins to the radiator, installing additional fans, attaching heat sinks to the base of the reservoir, and using more extensive tubing with larger couplers to enlarge the flow opening by 20%.
These adjustments would only cost a few pounds—perhaps a few pounds for tubing, and a few pounds for soldering and flux.
I'm not aiming for premium equipment; I just need to find a solution that works. It's clear I require more cooling, so I'll try these alternatives today. The radiator isn't a problem—it cost only £5, and replacing it would be unnecessary if it doesn't perform adequately. It could be moved to a secondary or GPU role.
Still, I'm not at the point where further modifications won't require spending money, so I'll attempt this today and update as I progress.