- last edited on by
ROGBot
This build is more of the cool performance than of the cool looks. You'll be sure of that after you see my workmanship. The first post is about motive, background and plans.
Back in August, there were discussions about chilled water cooling. Menthol, WhitePaw and Henkinator among several others led those discussions. One casual comment by HiVizMan really switched on the bulb for me about why that was worth the trouble.
At the same time, I was overclocking and benching a rather good Anniversary Edition Pentium. That low-powered two core CPU was an opportunity. I could try chilling and finally use some curiosity gadgets that were laying around – Peltier effect Thermo Electric Coolers (TEC).
I put one cooler between two universal waterblocks and split my water loop in two. The good EK reservoir with D5 pump, CPU block and the block on the cold side of the TEC formed one loop. The hot side TEC block, radiator and a not-so-good pump and reservoir I had replaced were the other loop. I had a simple on-off controller and MOSFET controlling the TEC. Theoretically, that arrangement should have chilled down to 15C with the Pentium idling. It barely chilled below room temperature. What the …
Oh, Duh. The D5 pump is water cooled and rivals an idling Pentium for power dumped. After I scrounged another two waterblocks, a second TEC went in. Here’s the PC with chiller parts outboard, literally.
That chilled down under 10C, but just once. The coldest point is the cold-side TEC waterblock. That’s where condensation shows up first. With the whole-house chiller (air conditioning) running in the Texas summer, condensation happened at around 14C. I set the controller to switch off the TEC power at 14C and back on at 16C.
The on-off controller almost worked. TECs pump heat from cold to hot when they are powered. That’s how they chill. They continuously conduct heat from hot to cold – conductivity about 4W/mK, about like a mediocre TIM. Peltier pumping overcomes the conduction when the TEC is powered, but when power is off conduction undoes the chilling. The off time was surprisingly short. Thirty seconds off and three minutes on.
I built up a PWM controller that works much better – the electronic gadget in the next picture.
Chilling performance was at last useable. In the Aida64 plot, yellow is the hot-side water temperature (used by the M6F to control radiator fans), green is the cold-side water temperature (used by the PWM to control the TECs) and red is the hottest core temperature in the CPU.
The chilled Pentium got some really good benchmark scores. With 650ml water in the cold side loop, core temperature would stay below 20C over a complete run of Realbench. The Pentium set a very nice mark for 2 cores in Realbench on HWBOT.
The downside was the impact of all that experimenting on my one and only PC. It was down for a day each time I changed anything. When the smoke got out of the MOSFET one time or I needed more waterblocks, there were days of downtime.
I need another rig for extreme OC without risking downtime to replace CPU or cooling components. OS damage from crashing is another risk I don’t want on the 24/7 PC. I also want to OC other CPUs on other motherboards while leaving the 24/7 in a stable configuration. The solution is an open testbench with a chilled watercooling arrangement built in.
Some objectives:
1. Bench to be built from shop scraps – I’m also into woodworking and have a lot of scraps.
2. TEC chiller cooling independent of any motherboard. The TECs need true PWM that responds to cold-side water temperature and radiator fans should respond to hot-side water temperature. Until the R5E, there were no motherboards with multiple temperature sensor inputs AND true PWM headers controlled by those inputs. (M7 series have PWM fan headers, but only one sensor input.) I plan to use the R4E and M6F I have on hand at the very least.
3. Chill cold-side water to 4C – a 20C drop from room temperature
4. No more than 1C temperature rise per minute when running CPU and GPU in benchmarks.
What those mean:
1. is easy for me. I envision a desk-shaped table with cooling equipment in a bay under the table, MB, monitor, keyboard etc on the top – made of wood and the needed scraps are on hand.
2. means that I have to make or buy a PWM controller, but that’s already designed and tested. The chiller will also need its own PSU for total power and for independent switching. It’s convenient to chill down the water before powering on the MB.
3. Operation below the dew point means that cold-side blocks, reservoir, pump and tubing have to be insulated. Motherboard, video card, RAM etc have to be protected from damage from condensation. That’s a combination of coating on the board and insulation.
4. I could go into the calculations, but it works out to at least a ¾ horsepower (560 max watts) chiller with at least 4 liters of water in the cold-side loop. I choose to use six TECs rated 110 watts each at 12v. That allows for 1/8 hp of operating margin.
The cooling plan:
One question is the relative cost between a custom-made chiller and buying one. Operating costs of the TECs is not considered here. First, what of all the above is ‘chiller’ and what would be in an ordinary water cooling loop. I think the stuff inside the blue dotted line is ‘chiller’ since I’d need one pump, a radiator of about the same cost, a reservoir and all the blocks around the MB anyway. The added parts are the dedicated PSU, PWM controller, TECs, their two waterblocks, the second pump, the large reservoir and maybe half the fans.
Planned costs, all in US$
80 Coolmax PSU at Microcenter after rebate
20 PWM controller bits and pieces
14 Six power MOSFETs
255 Six 50mm square TECs
85 Five 140mm fans
22 PVC pipe, fittings and barbs for 4 liter reservoir
80 D5 pump (Koolance 450)
400 Pair of TEC waterblocks custom made.
956 Total.
I think that’s less than ¾ horsepower chillers I’ve seen and probably rivals the cost of the ½ hp size. There’s also the value of my preference to build rather than buy. It’s a hobby after all.
Current state of the plan is that most of the pieces are bought or built and are ready to assemble. The next post should be soon.
Jeff
Back in August, there were discussions about chilled water cooling. Menthol, WhitePaw and Henkinator among several others led those discussions. One casual comment by HiVizMan really switched on the bulb for me about why that was worth the trouble.
At the same time, I was overclocking and benching a rather good Anniversary Edition Pentium. That low-powered two core CPU was an opportunity. I could try chilling and finally use some curiosity gadgets that were laying around – Peltier effect Thermo Electric Coolers (TEC).
I put one cooler between two universal waterblocks and split my water loop in two. The good EK reservoir with D5 pump, CPU block and the block on the cold side of the TEC formed one loop. The hot side TEC block, radiator and a not-so-good pump and reservoir I had replaced were the other loop. I had a simple on-off controller and MOSFET controlling the TEC. Theoretically, that arrangement should have chilled down to 15C with the Pentium idling. It barely chilled below room temperature. What the …
Oh, Duh. The D5 pump is water cooled and rivals an idling Pentium for power dumped. After I scrounged another two waterblocks, a second TEC went in. Here’s the PC with chiller parts outboard, literally.
That chilled down under 10C, but just once. The coldest point is the cold-side TEC waterblock. That’s where condensation shows up first. With the whole-house chiller (air conditioning) running in the Texas summer, condensation happened at around 14C. I set the controller to switch off the TEC power at 14C and back on at 16C.
The on-off controller almost worked. TECs pump heat from cold to hot when they are powered. That’s how they chill. They continuously conduct heat from hot to cold – conductivity about 4W/mK, about like a mediocre TIM. Peltier pumping overcomes the conduction when the TEC is powered, but when power is off conduction undoes the chilling. The off time was surprisingly short. Thirty seconds off and three minutes on.
I built up a PWM controller that works much better – the electronic gadget in the next picture.
Chilling performance was at last useable. In the Aida64 plot, yellow is the hot-side water temperature (used by the M6F to control radiator fans), green is the cold-side water temperature (used by the PWM to control the TECs) and red is the hottest core temperature in the CPU.
The chilled Pentium got some really good benchmark scores. With 650ml water in the cold side loop, core temperature would stay below 20C over a complete run of Realbench. The Pentium set a very nice mark for 2 cores in Realbench on HWBOT.
The downside was the impact of all that experimenting on my one and only PC. It was down for a day each time I changed anything. When the smoke got out of the MOSFET one time or I needed more waterblocks, there were days of downtime.
I need another rig for extreme OC without risking downtime to replace CPU or cooling components. OS damage from crashing is another risk I don’t want on the 24/7 PC. I also want to OC other CPUs on other motherboards while leaving the 24/7 in a stable configuration. The solution is an open testbench with a chilled watercooling arrangement built in.
Some objectives:
1. Bench to be built from shop scraps – I’m also into woodworking and have a lot of scraps.
2. TEC chiller cooling independent of any motherboard. The TECs need true PWM that responds to cold-side water temperature and radiator fans should respond to hot-side water temperature. Until the R5E, there were no motherboards with multiple temperature sensor inputs AND true PWM headers controlled by those inputs. (M7 series have PWM fan headers, but only one sensor input.) I plan to use the R4E and M6F I have on hand at the very least.
3. Chill cold-side water to 4C – a 20C drop from room temperature
4. No more than 1C temperature rise per minute when running CPU and GPU in benchmarks.
What those mean:
1. is easy for me. I envision a desk-shaped table with cooling equipment in a bay under the table, MB, monitor, keyboard etc on the top – made of wood and the needed scraps are on hand.
2. means that I have to make or buy a PWM controller, but that’s already designed and tested. The chiller will also need its own PSU for total power and for independent switching. It’s convenient to chill down the water before powering on the MB.
3. Operation below the dew point means that cold-side blocks, reservoir, pump and tubing have to be insulated. Motherboard, video card, RAM etc have to be protected from damage from condensation. That’s a combination of coating on the board and insulation.
4. I could go into the calculations, but it works out to at least a ¾ horsepower (560 max watts) chiller with at least 4 liters of water in the cold-side loop. I choose to use six TECs rated 110 watts each at 12v. That allows for 1/8 hp of operating margin.
The cooling plan:
One question is the relative cost between a custom-made chiller and buying one. Operating costs of the TECs is not considered here. First, what of all the above is ‘chiller’ and what would be in an ordinary water cooling loop. I think the stuff inside the blue dotted line is ‘chiller’ since I’d need one pump, a radiator of about the same cost, a reservoir and all the blocks around the MB anyway. The added parts are the dedicated PSU, PWM controller, TECs, their two waterblocks, the second pump, the large reservoir and maybe half the fans.
Planned costs, all in US$
80 Coolmax PSU at Microcenter after rebate
20 PWM controller bits and pieces
14 Six power MOSFETs
255 Six 50mm square TECs
85 Five 140mm fans
22 PVC pipe, fittings and barbs for 4 liter reservoir
80 D5 pump (Koolance 450)
400 Pair of TEC waterblocks custom made.
956 Total.
I think that’s less than ¾ horsepower chillers I’ve seen and probably rivals the cost of the ½ hp size. There’s also the value of my preference to build rather than buy. It’s a hobby after all.
Current state of the plan is that most of the pieces are bought or built and are ready to assemble. The next post should be soon.
Jeff
Making the pieces.
Here are the 800ml and 4 liter reservoirs on the table of the capacity adjustment tool. The bulky US IPT fittings are easier to get into round PVC. They’re in with epoxy.
I bought a Mo-Ra3 radiator core – US$80 less than a finished product. The idea is to make a fan board with shrouds for the sides of the core and mount the board to the M4 threaded holes profided. The routing template lets me make the right size holes in close to the right places.
Fan board on the radiator core.
Reverse view showing shrouds around the edges of the core.
Radiator with fans mounted in the cooler bay of the bench. I did say it was to be of scrap wood, didn’t I?
The most costly item in the chiller is the pair of waterblocks that can take six 50x50mm TECs. These are 3 inches (7.6cm) wide by 2 feet long (62cm). The outer wall with all the screw holes in it is 1cm. The lids are 1/8 inch (3mm) thick. The good ol’ boys running the local machine shop had to special order the G1/4 tap for the fitting holes.
A detailed look at the fitting end of a block.
Two blocks assembled. The G1/4 barbs attach to the cold plate side of the block.
Testing with the radiator, one of the waterblocks, the small reservoir and a D5 pump. The Coolmax PSU no longer has a warranty. Those are now 10 gauge wires carrying 12v to the TECs.
Three TECs with their hot sides on the waterblock. From left to right, TIM used was Prolimtech PK-1, Arctic Silver 5 and MX-4. I’d had some trouble with MX-4 on the small waterblocks in the first trial. The problem turns out to be curvature on the CPU block I was using – be sure to use flat coldplates with TECs. In this run, all three TECs had thick coats of ice and frosty fringes. All three TECs, and the waterblock worked.
On a 20C day in my garage, the radiator held the temperature down to 22.5C or so with the fans at full speed. That was working, too. The disappointment is the water flow rate – 5.4 liters/minute. I tried testing flow with just the radiator and pump in the loop and got exactly the same flow at the 5 settings of a D5 vario. Flow was 7.6 liters/minute with a PMP-500. I expected the D5 to be able to do that.
Six TECs on the coldplate of one block along with closed cell foam insulation between them. The foam is 1mm thicker than the TECs, so it will seal off much of the air that could flow in there. TIM on both sides of the TECs and on both blocks is PK-1. The PK-1 spreading tool is in the background. Sandwich ready to close.
And closed.
Through screws hold the sandwich together. Nylon shoulder washers try to prevent contact and thermal short circuit through the screws. TECs like 70 – 200 psi pressure in a mount such as this. Four Bellevelle spring washers on each screw, two screws straddling each TEC, give an estimated 140 psi.
Now with power rails and MOSFETS. The hot side waterblock is heatsink for the MOSFETs.
That's most of the stuff. Now to put it together.
Jeff
Here are the 800ml and 4 liter reservoirs on the table of the capacity adjustment tool. The bulky US IPT fittings are easier to get into round PVC. They’re in with epoxy.
I bought a Mo-Ra3 radiator core – US$80 less than a finished product. The idea is to make a fan board with shrouds for the sides of the core and mount the board to the M4 threaded holes profided. The routing template lets me make the right size holes in close to the right places.
Fan board on the radiator core.
Reverse view showing shrouds around the edges of the core.
Radiator with fans mounted in the cooler bay of the bench. I did say it was to be of scrap wood, didn’t I?
The most costly item in the chiller is the pair of waterblocks that can take six 50x50mm TECs. These are 3 inches (7.6cm) wide by 2 feet long (62cm). The outer wall with all the screw holes in it is 1cm. The lids are 1/8 inch (3mm) thick. The good ol’ boys running the local machine shop had to special order the G1/4 tap for the fitting holes.
A detailed look at the fitting end of a block.
Two blocks assembled. The G1/4 barbs attach to the cold plate side of the block.
Testing with the radiator, one of the waterblocks, the small reservoir and a D5 pump. The Coolmax PSU no longer has a warranty. Those are now 10 gauge wires carrying 12v to the TECs.
Three TECs with their hot sides on the waterblock. From left to right, TIM used was Prolimtech PK-1, Arctic Silver 5 and MX-4. I’d had some trouble with MX-4 on the small waterblocks in the first trial. The problem turns out to be curvature on the CPU block I was using – be sure to use flat coldplates with TECs. In this run, all three TECs had thick coats of ice and frosty fringes. All three TECs, and the waterblock worked.
On a 20C day in my garage, the radiator held the temperature down to 22.5C or so with the fans at full speed. That was working, too. The disappointment is the water flow rate – 5.4 liters/minute. I tried testing flow with just the radiator and pump in the loop and got exactly the same flow at the 5 settings of a D5 vario. Flow was 7.6 liters/minute with a PMP-500. I expected the D5 to be able to do that.
Six TECs on the coldplate of one block along with closed cell foam insulation between them. The foam is 1mm thicker than the TECs, so it will seal off much of the air that could flow in there. TIM on both sides of the TECs and on both blocks is PK-1. The PK-1 spreading tool is in the background. Sandwich ready to close.
And closed.
Through screws hold the sandwich together. Nylon shoulder washers try to prevent contact and thermal short circuit through the screws. TECs like 70 – 200 psi pressure in a mount such as this. Four Bellevelle spring washers on each screw, two screws straddling each TEC, give an estimated 140 psi.
Now with power rails and MOSFETS. The hot side waterblock is heatsink for the MOSFETs.
That's most of the stuff. Now to put it together.
Jeff
A few more of the pieces: The PWM controller takes thermal sensor inputs from the hot and cold water loops. Op amps set gain and levels. 555 timers perform pulse width modulation. Power is connected through a floppy disk PSU header – just had to find a use for one of those plugs in the PSU cable set. Radiator fans are connected via two 5-way splitter cables that get power from a PSU molex.
I set up the Rampage IV Extreme and R9 290X with air cooling to 1) check them out before investing any time in them and 2) provide some show and tell for visitors we had last month. Auto OC settings ran OCCT stability at up to 4.6GHz in the 3960X SB-E on Hyper 212 Evo air cooling. Little Sandy booted into Windows at 5.0GHz, but any testing produced more than 200 watts of power – the air cooler couldn’t handle that. The optical drive was only connected for installing the operating system onto the SSD. Hereafter, the SSD will be the only storage and most benchmark programs will run from RAMDrive.
With all the chiller parts assembled under the table, here was the first flow test. The only waterblock involved was the Koolance 380i CPU block, dangling by its tubes.
See those valves on the vertical board? They were intended to isolate the components above the table for easy changing – shut off the rest of the system with the valves, then drain everything above the table through the small plugged block in the foreground. One time I opened the block plug before closing the valves. That let air in and the water drained down into the reservoir – no need to isolate or waste any water as long as there is room in the reservoir. In the test shown above, cold loop water was chilled to 15.5C and hot side water was at 29.5C – both good enough for a short condensation-free test. Again water flow was disappointing – 5.4 liter/min with a PMP-500 and only one waterblock in the loop. The little 3/8 inch IPT barbs went with the valves.
Another issue cropped up with the hot loop 800ml reservoir. The short stubby tank sat well below the top of the radiator. The reservoir was never intended to hold all the water in that loop, but it should drain gracefully. It didn’t. Water rushed down from the radiator and the carpet got some before I could position the bucket. That reservoir is now a long tube – 1.5 inch (37mm) diameter and 22 inches long (56cm). It’s bottom is below the radiator so the radiator will drain and its top is above the radiator. It primes the pump well and radiator draining starts with I pour from the reservoir.
This picture shows water components above and below the bench top. The four liter reservoir with the 800ml tube behind it are at the right. PMP-500 for the hot loop and D5 vario for the cold side are mounted on the bottom board. The radiator is mounted behind all those and behind the dedicated 1000W PSU for the TECs. The TEC block sandwich is on the bottom, below the radiator. The HX-1050 PSU for the motherboard is on a shelf under the benchtop where all its cable can reach up through holes drilled in the bench.
The next test had no valves and I let it go long enough to really chill down. Eight liter/min flow is much better. Cold loop at 6.9C is in the target range, but check the condensation on everything. This might actually work, but it will take more prep.
Preparation of the VGA – Conformal coating was sprayed on both sides to protect against condensation. An extra layer was brushed onto the board around the VRMs, RAM and GPU. Little dabs of PK-1 TIM on VRM coils and RAMs transferred correctly in a contact test. The substance on the GPU chip and its contact area of the waterblock is liquid metal, hence the extra insulating protection.
With the waterblock installed.
The R4E got the same treatment. After the stock PCH/VRM cooler with fan and heat pipes was removed, the board was masked with tape covering electrical and thermal contact points – sockets, headers, PCH chip, VRMs, switches and the rear I/Os. Both sides were sprayed with conformal coating. It looks a little more glossy in these pictures from that coating. PK-1 TIM was used on the VRMs and PCH under the EK waterblocks.
Another test, this time to be sure all the handling and sprayed layers haven’t hurt the boards. Cold water temperature was kept above the dew point and everything ran. Water flow was down to 5.4 liter/min, but this is with two constrictive blocks in the loop – CPU and VGA.
Note the wooden support for the VGA. It has a cutout for the DVI sockets and the motherboard audio.
I set up the Rampage IV Extreme and R9 290X with air cooling to 1) check them out before investing any time in them and 2) provide some show and tell for visitors we had last month. Auto OC settings ran OCCT stability at up to 4.6GHz in the 3960X SB-E on Hyper 212 Evo air cooling. Little Sandy booted into Windows at 5.0GHz, but any testing produced more than 200 watts of power – the air cooler couldn’t handle that. The optical drive was only connected for installing the operating system onto the SSD. Hereafter, the SSD will be the only storage and most benchmark programs will run from RAMDrive.
With all the chiller parts assembled under the table, here was the first flow test. The only waterblock involved was the Koolance 380i CPU block, dangling by its tubes.
See those valves on the vertical board? They were intended to isolate the components above the table for easy changing – shut off the rest of the system with the valves, then drain everything above the table through the small plugged block in the foreground. One time I opened the block plug before closing the valves. That let air in and the water drained down into the reservoir – no need to isolate or waste any water as long as there is room in the reservoir. In the test shown above, cold loop water was chilled to 15.5C and hot side water was at 29.5C – both good enough for a short condensation-free test. Again water flow was disappointing – 5.4 liter/min with a PMP-500 and only one waterblock in the loop. The little 3/8 inch IPT barbs went with the valves.
Another issue cropped up with the hot loop 800ml reservoir. The short stubby tank sat well below the top of the radiator. The reservoir was never intended to hold all the water in that loop, but it should drain gracefully. It didn’t. Water rushed down from the radiator and the carpet got some before I could position the bucket. That reservoir is now a long tube – 1.5 inch (37mm) diameter and 22 inches long (56cm). It’s bottom is below the radiator so the radiator will drain and its top is above the radiator. It primes the pump well and radiator draining starts with I pour from the reservoir.
This picture shows water components above and below the bench top. The four liter reservoir with the 800ml tube behind it are at the right. PMP-500 for the hot loop and D5 vario for the cold side are mounted on the bottom board. The radiator is mounted behind all those and behind the dedicated 1000W PSU for the TECs. The TEC block sandwich is on the bottom, below the radiator. The HX-1050 PSU for the motherboard is on a shelf under the benchtop where all its cable can reach up through holes drilled in the bench.
The next test had no valves and I let it go long enough to really chill down. Eight liter/min flow is much better. Cold loop at 6.9C is in the target range, but check the condensation on everything. This might actually work, but it will take more prep.
Preparation of the VGA – Conformal coating was sprayed on both sides to protect against condensation. An extra layer was brushed onto the board around the VRMs, RAM and GPU. Little dabs of PK-1 TIM on VRM coils and RAMs transferred correctly in a contact test. The substance on the GPU chip and its contact area of the waterblock is liquid metal, hence the extra insulating protection.
With the waterblock installed.
The R4E got the same treatment. After the stock PCH/VRM cooler with fan and heat pipes was removed, the board was masked with tape covering electrical and thermal contact points – sockets, headers, PCH chip, VRMs, switches and the rear I/Os. Both sides were sprayed with conformal coating. It looks a little more glossy in these pictures from that coating. PK-1 TIM was used on the VRMs and PCH under the EK waterblocks.
Another test, this time to be sure all the handling and sprayed layers haven’t hurt the boards. Cold water temperature was kept above the dew point and everything ran. Water flow was down to 5.4 liter/min, but this is with two constrictive blocks in the loop – CPU and VGA.
Note the wooden support for the VGA. It has a cutout for the DVI sockets and the motherboard audio.
Next, thermal insulation. Kneaded eraser seals around the edges of the VGA waterblock – no air flow = less condensation.
Here is 3/16 inch (4.7 mm) closed cell foam applied to the back of the VGA. The foam collides with DRAM sockets, so has cutouts to clear.
Conformal coating appears as a thick, shiny layer in this shot of the back of the motherboard. Closed cell foam covers the back of the motherboard in the chilled area.
Edges of the motherboard waterblocks are also sealed to the board with eraser. The PCH area is insulated by a ½ inch (13mm) thick closed cell foam with armaflex tape over some of the seams. Insulation is cut away where the VGA will fit in. The CPU block, VGA and barb fittings limit the insulation over some parts, so there is only kneaded eraser or tape.
The VGA block is covered with ½ inch foam and armaflex tape over the seams and over water connections. The VGA is nestled into its cutout over the PCH cooler.
The relatively neat arrangement of tubing gets crowded with insulation over the tubes. That’s armaflex tape covering barbs and the ends of the insulation tubes.
Insulation under the bench – ½ inch foam and armaflex tape over the cold TEC block and cold loop tubing.
The cold reservoir is wrapped with foam. The PMP-500 impeller area is also insulated, but not the motor cooling fins.
Better view of hot and cold loop reservoirs
It all still works. I'm ready to declare victory and end the building phase. Now to see how the R4E and SB-e overclock and bench. The point of easy-to-change tubing and having the cooling independent of the motherboard will show up later when I switch to the M6F and then something with Haswell-E.
Here is 3/16 inch (4.7 mm) closed cell foam applied to the back of the VGA. The foam collides with DRAM sockets, so has cutouts to clear.
Conformal coating appears as a thick, shiny layer in this shot of the back of the motherboard. Closed cell foam covers the back of the motherboard in the chilled area.
Edges of the motherboard waterblocks are also sealed to the board with eraser. The PCH area is insulated by a ½ inch (13mm) thick closed cell foam with armaflex tape over some of the seams. Insulation is cut away where the VGA will fit in. The CPU block, VGA and barb fittings limit the insulation over some parts, so there is only kneaded eraser or tape.
The VGA block is covered with ½ inch foam and armaflex tape over the seams and over water connections. The VGA is nestled into its cutout over the PCH cooler.
The relatively neat arrangement of tubing gets crowded with insulation over the tubes. That’s armaflex tape covering barbs and the ends of the insulation tubes.
Insulation under the bench – ½ inch foam and armaflex tape over the cold TEC block and cold loop tubing.
The cold reservoir is wrapped with foam. The PMP-500 impeller area is also insulated, but not the motor cooling fins.
Better view of hot and cold loop reservoirs
It all still works. I'm ready to declare victory and end the building phase. Now to see how the R4E and SB-e overclock and bench. The point of easy-to-change tubing and having the cooling independent of the motherboard will show up later when I switch to the M6F and then something with Haswell-E.
