quad socket G34 water cooling list, need sugestions

G

geok1ng

2[H]4U
Joined
Oct 28, 2007
Messages
2,130
After completing the SR-2 based Chess server, now it is time to experiment with a quad Opteron Server for chess.

luck smiled on me and a fellow [H] sold a few Extra Spicy Opterons, which are pretty much the only CPUs that can (possibly:confused:) beat my SR-2 in price vs performance for chess servers.

The case is an Aerocool Strike X ST

water blocks: cuplex kryos Delrin für Sockel G34, G1/4 this was the only place where i could locate 4 blocks for G34

radiator and fans: Bundle airplex PRO 240 G1/4 + 2x airstream 120 mm

this one was easy as well, since i was looking for quiet operation. that ruled out high FPI and thick radiators. I do know that 2x240mm is overkill for a server without a gaming GPU, But 2x240 low FPI, not so thick gave me better sound/temps than a high performance 360mm rad. The airstream fans are so-so, but their price does not hurt


pump and monitoring: aquastream USB 12V pump - Ultra version I need remote monitoring and control, and the only software solution that i found was this Eheim based solution. If there a way to monitor and network control a loop using a D5 pump, i would like to know how it compars to AC software.

|Reservoir: aquainlet XT 150 ml with nano coating, fill level sensor and LED holder, G1/4 a fill level sensor monitored by software was too cool. Plus this reservoir combos nicely with the pump, saving space and reducing loop restriction

Subtotal: 352,27 EUR :(

Would you choose anything different?

Do i have enough pump?

martinsliquidlab say that dual pumps in series adds reliability, but i have heard that when one pump fails the other faces even higher restriction and ends up failing as well. i have an spare DDC ( inside an apogee drive block) that i could mount in series. Would you add another pump?

and finally, the part i do not get at all: choose barbs for blocks and tube.

i am loost in the alternatives. i tend to favor 3/8 and 7/16 ID, but won't refuse 1/2 ID.

I am too stupid to actually locate and source compatible barbs, adapters, fittings and tubes. what goes with what. by all means put me away from any stuff that can cause galvanic corrosion, only silver and brass connections please, not F. nickel plated gunk.:mad:

The pump has an optional flow sensor. What is the need of this? I already monitors the temps, pump and water levels. why do i need to monitor the flow? Is that worth the + 10% total build cost?

finally, loop design:

rad1 > reservoir/pump > CPUs 1-2 > rad2 >CPUs 3-4 >rad1

is that the optimal solution?
 
Last edited:
Thinner radiators will not be better than thicker radiators with 1200 RPM fans. It's at the sub-800 RPM range that thinner radiators will be better than thicker ones.

A 360 radiator will always outperform a 240 radiator, regardless of fans used or design. For radiators, there is no replacement for frontal surface area. Except for maybe fan speed.

There are D5 pumps with RPM monitoring like fans. You can use your motherboard fan header and any fan monitoring software.

For 4 CPU blocks, it will probably be fine, but be the bare minimum. I don't know where you got the misinformation about a pump causing another pump to die, a dead pump in a loop is no different than adding another CPU or GPU block. By that logic, you would need a pump for each component in your loop. Two identical pumps is better than two mismatched pumps, but you can use it for peace of mind.

3/8" ID 5/8" OD offers the best flexibility. 1/2" ID 3/4" OD IMO looks better, but offers nothing performance-wise over 3/8" ID.

Nickel plating problems only really occurred with EK blocks. 99.9% of nickel plated fittings are fine (I have had fittings last 3+ years, still no problems). Besides, nickel flaking causes gunk buildup, NOT galvanic corrosion. Galvanic corrosion, for the sake of computer watercooled systems, occurs between aluminum and copper parts.

You buy fittings to match your tubing. ID (and OD if you get compression fittings) has to match. That's all you need to know about that.
 
Thank you for your help.

You are correct about thick/ thin rads: for 4those concerned about noise, it is only below 1200 rpms that thinner rads beat thicker rads in noise/performance. cooling performance of high fpi rads vs low fpi rads is another matter: very dense fin spacing does not work well with lower fan speeds. looking at bundymania's round ups and another sources, 45mm thick is the turning point of performance. 60mm thick rads do not outperform 45mm thick rads at 1200rpm for large margins. on the other hand, not every 60mm rad is more expensive than 45mm ones.

the choice of 240mm rads is about flexibility . with 240mm rads i can put dual PSUs on this case, which is not possible with 280mm or bigger rads.

about dual pumps in series, martins liquid labs said that flow meters work like a health monitoring system on the loop, allowing the user to detect pump failures or gunk build up before temps raise significantly or a complete fail occurs. in a dual pump scenario, a flow meter makes a lot of sense. my chosen pump has a temp sensor included, so value and features are much better than RPM D5s

i also discovered that for such a complex loop a dedicated drain T line is essential for maintenance, hence i am adding a drain plug after the pumps.

if i decided to use quick disconnects, where would they be most useful?
 
Last edited:
If you wanted to use the minimum number of QDCs, use them where you're most likely to be swapping components. Since the components most likely to be swapped are your CPU blocks, use it around those.

Also, regarding loop order... loop order doesn't matter as long as you maintain a high enough flow rate.
 
You do know that AquaComputer makes an AquaStream version of the D5 that links in with their software right?

http://www.frozencpu.com/products/1...B_and_Aquabus_Interface_41093.html?tl=g30c107

It is even designed to be able to connect directly to your Aqualis reservoir.

http://www.frozencpu.com/products/1..._for_D5_Pumps_Compatible_w_Aqualis_41094.html

If you are dead set on using two pumps (you really don't need it) you could also run them in parallel instead of series. Parallel gives you more max flow, Series gives you more head pressure. If you have a failure in series the restriction in your loop goes up due to the dead pump, if you have a failure in parallel the restriction doesn't change, you just lose the extra flow. But really, you don't need two pumps. If your pump fails you should be able to configure the software to gracefully shut down the computer before anything bad happens. If the software fails the computer will shut itself down before any damage happens due to thermal safeguards that all CPU's have now.
 
Last edited:
Martin's Liquid Lab showed parallel pumps to actually be detrimental, as what happens is rather than going through the loop, the water goes through the dead pump. Two separate loops can be done, but don't do pumps in parallel.
 
Tsumi said:
Martin's Liquid Lab showed parallel pumps to actually be detrimental, as what happens is rather than going through the loop, the water goes through the dead pump. Two separate loops can be done, but don't do pumps in parallel.
Click to expand...

Good point, ultimately though two pumps isn't really necessary.

Only two reasons for dual pumps:

1) You need the performance that two pumps can provide.
2) You need 100% uptime and can't have the system turn off if a pump fails.

You don't need two pumps with your loop, so unless you can't live with your computer shutting down if a pump fails then just stick with one pump.
 
I agree with Erasmus, you don't need more than 1 pump for your loop. If you want redundancy (which in a home PC environment I don't see why you really would need 100% uptime), I would probbaly either buy or homebrew a system that checks the tach signal of 1 pump and if it goes to 0 a microcontroller switches the relay on for a second pump in series sitting idle. You can have a dead pump in a series loop and it doesn't really affect flow very much.

Also, you don't need to worry about the order of your loop. Even under 1000W of heatload, the inlet and outlet temperatures of the water are going to be within a few C of each other unless you have a trickle of flow. I'm speaking from personal exp. here.

And the D5 pumps I'm running now have been running 24/7 (probably 95%+ uptime) for over a year and they're still going strong.
 
Erasmus354 said:
You do know that AquaComputer makes an AquaStream version of the D5 that links in with their software right?

http://www.frozencpu.com/products/1...B_and_Aquabus_Interface_41093.html?tl=g30c107

It is even designed to be able to connect directly to your Aqualis reservoir.

http://www.frozencpu.com/products/1..._for_D5_Pumps_Compatible_w_Aqualis_41094.html
Click to expand...

thank you for point me towards this alternative :)

Even cheking prices in Euros ( which are a lot cheaper than US$ right now, go figure, going with a D5 based makes me lose a lot:

the D5 solution costs more than the Eheim based one, does not include a sensor temp and is not compatible with the flow sensor. adding all that with the cost of the reservoir adaptor and going D5 would cost twice as much as using a Eheim pump.

i already have a secondary pump, so using it in series on the loop will not increase costs.

reliability is essential on the build, because when i need it most, as in preparing for a game in between rounds of a chess tournament i am away from home, with a couple of hours to spare and no one at home to repair the loop. usually the server is rented 24x7, so uptime is money as well. a dedicated server machine earns 1 Euro/3h or more, depending on performance and market demand.
 
As you pointed out in your first post, you don't need the flow sensor. It tells you nothing from a reliability stand point that you don't already know from the pump tach reading.
 
You must log in or register to reply here.
Back
Top