The Z Build(s) 3.0 - Go Big and/or Go Home

[Zarathustra]

I have been considering making some changes to the water loop layout as well.

Consideration #1:

Splitting up dual D5's to two separate D5's and having each power just one block of the game machine (one for CPU, one for GPU.)

My plan has originally been to dedicate a dual pump top of the two blocks in the game machine, but I have been thinking. Dual pumps (in series) DO scale, but they do not scale well. Generally if you want better flow, it is better to reduce head pressure. I bet I could get better flow over both the blocks, if I run them as separate sub-loops from the same reservoir, with just one pump each.

I'll have to think about this. I'll have to buy more tubing, and one more set of QDC's but that might be a small price to pay...

Consideration #2:

I currently have all D5's. I was going to continue using D5's, but then I thought of something. I'll be using 5 D5 pumps. (one per radiator, one for the workstation CPU block, one for the game machine CPU block and one for the game machine GPU block)

D5's dump their heat into the loop. D5's can consume up to 25W each. That's an additional 125w of heat being dumped into my loop.

DDC pumps do not dump their heat into the loop. They can even be equipped with small heat sinks on the bottom to help them dispatch their heat tot he surrounding air.

DDC pumps also do not achieve as high of a flow rate as D5's do (unless under very high head pressure).

I still want my flow to be exceptionally high across the game systems blocks (and I should be able to accomplish that with a one D5 pump to one block ratio) but the other pumps output are a lot less flow dependent. Maybe I would switch out the radiator pumps and the pump for the workstation to all be DDC pumps? That would certainly limit the system wide pump heat load...

DDC pumps seem to ave become less popular lately though. I've been searching, and there aren't too many options that are available and in stock from trusted brands.

I am going to have to give this some thought.

 

[Grimlakin]

Loving this thread BTW. Thank you for sharing your amazing projects!!!

 

[Space_Ranger]

Agreed... I love your detailed analysis... Following this one...

 

[Brian_B]

Kinda surprised you didn’t leave the 230 outlet at 230 and run the PSUs off the higher voltage for that smidge better efficiency

Looks nice though!

 

[Zarathustra]

Loving this thread BTW. Thank you for sharing your amazing projects!!!

Agreed... I love your detailed analysis... Following this one...

Thank you sirs!

 

[Zarathustra]

Kinda surprised you didn’t leave the 230 outlet at 230 and run the PSUs off the higher voltage for that smidge better efficiency

Looks nice though!

I don't think any of my current UPS:es I have in there ( 2x APC SMT1500RM2U and 1x Cyberpower PR1500LCD) accept 240v inputs, or even anything above 15A. They are all 120v 15A units.

Long term I have been toying with the idea of replacing the three UPS:es I have in there with one bigger unit. At the very least a 120v 30A unit, but maybe even a 240v unit of some sort.

I'll have to make sure the unit can either provide 120v power, or maybe make sure the power supplies for everything int he rack can actually handle 240v inputs first. And this is further complicated by the fact that U.S. 240v is 60hz, It falls within the acceptable range of 230v European power, but the AC over there is at 50hz. Some devices support both 50 and 60hz, and others do not. (Though in general most DC devices have AC adapters that support both. This is mostly a direct AC device issue)

...but UPS:es like that cost way more money than the cheap decommed ones I usually get on eBay, and I am already spending a crazy amount of money on this project. As with all things water-cooling, all the little bits and pieces add up. Even fans add up in quantity. Consider that a 200mm Noctua fan costs ~$33 on Amazon. I'm going to need 18 of them. With sales tax that's going to be over $600 on just fans

So, yeah. I am going to finish stage 1 (moving the workstation into the rack) Then take a little financial breather, then finish stage 2 (add the game system). Then after that I am going to reassess what is next.

I'm probably going to go from being water-cooling limited to being room temperature limited. The "Server room", being a former laundry room is not large. I may need to think about what to do about cooling that room.

High level I am considering maybe repurposing that MWBC I just installed to instead provide two 120v circuits at 30amps, using one of them for a small mini-split unit to cool the room and the other one to power a 120v 30A capable UPS, though I am a little unclear on if that use is per code. HVAC units - per their listing - are typically supposed to operate on a "dedicated circuit", but it is unclear to me if one half of an MWBC actually qualifies as a "dedicated circuit", or if they consider it a shared circuit because they share a neutral and shut off together.

I'm likely going to have to consult with an actual electrician on that one. Or just do it in spite of code, and just remove the unit before some day in the future selling the house, because who in their right mind is going to want an HVAC unit permanently installed in a closet that used to be a laundry room?

I also have other home improvement projects planned for for 2025 that are going to drain the finances.

So between all of that, going to a bigger single UPS is a "future consideration".

 

[Dan_D]

There's no kill like overkill.

 

[Zarathustra]

There's no kill like overkill.

That's my favorite kind of kill!

I honestly don't know what kind of overkill this will wind up being.

Yes, those Mora IV 600 radiators are huge, but they are also very low fin density, optimized for silence rather than performance (which is almost a little wasted in my application)

It will be interesting to see just how close to ambient water temperatures I can sustain with both systems at full load, and then with just one system at full load.

My thought process is that this is big and a little pricy up front, but once I am done this will be the "final stage" cooling solution. Provided **** doesn't break all I'll ever have to do is replace water blocks every now and then when I upgrade GPU's or CPU's (which will be made very easy due to the QDC's) and replace the fluid every couple of years.

That's one of my favorite parts of water cooling. The fact that it is so universal. You build it once, and then you can keep using it, almost indefinitely.

 

[Brian_B]

High level I am considering maybe repurposing that MWBC I just installed to instead provide two 120v circuits at 30amps, using one of them for a small mini-split unit to cool the room and the other one to power a 120v 30A capable UPS, though I am a little unclear on if that use is per code. HVAC units - per their listing - are typically supposed to operate on a "dedicated circuit", but it is unclear to me if one half of an MWBC actually qualifies as a "dedicated circuit", or if they consider it a shared circuit because they share a neutral and shut off together.

You can change the breaker out from a double-pole 220V to a two 120V breakers, then it would be dedicated per each side. The hot wire for one of them will be red versus black, but that's technically in code. The issue would be the shared neutral line... which is really only there to provide "incidental" single phase power for appliances that normally require full 240V phase to phase power. So you would need to derate that circuit considerably (you probably should evaluate that even with a double ganged 240V breaker the way you are running it, depending on the wire size that was run to that outlet), otherwise you have two hots each with 30A output, but a common neutral that's not rated for 60A if you maxed out both those circuits.

 

[Brian_B]

Also, wow, can't believe you are trusting all that money in hardware to Ebay refurbished UPSes.

I mean, glad it's working for you, but that's a lot of trust in some random dudes on Ebay. I think I'd rather plug it straight to grid with just a surge suppressor than trust a cheap or used UPS.

I mean, you really should be considering the UPS at least in the same light as your power supply - and you don't bat an eye at dropping good money for a dependable PSU. You should do the same for the UPS, if you elect to use one. Otherwise why not just get your PSUs decommed from Ebay as well and save a few more bucks there?

 

[Zarathustra]

You can change the breaker out from a double-pole 220V to a two 120V breakers, then it would be dedicated per each side. The hot wire for one of them will be red versus black, but that's technically in code.

I was under the impression they had to shut off together to be per code due to the shared neutral? Maybe I could solve that with some sort of bridging device on the handles if the breakers are right next to each other?

The issue would be the shared neutral line... which is really only there to provide "incidental" single phase power for appliances that normally require full 240V phase to phase power. So you would need to derate that circuit considerably (you probably should evaluate that even with a double ganged 240V breaker the way you are running it, depending on the wire size that was run to that outlet), otherwise you have two hots each with 30A output, but a common neutral that's not rated for 60A if you maxed out both those circuits.

I'm not sure what you mean by de-rate here. The wire from the main panel to that old dryer outlet is all orange sleeved #10/3 NM cable, so it should be good up to 30 amps (which is how it was wired to the old 30amp receptacle)

Having dropped it down to 20amps, I should have a 10 amp safety margin, which is way more than enough.

Many people think the neutral return in an MWBC is additive. In other words, it returns the sum of both sides of the MWBC over the neutral, and thus requires a larger neutral, but according to my reading this is not the case due to line1 and line2 being 180 degrees offset from each other and canceling each other out on the neutral return.

In other words, in a worst case, if one side of the MWBC is fully loaded, that neutral will see the full 20A return load.

Best case - however - if both sides are fully loaded (or equally loaded at a lower rate) the neutral will have zero return current on it, as the two opposing line1 and line2 currents will cancel each other out, and sort of act as returns for each other, just like if you were wiring a 240v circuit.

So really, all the neutral does in an MWBC is act as the return for the difference between the two sides.

In other words, if I am pulling 5 amps on the side of the MWBC that is connected to line1 as a hot, and 10 amps on the side that is connected to line2 as the hot, the shared neutral will only be seeing a 5 amp return current (the difference between the two)

I'm not an electrician though. But I did take some electrical engineering classes in college (mostly DC circuits though) and I have read the NEC cover to cover at this point "for fun".

At this point, the MWBC is "done", but I can always change it if I need to. It's not that hard.

Another option would be to just install a small 30amp subpanel using the existing #10 wiring, but I think that might just be over-complicating things.

 

[Zarathustra]

Also, wow, can't believe you are trusting all that money in hardware to Ebay refurbished UPSes.

I mean, glad it's working for you, but that's a lot of trust in some random dudes on Ebay. I think I'd rather plug it straight to grid with just a surge suppressor than trust a cheap or used UPS.

I mean, you really should be considering the UPS at least in the same light as your power supply - and you don't bat an eye at dropping good money for a dependable PSU. You should do the same for the UPS, if you elect to use one. Otherwise why not just get your PSUs decommed from Ebay as well and save a few more bucks there?

I have had nothing but good experiences when it comes to Enterprise pulls on eBay.

Consumer stuff is a different story all together, but I have been buying used enterprise stuff (from sellers that have a storefront that makes it obvious they specialize in decommed enterprise equipment sales) for 15 years now. I have a huge sample size of parts spanning the segment under my belt at this point. You name it. Rackmountable UPS:es, Server NIC's, Network Transceivers, SFP+ (and QSFP+) DAC cables, server power supplies, server rackmount cases, server motherboards, server CPU's, registered server ECC RAM, enterprise SSD's (mostly Optane), PCIe SAS HBA's and over these 15 years and countless purchases my experience has been absolutely perfect. Not a single problem.

(of course, now - having made this statement publicly, Murphy's law is bound to kick in, so I'll need to knock on wood. I knew the walnut desk would come in handy some day )

So I have found the used market of decommed enterprise parts to be very reliable over the years.

I no longer buy consumer stuff used though. At least not unless I get a very good vibe from the seller (again, like if they ahve a large store-front, and sell identical things in large batches, which makes them come across as professionals). I have received too much broken and abused consumer **** from online sales to ever really try it again. Even things (like GPU's) just in a carboard box, with no padding or anti-static protection, just bouncing around in there and showing up all beaten up.

Even the Enterprise stuff I guess is a slight risk, but I always test the crap out of the stuff in my testbench rig before using it in the rack to mitigate that risk. The savings are so extreme compared to new enterprise stuff that it tends to be worth it, and I will take used decommed enterprise stuff for most applications over new consumer stuff any day.

 

[Zarathustra]

I have been considering making some changes to the water loop layout as well.

Consideration #1:

Splitting up dual D5's to two separate D5's and having each power just one block of the game machine (one for CPU, one for GPU.)

My plan has originally been to dedicate a dual pump top of the two blocks in the game machine, but I have been thinking. Dual pumps (in series) DO scale, but they do not scale well. Generally if you want better flow, it is better to reduce head pressure. I bet I could get better flow over both the blocks, if I run them as separate sub-loops from the same reservoir, with just one pump each.

I'll have to think about this. I'll have to buy more tubing, and one more set of QDC's but that might be a small price to pay...

Consideration #2:

I currently have all D5's. I was going to continue using D5's, but then I thought of something. I'll be using 5 D5 pumps. (one per radiator, one for the workstation CPU block, one for the game machine CPU block and one for the game machine GPU block)

D5's dump their heat into the loop. D5's can consume up to 25W each. That's an additional 125w of heat being dumped into my loop.

DDC pumps do not dump their heat into the loop. They can even be equipped with small heat sinks on the bottom to help them dispatch their heat tot he surrounding air.

DDC pumps also do not achieve as high of a flow rate as D5's do (unless under very high head pressure).

I still want my flow to be exceptionally high across the game systems blocks (and I should be able to accomplish that with a one D5 pump to one block ratio) but the other pumps output are a lot less flow dependent. Maybe I would switch out the radiator pumps and the pump for the workstation to all be DDC pumps? That would certainly limit the system wide pump heat load...

DDC pumps seem to ave become less popular lately though. I've been searching, and there aren't too many options that are available and in stock from trusted brands.

I am going to have to give this some thought.

So, I am considering buying some Bykski DDC pumps.

As a reminder, I am looking for a few DDC pumps, because I plan on using 5 pumps, and D5 pumps each dump as much as 25w into the loop. That's an extra 125w the loop has to handle, which just doesn't seem efficient. The D5 pumps will get better flow, so I still plan to use them for the parts that benefit the most from high flow (game system CPU block, game system GPU block) but the remaining three pumps (workstation CPU block, and one each per radiator) could go to slower DDC pumps that dump their heat external to the loop without a problem.

Anyway, the above statement is a very weird one for me. Normally Bykski wouldn't even be one of the brands on my shopping list, as I have been very wary of Chinese brands in the past.

My biggest concern has been that anything Chinese that connects to the internet is a potential spying tool, but that obviously isnt the case with a DDC pump. My second concern is that while I know Chinese stuff has gotten better in the last 20 years, I still have that nagging voice in the back of my head that says "it is Chinese crap. Everyone knows they make garbage. What if it leaks or fails after the first couple of weeks?"
I guess in this case, the pumps will all be pretty far from the sensitive and expensive PC parts, so leaks - while annoying - would be less critical.

I also don't want to reward the Chinese copies of western designs. That is a big pet peeve of mine.

But looking at all of my usual sources (Amazon, Tinanrig, Modmymods, Performance-PC's, Newegg), I just cant find any other fully dressed (pwm controlled pump, pump top, case, pump bottom with heatsink) DDC pumps in stock, so I guess I don't have much of a choice.

I have found a couple of others without a heatsink on the bottom, but from my reading, you really want a heatsink on DDC pumps that see a lot of use, or their won't last long.



They have a kind of ugly (but practical) top that is designed to either accept a screw in cylinder reservoir, or be used standalone with a G1/4 inlet.

I don't really want to buy this pump, but I don't see many other options. DDC pumps seem to have fallen out of favor in the water cooling community in later years, with most either discontinued or just not available.

I also have next to zero DDC experience. Every water cooling system I've built in the last decade has used D5 pumps. Maybe I can get the pump motor and tops and bottoms separately and assemble myself?


I'd appreciate any thoughts or recommendations for alternates from anyone who has DDC pump experience.

 

[Zarathustra]

I was under the impression they had to shut off together to be per code due to the shared neutral? Maybe I could solve that with some sort of bridging device on the handles if the breakers are right next to each other?



I'm not sure what you mean by de-rate here. The wire from the main panel to that old dryer outlet is all orange sleeved #10/3 NM cable, so it should be good up to 30 amps (which is how it was wired to the old 30amp receptacle)

Having dropped it down to 20amps, I should have a 10 amp safety margin, which is way more than enough.

Many people think the neutral return in an MWBC is additive. In other words, it returns the sum of both sides of the MWBC over the neutral, and thus requires a larger neutral, but according to my reading this is not the case due to line1 and line2 being 180 degrees offset from each other and canceling each other out on the neutral return.

In other words, in a worst case, if one side of the MWBC is fully loaded, that neutral will see the full 20A return load.

Best case - however - if both sides are fully loaded (or equally loaded at a lower rate) the neutral will have zero return current on it, as the two opposing line1 and line2 currents will cancel each other out, and sort of act as returns for each other, just like if you were wiring a 240v circuit.

So really, all the neutral does in an MWBC is act as the return for the difference between the two sides.

In other words, if I am pulling 5 amps on the side of the MWBC that is connected to line1 as a hot, and 10 amps on the side that is connected to line2 as the hot, the shared neutral will only be seeing a 5 amp return current (the difference between the two)

I'm not an electrician though. But I did take some electrical engineering classes in college (mostly DC circuits though) and I have read the NEC cover to cover at this point "for fun".

At this point, the MWBC is "done", but I can always change it if I need to. It's not that hard.

Another option would be to just install a small 30amp subpanel using the existing #10 wiring, but I think that might just be over-complicating things.

@Brian_B You never know who you are talking to online, and what their expertise is.

If you are confident I have gotten the above wrong, I'd love to hear it. As mentioned, I am not a pro, and if I missed somethign despite all of my research on the topic, I'd love to be set straight before I burn my house down

I have just read up on this stuff on my own time. (and I do lurk in a handful of electricians subreddits just to absorb things).

 

[Zarathustra]

So, I am considering buying some Bykski DDC pumps.

As a reminder, I am looking for a few DDC pumps, because I plan on using 5 pumps, and D5 pumps each dump as much as 25w into the loop. That's an extra 125w the loop has to handle, which just doesn't seem efficient. The D5 pumps will get better flow, so I still plan to use them for the parts that benefit the most from high flow (game system CPU block, game system GPU block) but the remaining three pumps (workstation CPU block, and one each per radiator) could go to slower DDC pumps that dump their heat external to the loop without a problem.

Anyway, the above statement is a very weird one for me. Normally Bykski wouldn't even be one of the brands on my shopping list, as I have been very wary of Chinese brands in the past.

My biggest concern has been that anything Chinese that connects to the internet is a potential spying tool, but that obviously isnt the case with a DDC pump. My second concern is that while I know Chinese stuff has gotten better in the last 20 years, I still have that nagging voice in the back of my head that says "it is Chinese crap. Everyone knows they make garbage. What if it leaks or fails after the first couple of weeks?"
I guess in this case, the pumps will all be pretty far from the sensitive and expensive PC parts, so leaks - while annoying - would be less critical.

I also don't want to reward the Chinese copies of western designs. That is a big pet peeve of mine.

But looking at all of my usual sources (Amazon, Tinanrig, Modmymods, Performance-PC's, Newegg), I just cant find any other fully dressed (pwm controlled pump, pump top, case, pump bottom with heatsink) DDC pumps in stock, so I guess I don't have much of a choice.

I have found a couple of others without a heatsink on the bottom, but from my reading, you really want a heatsink on DDC pumps that see a lot of use, or their won't last long.

View attachment 3510 View attachment 3511 View attachment 3512

They have a kind of ugly (but practical) top that is designed to either accept a screw in cylinder reservoir, or be used standalone with a G1/4 inlet.

I don't really want to buy this pump, but I don't see many other options. DDC pumps seem to have fallen out of favor in the water cooling community in later years, with most either discontinued or just not available.

I also have next to zero DDC experience. Every water cooling system I've built in the last decade has used D5 pumps. Maybe I can get the pump motor and tops and bottoms separately and assemble myself?


I'd appreciate any thoughts or recommendations for alternates from anyone who has DDC pump experience.

User pendragon1 over at the hardforums suggested the FreezeMod DDC pump as an alternative.

I spent hours searching and this one never came up. I swear to god Amazon is manipulating search results to try to hide inexpensive products from me and only show me expensive ones.

I'm still not thrilled about having Chinese brands in my loop, for all the reasons listed above, but I guess this one is low risk as it isn't going in the case with my PC parts, and pendragon says he's had his on 24/7 for 4.5 years without any issues. I think I am just going to go for it.



Their claim of 800l/h (~3.5 GPM) seems questionable, considering every other DDC I've seen on the market claims at most 600 L/h max and many below that, but they wouldn't be the first seller of water cooling pumps to lie about the specs.

EK claimed 1000L/H on some of their DDC pumps, but at the same time they posted a head/flow chart on the same page showing that they maxed out at like 550 L/H



omeone tell me how "up to 1000 L/h" is possible with that head/flow chart.

The "up to" is doing a lot of work in that statement.
That would likely be one of the less shady things EK has done over the years though.


All of that said, even if the Freezemod pump only hits a max of ~480 L/h like the traditional 3.2 DDC pump does in EK's chart above, that is still ~2.1 GPM. Even with some loss due to head pressure, I should wind up getting WELL above the 1GPM "rule of thumb" level considering all they will be pushing through is a single (albeit huge) radiator each (2 of them) and one CPU block (one of them).

And DDC pumps tend to have higher head pressure than D5 pumps, so while they start at a lower flow rate, they lose less of it to resistance in the loop than a D5 does.

According to some testing I've seen, a DDC pump ought to lose me less than 10% flow rate with only one block in the flow path. That's still a worst case of ~1.89 GPM for the workstation block. It's probably about the same for the huge radiators. And that is way more than sufficient.

By contrast, a typical D5 (regular variety, not D5 Strong or 24v D5) tends to be speced at 1500L/h (~6.6 GPM) but drops off way quicker due to loop resistance due to the lower head pressure. Effective flow with only one block in the loop ought to be ~3.9 GPM (depending on the block). In a more typical full loop with two blocks (CPU and GPU) and two (normal sized) radiators my experience is that they tend to drop down to between 0.85 and 1.3 GPM depending on the loop particulars and how many restrictive components are in there.

I'm still going to keep the D5's for the game system subloops. In that application, the up to 25w each they dump into the loop is probably warranted. I want to maximize the flow across those blocks to minimize the temperature delta between the core and the coolant.

In my original design I had intended to use two D5's in serial for a single loop with both blocks in it, but I am considering actually splitting it up with one D5 per block instead, as generally reducing resistance tends to have much more effect than adding pressure by adding pumps in series.

This has multiple benefits. On the one hand, my theoretical flow rate goes up (to a potential 3.9GPM per block) which ought to keep the delta between the cores and coolant as low as possible. But this is probably overkill. I plan on testing speed settings and see at what flow rate I start seeing seriously diminishing returns on delta T and cap the max PWM speed there, as this will mean less power used by the pumps, less pressure on my loop joints, and less heat (from the D5 pumps) dumped into the loop.

(I am going to try to do some similar optimization on the DDC's pushing through the radiators and the workstation block as well)

Another benefit from using one pump per block is that I don't need to use loop temperature sensors and the Aquaero as a speed control device. I can just use the motherboard fan header (based on actual core temps) to control the pump speed for the CPU block, and I can usually tap into the PWM signal originally going to the GPU's fans to control the pump for the GPU block.

(this works well in large part since I have decoupled the radiator flow and the block flow, otherwise this would be suboptimal, as at low core temps the fluid in the loop would be allowed to slowly heat up, and then if there is a sudden load, there will be a temp spike, before the fluid can be cooled down by the radiators.)

This is what I did (but for fan speed instead of pump speed) back when I used two AIO's to cool my dual 980ti's in 2015 using Corsairs HG10 N980 adapter bracket.





At least on those cards (EVGA 980ti's) I found that while they used a mini-fan header, it was still the same PWM signal as for regular PC fans, and still 4 pins, and the same pin layout as the standard PC fan header.

So I just made my own wiring harnesses to adapt the PWM and RPM signals to instead control the 140mm fans blowing through the Corsair H90 (though I didn't pull any power from the GPU as I didn't know how much power that header could handle)

This resulted in the fan speed being fully controllable by the GPU's firmware, or even GPU tools like MSI Afterburner.

Hopefully this is still the case.

My recollection from that project (though it has been a hot minute) is that the hardest part was actually finding a mating connector that would plug into the mini-fan header on the GPU. (I didn't want to destroy the stock coolers by cutting off and using their plugs). If I recall I did some exhaustive searching and eventually only found them as part of some completely unrelated wire harness, and bought two of those and lopped off the connectors.

 

[Zarathustra]

User pendragon1 over at the hardforums suggested the FreezeMod DDC pump as an alternative.

I spent hours searching and this one never came up. I swear to god Amazon is manipulating search results to try to hide inexpensive products from me and only show me expensive ones.

I'm still not thrilled about having Chinese brands in my loop, for all the reasons listed above, but I guess this one is low risk as it isn't going in the case with my PC parts, and pendragon says he's had his on 24/7 for 4.5 years without any issues. I think I am just going to go for it.

View attachment 3513

Their claim of 800l/h (~3.5 GPM) seems questionable, considering every other DDC I've seen on the market claims at most 600 L/h max and many below that, but they wouldn't be the first seller of water cooling pumps to lie about the specs.

EK claimed 1000L/H on some of their DDC pumps, but at the same time they posted a head/flow chart on the same page showing that they maxed out at like 550 L/H

View attachment 3514

omeone tell me how "up to 1000 L/h" is possible with that head/flow chart.

The "up to" is doing a lot of work in that statement.
That would likely be one of the less shady things EK has done over the years though.


All of that said, even if the Freezemod pump only hits a max of ~480 L/h like the traditional 3.2 DDC pump does in EK's chart above, that is still ~2.1 GPM. Even with some loss due to head pressure, I should wind up getting WELL above the 1GPM "rule of thumb" level considering all they will be pushing through is a single (albeit huge) radiator each (2 of them) and one CPU block (one of them).

And DDC pumps tend to have higher head pressure than D5 pumps, so while they start at a lower flow rate, they lose less of it to resistance in the loop than a D5 does.

According to some testing I've seen, a DDC pump ought to lose me less than 10% flow rate with only one block in the flow path. That's still a worst case of ~1.89 GPM for the workstation block. It's probably about the same for the huge radiators. And that is way more than sufficient.

By contrast, a typical D5 (regular variety, not D5 Strong or 24v D5) tends to be speced at 1500L/h (~6.6 GPM) but drops off way quicker due to loop resistance due to the lower head pressure. Effective flow with only one block in the loop ought to be ~3.9 GPM (depending on the block). In a more typical full loop with two blocks (CPU and GPU) and two (normal sized) radiators my experience is that they tend to drop down to between 0.85 and 1.3 GPM depending on the loop particulars and how many restrictive components are in there.

I'm still going to keep the D5's for the game system subloops. In that application, the up to 25w each they dump into the loop is probably warranted. I want to maximize the flow across those blocks to minimize the temperature delta between the core and the coolant.

In my original design I had intended to use two D5's in serial for a single loop with both blocks in it, but I am considering actually splitting it up with one D5 per block instead, as generally reducing resistance tends to have much more effect than adding pressure by adding pumps in series.

This has multiple benefits. On the one hand, my theoretical flow rate goes up (to a potential 3.9GPM per block) which ought to keep the delta between the cores and coolant as low as possible. But this is probably overkill. I plan on testing speed settings and see at what flow rate I start seeing seriously diminishing returns on delta T and cap the max PWM speed there, as this will mean less power used by the pumps, less pressure on my loop joints, and less heat (from the D5 pumps) dumped into the loop.

(I am going to try to do some similar optimization on the DDC's pushing through the radiators and the workstation block as well)

Another benefit from using one pump per block is that I don't need to use loop temperature sensors and the Aquaero as a speed control device. I can just use the motherboard fan header (based on actual core temps) to control the pump speed for the CPU block, and I can usually tap into the PWM signal originally going to the GPU's fans to control the pump for the GPU block.

(this works well in large part since I have decoupled the radiator flow and the block flow, otherwise this would be suboptimal, as at low core temps the fluid in the loop would be allowed to slowly heat up, and then if there is a sudden load, there will be a temp spike, before the fluid can be cooled down by the radiators.)

This is what I did (but for fan speed instead of pump speed) back when I used two AIO's to cool my dual 980ti's in 2015 using Corsairs HG10 N980 adapter bracket.

View attachment 3515

View attachment 3516

At least on those cards (EVGA 980ti's) I found that while they used a mini-fan header, it was still the same PWM signal as for regular PC fans, and still 4 pins, and the same pin layout as the standard PC fan header.

So I just made my own wiring harnesses to adapt the PWM and RPM signals to instead control the 140mm fans blowing through the Corsair H90 (though I didn't pull any power from the GPU as I didn't know how much power that header could handle)

This resulted in the fan speed being fully controllable by the GPU's firmware, or even GPU tools like MSI Afterburner.

Hopefully this is still the case.

My recollection from that project (though it has been a hot minute) is that the hardest part was actually finding a mating connector that would plug into the mini-fan header on the GPU. (I didn't want to destroy the stock coolers by cutting off and using their plugs). If I recall I did some exhaustive searching and eventually only found them as part of some completely unrelated wire harness, and bought two of those and lopped off the connectors.

For what it is worth, I just reviewed the pictures I took back when I put the water block on my 4090.

It appears as if at least MSI still used the same mini-fan headers, at least as of the 40-series.



With any luck, the AIB 5090's will do the same.

Even if I see a connector like this - however - I'm going to have to measure carefully, as this is not a standard, and there is no guarantee they use them the same way.

As an example, the MSI 4090 had two of them, a white one and a black one. If memory serves (though I'd have to go dig out the cooler to confirm) the black one was for the RBG LED's.

 

[Brian_B]

Best case - however - if both sides are fully loaded (or equally loaded at a lower rate) the neutral will have zero return current on it, as the two opposing line1 and line2 currents will cancel each other out, and sort of act as returns for each other.

So really, all the neutral does in an MWBC is act as the return for the difference between the two sides.

It carries unbalanced load if you are going phase to phase … but you are going phase to neutral on both legs. The currents do not cancel — it’s like carrying a 50 lb barbell in each hand. The weights don’t cancel out to 0 lbs total just because you have them out of phase on each arm. If anything, it would m

@Brian_B You never know who you are talking to online, and what their expertise is.

If you are confident I have gotten the above wrong, I'd love to hear it. As mentioned, I am not a pro, and if I missed somethign despite all of my research on the topic, I'd love to be set straight before I burn my house down

I have just read up on this stuff on my own time. (and I do lurk in a handful of electricians subreddits just to absorb things).

Nope, I went back and did some more reading on it - you are spot on. I am used to dealing with 3P which is a bit different, this isn't that though. So you look good.

The ganged breaker to single breaker would work, however, I think as you suspect, it wouldn't be per code due to the single neutral line. (although you could drop a breaker on that line too, I wouldn't do that. Easier to just run a new line for a HVAC)

 

[Zarathustra]

It carries unbalanced load if you are going phase to phase … but you are going phase to neutral on both legs. The currents do not cancel — it’s like carrying a 50 lb barbell in each hand. The weights don’t cancel out to 0 lbs total just because you have them out of phase on each arm. If anything, it would m

Nope, I went back and did some more reading on it - you are spot on. I am used to dealing with 3P which is a bit different, this isn't that though. So you look good.

The ganged breaker to single breaker would work, however, I think as you suspect, it wouldn't be per code due to the single neutral line. (although you could drop a breaker on that line too, I wouldn't do that. Easier to just run a new line for a HVAC)

Thank you, I appreciate you checking!

I have never worked with 3-phase. I imagine that is more complicated due to the phases not being perfectly offset by 180 degrees.

 

[Brian_B]

Thank you, I appreciate you checking!

I have never worked with 3-phase. I imagine that is more complicated due to the phases not being perfectly offset by 180 degrees.

It's all PFM (pure ****ing magic)

 

[LeRoy_Blanchard]

I've been thinking about replacing the MO-RA3 420 I have for my build for the new big boy, but I don't think I'll need that kind of cooling. The 420 should work for my needs, more than enough. Its going to be cooling dual AMD 9654 EPYCs and a pair of either 3090s or 2 - 4 V100s depending on which direction I decide to go.