• Welcome to TechPowerUp Forums, Guest! Please check out our forum guidelines for info related to our community.

Watercooling options and advice

cdawall

where the hell are my stars
Joined
Jul 23, 2006
Messages
27,622 (5.94/day)
Likes
8,488
Location
Houston
System Name All the cores
Processor 2990WX
Motherboard Asrock X399M
Cooling CPU-XSPC RayStorm Neo, 2x240mm+360mm, D5PWM+140mL||GPU-2x360mm, 3xbyski, D4+D5+100mL
Memory 4x8GB G.Skill Trident Z 3200 CL16
Video Card(s) (2) EVGA SC BLACK 1080Ti's+EVGA 1080Ti FE
Storage 2x Samsung SM951 512GB, Samsung PM961 512GB
Display(s) Dell UP2414Q 3840X2160@60hz
Case Caselabs Mercury S5+pedestal
Audio Device(s) Fischer HA-02->Fischer FA-002W High edition/FA-003/Jubilate/FA-011 depending on my mood
Power Supply Seasonic Prime 1200w
Mouse Thermaltake Theron || Steam controller
Keyboard Thermaltake Poseidon ZX || Steam controller
Software W10P
#1
I have finally ordered the waterblocks for my GPU. Have three of the chinesium byski 1080Ti blocks heading this way and already got a pair of the barrow 3*120*60mm rads in as well as another pump res top.

It will be cooling 3 1080Ti's all overclocked all on stockish PCB's (2 EVGA SC blacks on titan XP PCB's and 1 FE), plan is to paint mod these so they can push higher clocks and cooling the beast that is my 2990WX. I am wanting this to be pretty quiet running using decent fans all around, but with the amount of radiator surface area I cannot imagine it will make much difference.

Current loop:

EK pump/res D5 PWM 150mL
XSPC Raystorm Neo WB
Swiftec MCR220-QP
Swiftec MCR220-QP-Stak
Swiftec MCR320-QP

Adding
EK pump/res D5 vario 100mL
3 Byski N-TITAN-PAS-X
2 Barrow Dabel-60A-360 (3*120mm*60mm thick)
HWlabs N160GTX-F2PB (2*80mm*54mm thick)

Now with this I could absolutely run the second batch as its own loop with zero effort or I could make one giant loop, I need to flush the old loop anyway, but what is the thoughts on the best performance, dual loops or single loop?

Couple of lazy pics of what we are dealing with, using the HDD mounts for the caselabs pedestal because that is all that was available. Rest of parts just set in the bottom loosely, for test fitting.





 

infrared

Moderator
Staff member
Joined
May 28, 2005
Messages
4,650 (0.92/day)
Likes
2,526
Location
Southampton, UK
System Name The Distractor
Processor Ryzen 7 1800X
Motherboard Crosshair VI Hero
Cooling 2x 240mm rads, Corsair ML fans, EKWB pump/res and GPU block. XSPC Raystorm pro CPU block
Memory G.Skill TridentZ DDR4-4266 16GB @ 3466MHz CL14 1T
Video Card(s) EVGA 1080Ti FE. WC'd & TDP limit increased to 360W.
Storage Samsung 960 Evo 500GB + Corsair 120GB ssd (linux) + WD Black 2TB storage drive.
Display(s) Asus ROG Swift PG278QR 27" 1440P 165hz Gsync
Case Phanteks Enthoo Pro M
Audio Device(s) Phillips Fidelio X2 headphones / basic Bose speakers
Power Supply EVGA Supernova 750W G3
Mouse Logitech G602
Keyboard Cherry MX Board 6.0 (mx red switches)
Software Win 10 & Linux Mint 18.3
Benchmark Scores https://hwbot.org/user/infrared
#2
I'd set it up with separate loops personally, cpu temps get pretty rubbish when multiple gpus are already dumping loads of heat into the loop. You end up cranking the fans up just to try to keep the cpu cool. If you separate the loops the cpu fans can run gently, and the gpus will be happy enough on warmer coolant so can run those fans pretty low too.

Cool build mate :D
 

cdawall

where the hell are my stars
Joined
Jul 23, 2006
Messages
27,622 (5.94/day)
Likes
8,488
Location
Houston
System Name All the cores
Processor 2990WX
Motherboard Asrock X399M
Cooling CPU-XSPC RayStorm Neo, 2x240mm+360mm, D5PWM+140mL||GPU-2x360mm, 3xbyski, D4+D5+100mL
Memory 4x8GB G.Skill Trident Z 3200 CL16
Video Card(s) (2) EVGA SC BLACK 1080Ti's+EVGA 1080Ti FE
Storage 2x Samsung SM951 512GB, Samsung PM961 512GB
Display(s) Dell UP2414Q 3840X2160@60hz
Case Caselabs Mercury S5+pedestal
Audio Device(s) Fischer HA-02->Fischer FA-002W High edition/FA-003/Jubilate/FA-011 depending on my mood
Power Supply Seasonic Prime 1200w
Mouse Thermaltake Theron || Steam controller
Keyboard Thermaltake Poseidon ZX || Steam controller
Software W10P
#3
I'd set it up with separate loops personally, cpu temps get pretty rubbish when multiple gpus are already dumping loads of heat into the loop. You end up cranking the fans up just to try to keep the cpu cool. If you separate the loops the cpu fans can run gently, and the gpus will be happy enough on warmer coolant so can run those fans pretty low too.

Cool build mate :D
Kind of what I was thinking it already does a great job keeping the cpu cool and I think once I get the case temps down with the gpus not putting 500w worth of hot air in the case it’ll be a lot happier.

One day I might even hard tube it at that point I think it can be a cool build lol.
 
Joined
Mar 18, 2015
Messages
1,748 (1.17/day)
Likes
1,001
Location
Long Island
#4
1. The reference PCBs will limit some of the gains you get from water cooling, saving you about 50 watts per card. You will also lose 0.5C or so on the GPU and about 3C on th VRMS with the Bykski block as compared to say a EK

2. Let's "do the math" ... the + side is that the EVGA cards will produce less heat . I can tell you from the get go, your radiators are by no means oversized for this application. I'm using 37% more for my 90 watt CPU and twin 225 watt GFX cards and I have half your load

1080 Ti's = 267 watts in peak gaming. The MSI AB Power Slider I believe offers a 20% boost so that's 320, let's use 300 watts each
CPU = 250 watts ... I have seen reviews overclocking your CPU saying the CPU package under stress testing was pulling 590 watts ... let's use 300 here too.

So essentially we have 300 watts x 4 components of 1200 watts (overclocked) ...even at stock settings, I don't think you have enough rad for the usual target Delta T of 10C. Given the 1200 watt PSU, I'm gonna do the calcs at stock everything

3 x 267 (GFX) + 250 (CPU) + 23 (pump) = 1074 say 1075 watts

You can download the spreadsheets for various radiator lines and follow the methods here. All data came from Martinsliquidlab test site. Before getting started, it must be recognized that a) all the heat is not removed by the radiators ... the radiator shrouds, tubing, block surfaces, back plates are all radiating heat into the case and b) not very component is at maximum load at the same time. When we factor that in, we find that our actual heat load averages about 60% being handled by the radiators.

https://www.overclock.net/forum/61-water-cooling/1457426-radiator-size-estimator.html

So 1050 watts x 60% = 645 watts. Im using the Alphacool Radiator test data downloaded from that link.

With a 360 + 240 60mm thick rads ....

1000 rpm fans will net you 151 watts on the 360 and 101 for the 240 = 252 watts
1250 rpm fans will net you 188 watts on the 360 and 125 for the 240 = 313 watts
1400 rpm fans will net you 210 watts on the 360 and 140 for the 240 = 350 watts
1800 rpm fans will net you 269 watts on the 360 and 179 for the 240 = 448 watts ... putting fans in push pull gets ya 542
2200 rpm fans will net you 326 watts on the 360 and 217 for the 240 = 543 watts ... putting fans in push pull gets ya 686

Using 140mm fans would get you about 36% more cooling

Splitting the difference, 2000 rpm should get ya 614 watts of cooling at 2.000 rpm ... and should deliver a Delta T (ambient => coolant) of 10.5C ... but again, overclocking will push this higher. I would strongly suggest you get a "kil-o-watt" meter and see what this system is pulling.

http://www.p3international.com/products/P4400.html

Again, the sizing estimator is based upon anticipated heat loads at near maximum GPU and CPU loads ... not theoretical loads but things you can actually produce on the desktop. We use Furmark / RoG Real Bench to set up pump and fan curves and those are set up to provide adequate cooling at lowest possible fan speeds. We design our cooling systems to be close to 1250 rpm under such conditions ... we never hit 850 rpm at which point you can barely tell the system is running....in typical gaming, I won't see > 650 rpm at which point the fans are dead silent.

Again, before setting up your system, I'd give it a run with the system plugged into the meter.

If that's the Seasonic Gold Prime your efficiencies are 87% at 20% load, 90% at 50% load and 87% at 100% load. So at 1050 watt output , you'd be about 89% efficiency ... that would show as 1179 watts at the wall.

For a quite system, you'd want to use no more than 1250 rpm fans. In the box I am typing from, the calculated wattage is 796, so I designed around 480 watt load on the rads.... wound up with 508 (10) 140mm fans w/ max speed range of 325 - 1240 or so.

If this was a 2 -way SLI system, the loop strategy would be obvious. GPU blocks are massive and therefore are able to provide much better heat removal. With the CPU block being so much smaller, it benefits from higher flows in a way that GPUs don't require. So with twin GPUs, the proven strategy is single loop ... but split the flow before the GPUs such that they only receive half the flow. This system is piped that way ... with fan speeds at 100% I see GPU temps of 39C under Furmark .... at 800ish rpm I see 42C. The CPU sees core temps of 69 - 76C. Neither changes all that much with different pump speeds. When gaming I tyically see about 0.50 - 0.75 gpm going thru the GPU water blocks. As you can see here... from about 0.75 gpm and up, cooling does not increase all that much. That leaves me with 1.00 - 1.50 gpm ging thru my CPU block

https://www.xtremerigs.net/2015/04/07/gtx-980-water-block-round-up/3/

I don't know much about the block you have on and what the flow / pressure loss curve looks like ... so without that info hard to judge. Some blocks are very restrictive

https://www.xtremerigs.net/2015/04/07/gtx-980-water-block-round-up/2/

Now looking at the CPU, we see that pressure can get to be an issue after 1 gpm on threadripper for some blocks so what block you have matters.

https://www.xtremerigs.net/2018/08/29/amd-threadripper-cpu-block-review-round-up/9/

With the EK block, or even the XSPC you could do 2 GPM thru the block w/o concern giving the cards 0.66 gpm which will suffice. Remember putting 2 cards in parallel reduces pressure loss by 3.6 .... 3 blocks by 7.6. So if you go this route ... look at the pressure loss of CPU block at any given flow.... then look at loss thru the 3 water blocks at 1/3 that flow to see if pump can handle. I haven't run all the numbers o checked all the graphs but for this build I'd likely use

a) Swiftech 35x2 pump for GFX cards w/ cards in parallel
b) Swiftech MCP-655 D5 pump on CPU loop

https://martinsliquidlab.wordpress.com/2012/01/29/swiftech-mcp35x2-pump/6/
https://martinsliquidlab.wordpress.com/2012/12/25/swiftech-mcp-655-pwm-drive-pump-review/4/

Size the rads using 3 x 300 watts for overclocked x 60% = 540 watts
Size the CPUs using 1 x 300* watts for overclocked x 80% = 240 watts

or whatever ya think your OC'd CPU wattage is. If using a Mono block (recommended) , use entire CPU package.
 

cdawall

where the hell are my stars
Joined
Jul 23, 2006
Messages
27,622 (5.94/day)
Likes
8,488
Location
Houston
System Name All the cores
Processor 2990WX
Motherboard Asrock X399M
Cooling CPU-XSPC RayStorm Neo, 2x240mm+360mm, D5PWM+140mL||GPU-2x360mm, 3xbyski, D4+D5+100mL
Memory 4x8GB G.Skill Trident Z 3200 CL16
Video Card(s) (2) EVGA SC BLACK 1080Ti's+EVGA 1080Ti FE
Storage 2x Samsung SM951 512GB, Samsung PM961 512GB
Display(s) Dell UP2414Q 3840X2160@60hz
Case Caselabs Mercury S5+pedestal
Audio Device(s) Fischer HA-02->Fischer FA-002W High edition/FA-003/Jubilate/FA-011 depending on my mood
Power Supply Seasonic Prime 1200w
Mouse Thermaltake Theron || Steam controller
Keyboard Thermaltake Poseidon ZX || Steam controller
Software W10P
#5
You assume I haven’t checked power consumption. CPU at full load on its own pulls over 860w at the wall. Power supply is a prime platinum which puts me around 90% efficiency at that load take away a bit for idle parts and fans and the CPU pulls around 675-700w at my clocks.

GPUs will be modded to allow more power consumption. That being said with a three way sli setup they aren’t going to pull more than 250w or so just because sli life.

Normal gaming load with a pair of cards and current cpu clocks is around 800-900w depending on game with the additional card it will be 1050-1100w (at the wall) so heat output will be around 950-1000w shooting numbers loosely.

I could easily cool this with a single 480mm radiator, assuming I didn’t load the cpu at the same time. The current setup even using just the single d5 pwm would be plenty of flow. The 2*240 and 1*360 radiator setup would easily suffice for cooling power.

I am adding an addition pair of 360mm rads and a 160mm rad plus an additional pump. The question for me was do I keep it as a single loop or split them out. Sounds like I should split them out. That will give me around 700w cooled by the 2*240 and 360mm rads which should be able to do at nearly passive fan speeds. The gpu loop would be around 750w cooled by 2*360mm and a 160mm rad. This again should allow super low fan speeds even under load.

It does have me curious if I could add a silent switch and run it passive at times.
 
Joined
Jul 5, 2013
Messages
5,495 (2.60/day)
Likes
3,607
Location
USA
#6
I'd set it up with separate loops personally
Agreed. Use a loop for your CPU and a second loop for GPU's otherwise your CPU/GPU temps could get unpleasant under certain types of loads. I'd also get bigger fluid tanks
 

cdawall

where the hell are my stars
Joined
Jul 23, 2006
Messages
27,622 (5.94/day)
Likes
8,488
Location
Houston
System Name All the cores
Processor 2990WX
Motherboard Asrock X399M
Cooling CPU-XSPC RayStorm Neo, 2x240mm+360mm, D5PWM+140mL||GPU-2x360mm, 3xbyski, D4+D5+100mL
Memory 4x8GB G.Skill Trident Z 3200 CL16
Video Card(s) (2) EVGA SC BLACK 1080Ti's+EVGA 1080Ti FE
Storage 2x Samsung SM951 512GB, Samsung PM961 512GB
Display(s) Dell UP2414Q 3840X2160@60hz
Case Caselabs Mercury S5+pedestal
Audio Device(s) Fischer HA-02->Fischer FA-002W High edition/FA-003/Jubilate/FA-011 depending on my mood
Power Supply Seasonic Prime 1200w
Mouse Thermaltake Theron || Steam controller
Keyboard Thermaltake Poseidon ZX || Steam controller
Software W10P
#7
Agreed. Use a loop for your CPU and a second loop for GPU's otherwise your CPU/GPU temps could get unpleasant under certain types of loads. I'd also get bigger fluid tanks
I thought about doing the CPU loop with a bigger res purely because it would fit, but my understanding is it makes zero real world difference once the loop hits equilibrium temp. There is not room in the bottom to do bigger than the 100mL tank anyway.
 
Joined
Jul 5, 2013
Messages
5,495 (2.60/day)
Likes
3,607
Location
USA
#8
I thought about doing the CPU loop with a bigger res purely because it would fit, but my understanding is it makes zero real world difference once the loop hits equilibrium temp.
For a CPU, generally yes. I was referring more to your GPU loop. With three of them and being OC'd, you'll need beefy cooling and a big tank.
 

cdawall

where the hell are my stars
Joined
Jul 23, 2006
Messages
27,622 (5.94/day)
Likes
8,488
Location
Houston
System Name All the cores
Processor 2990WX
Motherboard Asrock X399M
Cooling CPU-XSPC RayStorm Neo, 2x240mm+360mm, D5PWM+140mL||GPU-2x360mm, 3xbyski, D4+D5+100mL
Memory 4x8GB G.Skill Trident Z 3200 CL16
Video Card(s) (2) EVGA SC BLACK 1080Ti's+EVGA 1080Ti FE
Storage 2x Samsung SM951 512GB, Samsung PM961 512GB
Display(s) Dell UP2414Q 3840X2160@60hz
Case Caselabs Mercury S5+pedestal
Audio Device(s) Fischer HA-02->Fischer FA-002W High edition/FA-003/Jubilate/FA-011 depending on my mood
Power Supply Seasonic Prime 1200w
Mouse Thermaltake Theron || Steam controller
Keyboard Thermaltake Poseidon ZX || Steam controller
Software W10P
#9
For a CPU, generally yes. I was referring more to your GPU loop. With three of them and being OC'd, you'll need beefy cooling and a big tank.
Lol well a pair of 360mm and a 160mm should comfortably handle that.
 
Joined
Mar 18, 2015
Messages
1,748 (1.17/day)
Likes
1,001
Location
Long Island
#11
You assume I haven’t checked power consumption. CPU at full load on its own pulls over 860w at the wall. Power supply is a prime platinum which puts me around 90% efficiency at that load take away a bit for idle parts and fans and the CPU pulls around 675-700w at my clocks.

GPUs will be modded to allow more power consumption. That being said with a three way sli setup they aren’t going to pull more than 250w or so just because sli life.

Normal gaming load with a pair of cards and current cpu clocks is around 800-900w depending on game with the additional card it will be 1050-1100w (at the wall) so heat output will be around 950-1000w shooting numbers loosely.

I could easily cool this with a single 480mm radiator, assuming I didn’t load the cpu at the same time. The current setup even using just the single d5 pwm would be plenty of flow. The 2*240 and 1*360 radiator setup would easily suffice for cooling power.

I am adding an addition pair of 360mm rads and a 160mm rad plus an additional pump. The question for me was do I keep it as a single loop or split them out. Sounds like I should split them out. That will give me around 700w cooled by the 2*240 and 360mm rads which should be able to do at nearly passive fan speeds. The gpu loop would be around 750w cooled by 2*360mm and a 160mm rad. This again should allow super low fan speeds even under load.

It does have me curious if I could add a silent switch and run it passive at times.
I didn't assume anything .... I had trouble understanding what was staying and what was going ... did best i could with what i could inerpret erring on the conservative when not clear.

I have SLI on 2 cards and they are both pulling 20% over TDP

Can only work with the information provided, as it didn't say whether it was gold or platinum so took worse case. 860 watts at the wall CPU only would be 775 output The folks at hothwardware were not able to get more than 590 in their test, and to see that 860 at the wall... then we have also got storage, the pump, RAM, rest of the CPU package \ and whatever else in there. I'd be surprised to see what you are expecting from the CPU while gaming ... as won't see any load past 4th core. Did a TR build for a buddy he wanted it in a single box but I couldn't accommodate.... he has a 2990WX which he now uses as his game server. He tried to have it be dual usage as his gaming and server box but he found gaming performance less than satisfactory.

What you have and what you were adding was a bit hard to follow so if I understand correctly..... we'll do the math again

Swiftec MCR220-QP (2 x 120mm 34mm thick)
Swiftec MCR220-QP-Stak (2 x 120mm 34mm thick)
Swiftec MCR320-QP (3 x 120mm 34mm thick)
Barrow Dabel-60A-360 (3 x 120mm 34mm thick)
Barrow Dabel-60A-360 (3 x 120mm 34mm thick)
HWlabs N160GTX-F2PB (2*80mm*54mm thick)

The 2 x 80mm wont deliver very much ... < 50% of the 2 x 120s

Frankly I am surprised you can 13 x 120mm of rad in that mATX case.... It's a huge mATX but still a bit crowded.

Lets give the skinny 34mm the benefit of the doubt and treat them as if they were 45 mm ... near passive is 350 rpm, let's use 1250 rpm. Each 120mm @ 45mm is going to give you 61 watts of cooling at 1250 rpm, and you have 7 x 120 ... so 1,000 rpm gets you 7 x 61 watts for 427 watts of cooling. Each 60mm of 120 will give 63 watts of cooling so 6 of those will get you 378 watts. Being generous on the 80s, lets call those 30 watts per and assign 600 watts

427 + 378 + 60 = 865 watts of cooling at 1250 rpm

860 at the wall is @ 775 output

so 775 (CPU) + 3 x 250 (GPUs) + 2 x 24 watt pumps = 1573 of theoretical max power draw for water cooling ... if as you said, that 775 is just CPU, than have to add in 40 for MoBo, 10 for RAM, 10 for storage.... 15 rad fans (20 watts) , the USB stuff and max possible load would b around 1675 .... I certainly would not run Furmark and Blender at the same time on this build

No that case is big but still a bit crowded even is it is mATX so I dont know that 60% heat load is enough to account for the fact that not everything is a peak and heat radiation from hot components... but less use it anyway.

1575 x 60% 965 watts > 865 so rpm would need to be > 1250. At 1400 rpm, you should get to about 930 watts of cooling . But again... the assumption here is your 860 watts at the wall is with CPU's 32 cores under a stress test ... when gaming, if not going anything else in background, I would expect no more than 250 - 300 watts.

So yes, if you want to have your 3 GPUs and 32 cores all running full tilt, you'd need to use 1350 - 1500 rpm fans. I know a few folks with TR builds who have run games on it but just for shitz and giggles. I'd be surprised to see it break 325 watts at stock playing a game and not much more with a heavy OC. If ya are running a workstation app w/o CUDA and the GPUs are relatively idle, you should be fine. If you gaming w/ cards going full tilt and CPU yawning with just a few threads in play you should be fine. In both cases I'd expect you can keep fan rpms down at around 550 - 850.

If you go single loop, I'd use a 35x2 pump....even that is tight

Remember that fittings add a lot of resistance....

90 Elbow = 0.64
45 Elbow = 0.32
Pass thru 0.40 - 0.75.... Now add the loss thru each block, each radiator, pump ,reservoir and it gets up there
CPU block 1.8 psi (0.5 to 3.8)
GPU Block 3.0 psi (1.2 - 5.0) )

A D5 pump produces about 4.5 psi at 100% speed @ 1.0 gpm ... so no ... I don't see a D-5 pump pushing thru 7 or 8 rads, lotsa fittings , a CPU block and 3 GPU blocks. A 35x2 can do about 11.5 psi but that is a bit scary. Even with 2 loops,

At 1.25 gpm ...(links in previous post)

CPU Block = 0.9 psi
GPU Block = 1.0 psi per block + 3.0 psi
Radiator = 0.6 psi

Loop 1 = CPU + 3 rads + rad fittings = 1.8 + 1.8 + 2.4 for rad fittings = 5.0
Loop 2 = 3 GPU + 4 rads + rad fittings = 3.0 + 2.4 + 3.2 = 8.6 psi

All in one loop = 9.0 psi and that's with no fittings. Each set (2) of rad fittings adds at least 0.80
 
Last edited:

cdawall

where the hell are my stars
Joined
Jul 23, 2006
Messages
27,622 (5.94/day)
Likes
8,488
Location
Houston
System Name All the cores
Processor 2990WX
Motherboard Asrock X399M
Cooling CPU-XSPC RayStorm Neo, 2x240mm+360mm, D5PWM+140mL||GPU-2x360mm, 3xbyski, D4+D5+100mL
Memory 4x8GB G.Skill Trident Z 3200 CL16
Video Card(s) (2) EVGA SC BLACK 1080Ti's+EVGA 1080Ti FE
Storage 2x Samsung SM951 512GB, Samsung PM961 512GB
Display(s) Dell UP2414Q 3840X2160@60hz
Case Caselabs Mercury S5+pedestal
Audio Device(s) Fischer HA-02->Fischer FA-002W High edition/FA-003/Jubilate/FA-011 depending on my mood
Power Supply Seasonic Prime 1200w
Mouse Thermaltake Theron || Steam controller
Keyboard Thermaltake Poseidon ZX || Steam controller
Software W10P
#12
I didn't assume anything .... I had trouble understanding what was staying and what was going ... did best i could with what i could inerpret erring on the conservative when not clear.

I have SLI on 2 cards and they are both pulling 20% over TDP
Tons of testing has been done with three cards showing the will not typically pull over TDP even with clocks pushed up. It is much more difficult to get three cards to play nice with clockspeed and GPU Boost than it is with two.

Can only work with the information provided, as it didn't say whether it was gold or platinum so took worse case. 860 watts at the wall CPU only would be 775 output The folks at hothwardware were not able to get more than 590 in their test, and to see that 860 at the wall... then we have also got storage, the pump, RAM, rest of the CPU package \ and whatever else in there. I'd be surprised to see what you are expecting from the CPU while gaming ... as won't see any load past 4th core. Did a TR build for a buddy he wanted it in a single box but I couldn't accommodate.... he has a 2990WX which he now uses as his game server. He tried to have it be dual usage as his gaming and server box but he found gaming performance less than satisfactory.
Picture show the PSU and clearly says platinum, not that the less than ~2% difference makes much of any difference for the conversation.

As for the hothardware review, they overclocked using an AIO and hit much lower clockspeeds than even my everyday settings. I can and do pull substantially more power than they do.

As for gaming performance, don't care. Built the rig because I could, overclock it because I can and quite honestly it should perform substantially worse than my 5960x@4.8ghz did. In the real world I see next to no performance difference in any of the games I play. I do however occasionally have to reboot with cores shut off because applications will not load with that many threads available (Carmageddon being an example). Mine is not low clocked in games all 32 cores hum along at 4.2ghz which means it runs just like a set of 4 2700x's with an OC.

What you have and what you were adding was a bit hard to follow so if I understand correctly..... we'll do the math again

Swiftec MCR220-QP (2 x 120mm 34mm thick)
Swiftec MCR220-QP-Stak (2 x 120mm 34mm thick)
Swiftec MCR320-QP (3 x 120mm 34mm thick)
Barrow Dabel-60A-360 (3 x 120mm 34mm thick)
Barrow Dabel-60A-360 (3 x 120mm 34mm thick)
HWlabs N160GTX-F2PB (2*80mm*54mm thick)

The 2 x 80mm wont deliver very much ... < 50% of the 2 x 120s

Frankly I am surprised you can 13 x 120mm of rad in that mATX case.... It's a huge mATX but still a bit crowded.
2 Barrow Dabel-60A-360 (3*120mm*60mm thick)
Two are 60mm rads...with the same 12FPI surface area to the MCR's...The MCR's are tested as performing similar to the HWlabs 54mm thick radiators when the proper fans are used. I will be curious to see how the performance of them is with a similar design to the MCR's performance should be quite good with the 1500RPM


Lets give the skinny 34mm the benefit of the doubt and treat them as if they were 45 mm ... near passive is 350 rpm, let's use 1250 rpm. Each 120mm @ 45mm is going to give you 61 watts of cooling at 1250 rpm, and you have 7 x 120 ... so 1,000 rpm gets you 7 x 61 watts for 427 watts of cooling. Each 60mm of 120 will give 63 watts of cooling so 6 of those will get you 378 watts. Being generous on the 80s, lets call those 30 watts per and assign 600 watts

427 + 378 + 60 = 865 watts of cooling at 1250 rpm

860 at the wall is @ 775 output

so 775 (CPU) + 3 x 250 (GPUs) + 2 x 24 watt pumps = 1573 of theoretical max power draw for water cooling ... if as you said, that 775 is just CPU, than have to add in 40 for MoBo, 10 for RAM, 10 for storage.... 15 rad fans (20 watts) , the USB stuff and max possible load would b around 1675 .... I certainly would not run Furmark and Blender at the same time on this build

No that case is big but still a bit crowded even is it is mATX so I dont know that 60% heat load is enough to account for the fact that not everything is a peak and heat radiation from hot components... but less use it anyway.

1575 x 60% 965 watts > 865 so rpm would need to be > 1250. At 1400 rpm, you should get to about 930 watts of cooling . But again... the assumption here is your 860 watts at the wall is with CPU's 32 cores under a stress test ... when gaming, if not going anything else in background, I would expect no more than 250 - 300 watts.

So yes, if you want to have your 3 GPUs and 32 cores all running full tilt, you'd need to use 1350 - 1500 rpm fans. I know a few folks with TR builds who have run games on it but just for shitz and giggles. I'd be surprised to see it break 325 watts at stock playing a game and not much more with a heavy OC. If ya are running a workstation app w/o CUDA and the GPUs are relatively idle, you should be fine. If you gaming w/ cards going full tilt and CPU yawning with just a few threads in play you should be fine. In both cases I'd expect you can keep fan rpms down at around 550 - 850.

If you go single loop, I'd use a 35x2 pump....even that is tight

Remember that fittings add a lot of resistance....

90 Elbow = 0.64
45 Elbow = 0.32
Pass thru 0.40 - 0.75.... Now add the loss thru each block, each radiator, pump ,reservoir and it gets up there
CPU block 1.8 psi (0.5 to 3.8)
GPU Block 3.0 psi (1.2 - 5.0) )

A D5 pump produces about 4.5 psi at 100% speed @ 1.0 gpm ... so no ... I don't see a D-5 pump pushing thru 7 or 8 rads, lotsa fittings , a CPU block and 3 GPU blocks. A 35x2 can do about 11.5 psi but that is a bit scary. Even with 2 loops,

At 1.25 gpm ...(links in previous post)

CPU Block = 0.9 psi
GPU Block = 1.0 psi per block + 3.0 psi
Radiator = 0.6 psi

Loop 1 = CPU + 3 rads + rad fittings = 1.8 + 1.8 + 2.4 for rad fittings = 5.0
Loop 2 = 3 GPU + 4 rads + rad fittings = 3.0 + 2.4 + 3.2 = 8.6 psi

All in one loop = 9.0 psi and that's with no fittings. Each set (2) of rad fittings adds at least 0.80
For all of this I will put it this way I currently have and have had for months run the pair of MCR220's and MCR320 with the single laing D5 PWM. This has been able to maintain good temps with out pushing fan speeds up. Keeping all of the non-static pressure fans running around 1000RPM and if I am looking to push over the 4.2ghz mark I have a pair of 32mm fans in the radiator sandwhich I can push up to >3000RPM. Those normally are shut off. Using your math this shouldn't even function as a computer, yet it has pushed 4k rendering loads so all 32 cores loaded and where I got my 860w wall load from, for 48 hour hauls when I was converting my media library over. All of that being said in games the CPU does next to nothing. I see gaming loads with the pair of 1080Ti's actually pulling less than the CPU does by itself under load. Typically gaming loads run around 775-800w at the wall. Adding another should put me around 1100w under load. I don't see it increasing CPU load any so I am not worried on that batch.

The loop will either be setup with a D5 vario and D5 PWM in series or as independent loops with a D5 in each loop. I already have these pumps and actually have a spare D4 I was contemplating adding into the GPU loop, just to be sure flow rate was enough. The MATX case is larger than most EATX cases and can easily accommodate the additional pump. I have zero HDD racks and use all nvme storage so there is not a loss of space to that. It has gobs of room for this loop...This honestly is probably the most spacious build I have done. The cool little pedestal could probably have handled a third 360mm 60mm radiator if I wanted to do that over running the extra pumps down there. It would have also been another rad sandwich which would have required more thought on my part to make sure it was performing as well as I would want.
 

cdawall

where the hell are my stars
Joined
Jul 23, 2006
Messages
27,622 (5.94/day)
Likes
8,488
Location
Houston
System Name All the cores
Processor 2990WX
Motherboard Asrock X399M
Cooling CPU-XSPC RayStorm Neo, 2x240mm+360mm, D5PWM+140mL||GPU-2x360mm, 3xbyski, D4+D5+100mL
Memory 4x8GB G.Skill Trident Z 3200 CL16
Video Card(s) (2) EVGA SC BLACK 1080Ti's+EVGA 1080Ti FE
Storage 2x Samsung SM951 512GB, Samsung PM961 512GB
Display(s) Dell UP2414Q 3840X2160@60hz
Case Caselabs Mercury S5+pedestal
Audio Device(s) Fischer HA-02->Fischer FA-002W High edition/FA-003/Jubilate/FA-011 depending on my mood
Power Supply Seasonic Prime 1200w
Mouse Thermaltake Theron || Steam controller
Keyboard Thermaltake Poseidon ZX || Steam controller
Software W10P
#13
Slight update, got the CPU loop cleaned and rebuilt. Went ahead and swapped to all matching clear tubing with mayhem's red fluid. Cleaned with lemon juice to not damage anything nickel plated in the loop. Couple of flushes and some nasty black junk out of the loop and its all refilled and happy. Corrected some fan arrangement stuff while I had everything torn down. The front rad sandwich is now in a full push/pull/pull arrangement. All things said and done saw a couple C's drop in temp. Also got the lower pedestal prepped. Mounted the D5 vario and went ahead and deleted the 2*80mm rad in favor of a laing D4 pump.

Will post pics of the completed project when I get the water blocks for the GPU's in. So far happy with the rebuild fixed a few issues that drove me nuts with the old setup.
 
Top