My System - An AIO Radiator Position and Fan Placement Analysis

[JrClocker]

I am the ultimate tweaker...and to that end, I am searching for the best cooling setup for my system...without going to a custom water loop.

My rig is in my signature, highlights:
- 5820 K running at 4.3 GHz
- Corsair H110i GTX AIO for CPU
- 2 GTX 970 running in SLI

The aim for my system is to have quiet. I don't want to have a high performing system at the cost of lots of fan noise. I read a lot of the forums about the best way to setup a water cooling system, but I really couldn't find anything about optimizing things around an AIO. When you add in the heat generated by graphics cards, you start to get an interesting problem that needs to be solved.

I devised a set of experiments to keep the CPU power draw constant, and vary the GPU power to see what effect the GPU power had on overall cooling system efficiency. I kept the primary "guts" of the system the same, and setup different configurations of these within the case.

The result for my system:

1. Radiator as exhaust for hot case air is the worst setup

2. The Noctua NF-A14 iPPC-2000 does not move as much air as the Corsair SPL140L (but it's quieter)


3. A fan "pushing" air through the radiator is superior to a fan "pulling" air through the radiator
--- This is probably due to the "pushing" of air raising the pressure between the fan and the end of the radiator. Higher pressure means more air molecules to grab and move heat.
--- A pull setup only will create a slight vacuum between the fan and the end of the radiator. Lower pressure means less air molecules to grab and move heat.

4. A push/pull configuration is superior to a fan pushing air through the radiator

The change in coolant and CPU temperature is quite astonishing...more than I thought it would be. GPU temperatures do increase some, but not much.


The experimental runs were setup as the following:

--------
Condition #1:
- SETI@Home at 90% of CPU, turn off GPU
- CPU Power at 153 W, GPU power at 95 W (idle)

Condition #2:
- SETI@Home at 90% of CPU, both GPUs running SETI@Home
- CPU Power at 155 W, GPU power at 171 W

Condition #3:
- SETI@Home at 60% of CPU, turn off GPU, Heaven 4.0 running continuously
- CPU Power 152 W, GPU power 290 W

Each run was 30 minutes long.

Before the start of each run, the coolant temperature was brought to 36.2 ± 0.2 °C.

Data was logged with the Open Hardware Monitor and Corsair Link software.

--------

The cooling system configurations where as follows:

Setup #1 (Radiator Exhaust, Corsair Fan Push):
- (3) 120 mm Intake fans in front of case
- (1) 120 mm Exhaust fan in back of case
- AIO radiator on top of case
- Fans pushing hot case air through the radiator to exhaust to outside
- Stock Corsair Fans (SPL140L)

Setup #2 (Radiator Intake, Corsair Fan Pull):
- AIO radiator in front of case
- Fans pulling outside air through radiator
- (2) 120 mm Exhaust fans on top of case
- (1) 120 mm Exhaust fan in back of case
- Stock Corsair Fans (SPL140L) on radiator

Setup #3 (Radiator Intake, Noctua Fan Pull):
- AIO radiator in front of case
- Fans pulling outside air through radiator
- (2) 120 mm Exhaust fans on top of case
- (1) 120 mm Exhaust fan in back of case
- Noctua NF-A14 iPPC-2000 Fans on radiator

Setup #4 (Radiator Intake, Noctua Fan Push):
- AIO radiator in front of case
- Fans pushing outside air through radiator
- (2) 120 mm Exhaust fans on top of case
- (1) 120 mm Exhaust fan in back of case
- Noctua NF-A14 iPPC-2000 Fans on radiator

Setup #5 (Radiator Intake, Noctua Fan Push, Corsair Fan Pull)
- AIO radiator in front of case
- Push/Pull fan configuration, Noctua Push, Corsair Pull
- (2) 120 mm Exhaust fans on top of case
- (1) 120 mm Exhaust fan in back of case
- Noctua NF-A14 iPPC-2000 Fans on radiator
- Stock Corsair Fans (SPL140L) on radiator

--------

Here are the results for max temperatures:

Max AIO Coolant Temperature:




Max CPU Temperature:




Max GPU Temperature:



--------

Next steps:

1. I just ordered 2 more Noctua NF-A14 iPPC-2000 to replace the Corsair SPL140L pull fans on the radiator. Will redo results
2. Replace the 2 GTX 970 graphics cards with 2 GTX 980i Hybrids (i.e. get as much of the graphics heat out of the case as possible)


Thoughts and feedback please (I have lots more pretty graphs if you want to see them!)

 

[Mandrake4565]

Interesting read Jr, I briefly looked through it but am too tired right now to really look into the details. It is pretty interesting how much difference there is in temps with the rad as intake vs exhaust.

 

[JrClocker]

Interesting read Jr, I briefly looked through it but am too tired right now to really look into the details. It is pretty interesting how much difference there is in temps with the rad as intake vs exhaust.

Two measures of that one...AIO temp and CPU temp. My basic goal is to keep the AIO coolant temp as low as possible...so that CPU temp will be low.

The interesting result is that with the radiator as intake versus exhaust, both the AIO coolant temp and CPU temp dropped a lot:

High CPU Load, Low GPU Load:
- AIO temp dropped ~3.5 °C
- CPU temp dropped ~ 7 °C

High CPU Load, Medium GPU Load:
- AIO temp dropped ~ 5 °C
- CPU temp dropped ~10 °C

High CPU Load, High GPU Load:
- AIO temp dropped ~5.4 °C
- CPU temp dropped ~11 °C


With the radiator as intake, going from a fan pulling through radiator to pushing through radiator:

High CPU Load, Low GPU Load:
- AIO temp dropped by ~1.7 °C
- CPU temp dropped by ~1.3 °C

High CPU Load, Medium GPU Load:
- AIO temp dropped by ~1.5 °C
- CPU temp about the same

High CPU Load, High GPU Load:
- AIO temp about the same
- CPU temp dropped by ~1.2 °C


We have to look closely at some of the smaller number deltas (for max CPU temp) as I ran these tests over multiple days, and ambient is was only controlled by the thermostat about 20 feet away.

However, if we look at the delta in max AIO coolant temp (a better measure), we see a good story!

 

[doyll]

Very interesting bit of work!
Look forward to seeing more of it.

You may not care, but I use a cheap digital indoor / outdoor wired sensor thermometer to monitor CPU and GPU cooler/radiator intake air temperature. This allows me to see more accurately just how the temp and case airflow is effected by different combinations. If interest, I can give more info. If not, not problem.

Good work!

 

[EarthDog]

The interesting result is that with the radiator as intake versus exhaust, both the AIO coolant temp and CPU temp dropped a lot:

Sounds like you need better case airFLOW?

Typically, I only see differences of around 2-4C between my intake temperature and the intake of my radiator(s) on top.

You should also be seeing a 1:1 for rad intake temps to CPU temps. So if you lower the temperature 4C, you will see a 4C drop in temperatures, why your system is doubling that and bunking science, I have no idea.

 

[JrClocker]

Thanks for the feedback!

You may not care, but I use a cheap digital indoor / outdoor wired sensor thermometer to monitor CPU and GPU cooler/radiator intake air temperature.

My AIO cooler gives me coolant temperature through Corsair Link. I also get various motherboard temperatures.

I have a thermalcouple and meter at home...never thought of putting it at various places in the case...thanks for the idea!

Sounds like you need better case airFLOW?

Typically, I only see differences of around 2-4C between my intake temperature and the intake of my radiator(s) on top.

You should also be seeing a 1:1 for rad intake temps to CPU temps. So if you lower the temperature 4C, you will see a 4C drop in temperatures, why your system is doubling that and bunking science, I have no idea.

I don't think I am debunking science, but I do think I need better airflow...that's what caused me to start this whole "analysis" and "tweaking".

As I said above, some of the numbers may be off ± a few C as I was not controlling my ambient temperature tightly. In fact, as the AC kicks on in my house (got to love Florida), I can see the AIO temperature drop by around 0.5 C or so over 5 minutes.

The big factor is the heat being dumped into the case by the 2 air-cooled graphics cards. The AIO and CPU temperature rises with GPU power output.

In my original setup, the air being drawn through the AIO was already heated by the graphics cards. When I reversed the air flow through the radiator so that is was drawing in cooler outside air, the AIO and CPU temperatures dropped (at the cost of increased GPU temperatures.)

I would not expect a true 1:1 between AIO temp and CPU temp for the basic reason that the GPU power is still going into the case...heating up everything. The CPU temp I am citing is the package temperature...which seems to always higher than the individual core temperatures (I can look through my data to verify this). I'm not sure where this temperature sensor is on the chip.

It takes the loop more than 30 minutes to saturate. But, running 3 30 minute runs, with cool-down times, takes me a bit over 2 hours. I was just looking for trends.

So, while the temperature improvements may be ± due to uncontrolled ambient, the trends are still correct.

 

[EarthDog]

I would not expect a true 1:1 between AIO temp and CPU temp

I need to clarify my point a bit...

Whatever the radiator intake air temp is, your CPU temps should raise in a 1:1. There are other variables that will complicate things such as multiple radiators and other heat producing items.

There are links of this testing (ambient to CPU temp 1:1) from martin's/Skinee labs. It is how we, and a lot of sites, 'normalize' temperatures in reviews because of that fact. I will have to dig them up... but the more I am thinking about it the more that would NOT apply with multiple radiators, etc as it would be impossible to pin down intake temps and I believe an average would be useless in this case. It would take mathzzzz that I haven't used since High School to figure it out...

 

[JrClocker]

I need to clarify my point a bit...

Whatever the radiator intake air temp is, your liquid and CPU temps should raise in a 1:1. There are other variables that will complicate things such as multiple radiators and other heat producing items.

There are links of this testing (ambient to CPU temp 1:1) from martin's/Skinee labs. It is how we, and a lot of sites 'normalize' temperatures because of that fact.

Ahhh - that makes more sense. I'll pull out my thermocouple and play around with that.


For the record, I will be getting a 980ti hybrid next year...can't wait to not have so much GPU heat in the case. However, I have to pay for my daughter's wedding first!

 

[doyll]

JrClocker
I do a lot of case airflow work on my own and many others use my guidance setting good case airflow. Sorry if that sounds pompous. There is far more I don't know than I do know about airflow and cooler. If you would like, I can give you links to them, but I do not have them posted here. Basically case has to flow at least a little more air volume than components use, but the paths / channels the air flow through the case are just as critical. The object of the game is to flow all the component heated exhaust out of case without it contaminating and heating up the cool air being supplied to components.

EarthDog
While we do normally see temperature changes at about 1:1, I do not think the radiator, coolant and CPU are on an scientific 1:1 temperature ratio. Heat from CPU does not move for CPU to water block at 1:1, from water block to water, nor does radiator release heat to air at 1:1. There are many variables involved including air speed, humidity level, etc. While I can't scientifically prove this hypothesis, my daily use of cooling systems seems to show it to be true. Like you my college math may be in the grey matter somewhere, but I doubt I could find it.

 

[EarthDog]

Sorry, I should clarify again... ambient temp to CPU temp will go up in 1:1. Not sure what the cooling medium does.

 

[JrClocker]

doyll

I 100% agree with everything you said. I would appreciate the links.

I asked questions, did research, and basically I could not find 100% agreement from the "experts" on the "subtle" questions I was asking....and I have learned over my career that this is almost always the case. Sometimes you just need to get in there and run experiments.

That is what I did...and I was trying to share results that others could try. I was not looking for the last 1% of 1 C, just the trends that made things better. These are listed in the first 4 points of my post.

Using the general rule that heat creates more heat with electronics, I'm on a mission to channel the heat away from them. In hindsight, I should have guessed that the biggest factor for heat was the heat load being put into the system by the graphics cards. Makes sense now that I have data which shows this. And, the graphics cards I am using don't really direct their heated airflow well (EVGA ACX 2.0 cooler).

I am going to play around with different options to "channel" this heat better...so again, I would appreciate the links!

Please keep up with any feedback, and thanks again!

 

[doyll]

link to basic guide here.
http://www.overclockers.com/forums/...nd-Lighting-Suggestions?p=7678268#post7678268

I should probably start a new thread and add more posts

I also PM'ed you a link to more

 

[JrClocker]

Thanks!

 

[JrClocker]

Mental note:

- When you put your desktop on top of the desk instead of under the desk, temperatures drop

DUH

It makes sense as with the PC under the desk, some of the warm exhaust air was getting trapped under the desk...causing the PC to intake this warm air...get hotter...until it reaches equilibrium.


I avoided doing this all of the years as I didn't have long enough cables...and there are A LOT of cables coming out of a PC! So, the other day, I bit the bullet, bought longer cables, and spent 2.5 hours moving and dressing all of the cables (so that it didn't look like a rat's nest). So now I have this big honking case sitting on top of my desk...good thing it has bling! Bad thing, needs more bling!

The result, temperatures dropped for the AIO coolant temp, CPU temp, and 980 Ti hybrid graphics temp. In fact, I was able to lower the speeds on all of my fans...and I am still running 2 C lower on the AIO coolant temp and about 4 C lower on the CPU temp.

My usual routine is to let my PC fold over night. In the past, my office would be noticeably warmer than the rest of the house. This morning, my office was much cooler.

 

[Mandrake4565]

Additional mental note, if trying to Oc a Pc that is in a desk cubby take it out. My first Ocing adventure I was turning the knobs on my rig sitting in my desk cubby. Even though the door was open and the entire back side of the desk was open my temps still dropped by near 8c with it out.