I wanna to test my stuff, but also post it online, and find best cooling for my case.

[Kwadratowicz]

Hello,
I'm trying to benchmark my case cooling capabilities, but I would like to post it online and I'm asking for help with testing methodology and data gathering ideas.

I'm using gpuz as main source of info, but I will be adding hwinfo for data that is not avaliable in gpuz.
I'm using steel nomad 20x 3dmark stress test as main benchmark.

Case is Lian Li Lancool 207, I've got stock fans. as well a lot of lying around fans from noctua (lower tier), arctic p120 max and p140max,

i wanna to find best combo for gpu lowest temperatures, it means that reconfigure fans, change orientation of cpu cooler from horizontal to vertical, 2 fan on cpu vs 1 fan, adding ducts, and closing mesh where I think the air is escaping without helping etc.


But all of that can be scrapped if I don't have standardized test, and that is what I'm asking, ideas how to make it repetitive as much as it can be.

This picture is for now as what my system looks like, but I will add later normal pictures done with camera.


Example test (unfortunatelly I forgot to powerup hwinfo for ram and other temperatures).

Settings of the gpu.

Max data (for now it is screen from gpuz, later I wanna add max from raw data in spreadsheet.)


Average data from test.

The test scores.


Everything meaningful will be added to spreadsheet and compared with other runs.

 

[Chrispy_]

I applaud your wish to test scientifically and consistently, but not all temperature sensors are equally important.

For gaming, your GPU hotspot temp is the priority, that affects your max boost clock and everything else is secondary. If you're cooling your GPU optimally, the GPU's memory temperature shouldn't be of concern, neither should your GPU's VRM temperature. The two basement fans in the Lancool 207 are the most important fans in the case, just make sure there's not a rats nest of wiring impeding their air intake for the GPU.

Your system's biggest weakness is the blow-through cooler of your graphics card dumping a ton of hot air straight into your CPU cooler. It won't actually matter that much as when this is happening you will be GPU bottenecked and your CPU is unlikely to be running at its thermal limits anyway. However, if you want to improve the CPU cooling situation as best as you can, do this.

First of all, put three 120mm fans in the roof of the case. Make sure the front one is an intake to feed air to the CPU. The other two should be exhausts and their main purpose is to increase the overall CFM of your case. There will be a small percentage of recirculation from the intake and the exhaust right next to each other, but it really is negligible and not worth worrying about.

Next, 3D print yourself something similar to this and bolt it to your CPU coolers fan, with the intake pointing up right at the new intake fan you've fitted in the roof.

NR200 CPU Intake Duct by darknommer

Intake fan duct for air coolers for NR200, prevents CPU air cooler from ingesting warm case air (from GPU or other components)Fits 120mm fans, can likely be scaled up to fit 140mm fans

www.thingiverse.com

What you'll have is this, blue is intake, red is exhaust:



You'll have an imbalance of two fans positive pressure, which will push air out of the all the venting on the rear panel (that's why it's there). That CPU intake cowl is important. It can be either a straight pipe to the front intakes, or a box section that grabs air from the front and top intakes. It doesn't really matter what it is as long as it blocks hot air from the GPU from going straight into the CPU fan. If you don't have access to a 3D printer, just fashion anything that can be screwed into the CPU fan's unused holes with fan screws. Plastic packaging, a hole-punch, and a pair of scissors and some sticky tape will work to make a rudimentary box pipe in a pinch.

 

[Shrek]

i wanna to find best combo for gpu lowest temperatures

If that were your only objective would that not imply having all case fans pump inward, so the card flow is maximized?

 

[Carlyle2020hs]

Please elaborate.

As far as i have come to understand it, an over-pressured case has a rather chaotic flow pattern, no?

Sure it helps to churn air in dead spots that normal airflow doesn't get to properly, but to call it maximized?


... If so there must be a fan rpm tipping point - when the pressure gets too low for the over-pressure to combat the missing active outlets

Something like ... :

lower than xxx general case-fan rpm and oulet fans make sense again?

 

[Shrek]

I actually think that concentrating on 'gpu lowest temperatures' alone may be an error.

My point being that to use this criterion could lead to a highly non-optimal situation.

 

[ir_cow]

Proper case reviews are far and few between. I commend you for trying.

Take your time and do research on thermal dynamics and aerodynamics to understand what may or may not be important to the end user.

 

[ShrimpBrime]

What is even really the focus point here?

Ok so do 100 (edit: 10) runs each benchmark for each different setting or hw configuration you change. Should get us some averages.

No. That is not positive case pressure. lol. You have some pressure at the fan blades maybe....

 

[Kwadratowicz]

the GPU's memory temperature shouldn't be of concern,

It's new stuff, memory is hotter than gpu hotspot, and that's concerning for me and it is main focus point in My testing, and with heat by proximity gpu hotspot affects vram temp, and vice versa.

hot air straight into your CPU cooler

Cpu Cooler is specially noctua nh d15 g2 cause of thermal mass, and it will be rotated to pump cold air to gpu backplate.

First of all, put three 120mm fans in the roof of the case. Make sure the front one is an intake to feed air to the CPU. The other two should be exhausts and their main purpose is to increase the overall CFM of your case. There will be a small percentage of recirculation from the intake and the exhaust right next to each other, but it really is negligible and not worth worrying about.

That will be tested , for now I'm trying to get best base in case of how to manage data, like what heat load to gpu add etc.

If that were your only objective would that not imply having all case fans pump inward, so the card flow is maximized?

I've tested with case before, that all positive pressure inwards is making to much turbulent flow and is in fact choking gpu, instead of adding airflow.

... If so there must be a fan rpm tipping point - when the pressure gets too low for the over-pressure to combat the missing active outlets

Something like ... :

lower than xxx general case-fan rpm and oulet fans make sense again?

After making good spread sheet that will automatically manage data inserted I would love to test different fan speeds, (but for that I need really good spreadsheet that do everything automatically, cause manually comparing data in this case scenario could be dangerous to mental health )

Take your time and do research on thermal dynamics and aerodynamics to understand what may or may not be important to the end user.

That's irrelevant for now, cause I wanna test even the stupid thoughts , and some idea of aerodynamics, and thermal dynamics I posses it is of course a little, but I know what generates and how heat flow in grand scale , cause those tiny scales are always problematic .

What is even really the focus point here?

How to test and how to manage data.
For example I thought of changing the power limit of gpu to 70% cause with 100% it will bang ahead of thermal limit, and the clocks won't be as stable as in lower PL, but it could be mine misinterpretation.
with PL 70% hotspot was over 70C and Vram was around 73.

For now I've changed the hw info to gather only handfull of data.



So now questions are:
Is power limit 70% which has lowered temperatures a good solution to gather data that will be comparable with runs?
What data should be collected and what is most important? (For now I think Vram C is first, Hotspot C is second, and CPU C is last, I don't care about cpu cause it is ryzen 5700x with nh d15, it has a lot of headroom to even soak heat from gpu).
If I distract the room temperature from the gathered data is it ok, or I should add room temperature, or distract the room temperature delta between runs to get best results (to normalize it)?

 

[ShrimpBrime]

It's new stuff, memory is hotter than gpu hotspot, and that's concerning for me and it is main focus point in My testing, and with heat by proximity gpu hotspot affects vram temp, and vice versa.

Cpu Cooler is specially noctua nh d15 g2 cause of thermal mass, and it will be rotated to pump cold air to gpu backplate.

View attachment 401988



That will be tested , for now I'm trying to get best base in case of how to manage data, like what heat load to gpu add etc.

I've tested with case before, that all positive pressure inwards is making to much turbulent flow and is in fact choking gpu, instead of adding airflow.

After making good spread sheet that will automatically manage data inserted I would love to test different fan speeds, (but for that I need really good spreadsheet that do everything automatically, cause manually comparing data in this case scenario could be dangerous to mental health )


That's irrelevant for now, cause I wanna test even the stupid thoughts , and some idea of aerodynamics, and thermal dynamics I posses it is of course a little, but I know what generates and how heat flow in grand scale , cause those tiny scales are always problematic .

How to test and how to manage data.
For example I thought of changing the power limit of gpu to 70% cause with 100% it will bang ahead of thermal limit, and the clocks won't be as stable as in lower PL, but it could be mine misinterpretation.
with PL 70% hotspot was over 70C and Vram was around 73.

For now I've changed the hw info to gather only handfull of data.

View attachment 401989

So now questions are:
Is power limit 70% which has lowered temperatures a good solution to gather data that will be comparable with runs?
What data should be collected and what is most important? (For now I think Vram C is first, Hotspot C is second, and CPU C is last, I don't care about cpu cause it is ryzen 5700x with nh d15, it has a lot of headroom to even soak heat from gpu).
If I distract the room temperature from the gathered data is it ok, or I should add room temperature, or distract the room temperature delta between runs to get best results (to normalize it)?

All the data. Include everything always. Drive temps, cpu, gpu, motherboard, fan rpm, fan orientations, all of the data.

Needs a control baseline. Anything changed is compared against your control.

Must state the objective and the change for each parameter you want to compare against your control. What exactly was changed and what exactly is the desired outcome and what exactly is the actual outcome.

And the more runs accomplished, the more accurate the average data will become. Or you could do like best of three runs and include only that data per configuration change.

 

[Kwadratowicz]

Needs a control baseline. Anything changed is compared against your control.

As base I will test temperatures with all case fans off. Thanks for idea, cause I forgot that I should have some baseline , and I will do it 3 times, first, somewhere between and last tests.
I really thank You for the ideas , I will add Your comment on top of the spreadsheet to remember it for changes
And other comments that I find interesting to look while testing to not forget the important things, thanks everyone for ideas

just make sure there's not a rats nest of wiring impeding their air intake for the GPU.

I've tidied cables and made them to not obstruct any of the airflow.
Before:

After:

 

[ShrimpBrime]

As base I will test temperatures with all case fans off. Thanks for idea, cause I forgot that I should have some baseline , and I will do it 3 times, first, somewhere between and last tests.
I really thank You for the ideas , I will add Your comment on top of the spreadsheet to remember it for changes
And other comments that I find interesting to look while testing to not forget the important things, thanks everyone for ideas

I've tidied cables and made them to not obstruct any of the airflow.
Before:
View attachment 402086View attachment 402087
After:
View attachment 402089View attachment 402088

Since you have the case open, another idea for cooling is to see how fanning the back of the socket helps. We used to do this on the old FX platforms where AMD recommended 65c max temp for extended load periods. Im curious if this would still be an effective cooling method on more modern hardware.

Nice looking rig BTW. I like the all white look.

 

[Chrispy_]

It's new stuff, memory is hotter than gpu hotspot, and that's concerning for me and it is main focus point in My testing, and with heat by proximity gpu hotspot affects vram temp, and vice versa.

Cpu Cooler is specially noctua nh d15 g2 cause of thermal mass, and it will be rotated to pump cold air to gpu backplate.

View attachment 401988

The 9070XT has no memory modules on the back of the card, and all of the teardowns I've seen have no significant contact between the backplate and the PCB - not that cooling components through the backside of the PCB has much effect anyway....

So, using the CPU cooler to get airflow across the back of the card is likely to do nothing or at least almost nothing.

IMO you are most likely to get the best GPU temperatures by ensuring that there's a very strong bottom-to-top case airflow and that everything in your case is pushing air in the same general direction to maximise the speed at which hot exhaust air coming out of the CPU and GPU's coolers is evicted from your case.

For you, the entire bottom and front are your intake areas, so the rear and top are exhust. The yellow line is your imaginary divider between intake and exhaust zones, so you want airflow to be generally going across this line in the most direct and perpendicular route as possible (white arrows):

 

[phanbuey]

I've actually found top intake to be the best with AIO - AIO fans and pump to ramp with liquid temp (I like to keep it at ~26C and below) and then 3 intake fans just pushing heat out, ramping based on GPU memory temps (I usually use Argus for fan control for an all-in one lightweight system - AIO & pump on liquid temp scale, front intake fans on gpu mem ramping - since GPU mem thermal sensor is the best at knowing when the card is dumping 400W of heat into the case).

This tends to be the coolest and quietest setup -- the ram and vrm get airflow from the rad, the heat from everywhere gets pressured out the back. Scaling AIO with liquid temps instead of CPU has a REALLY nice acoustic profile - water heats up slowly and the fans ramp slowly to match, it's really quite pleasant.

Top exhaust doesnt actually work so well through fine mesh, neither does trying to cool your AIO by hot gfx exhaust -- intake ends up working much better and the fine mesh acts as a dust filter.

 

[Kwadratowicz]

So, using the CPU cooler to get airflow across the back of the card is likely to do nothing or at least almost nothing.

Gigabyte is strange card that has connected with thermal pad gpu die area with back plate, and I've got some old gpu radiator that should fit under the cpu cooler, that I can attach to back plate with thermal pad, cause while gaming back plate is almost to hot to touch. (but this setup will be almost the last to test , cause It could be the best, so I don't want to disassembly it later . )

For you, the entire bottom and front are your intake areas, so the rear and top are exhust. The yellow line is your imaginary divider between intake and exhaust zones, so you want airflow to be generally going across this line in the most direct and perpendicular route as possible (white arrows):

The idea bout it is to study every possible combination that I would think of, and have time to set up , I know that what You showed is likely to be the best solution, but the topic is not on best solution, but to make tests , (I did not thought about 1 fan in top to be intake, and now I'm testing it with intake, thanks for showing it )

So, using the CPU cooler to get airflow across the back of the card is likely to do nothing or at least almost nothing.

I see another problem, cause with that high cpu cooler it would be suffocating the gpu exhaust, but I will test push and later pull configuration in fans .

I've actually found top intake

That will be tested .

front intake fans on gpu mem ramping - since GPU mem thermal sensor is the best at knowing when the card is dumping 400W of heat into the case).

when I'm not testing, all case fans are steering by the vram temps, cause it is hottest component of my PC.
Thanks again, and see You tomorrow (cause I was thinking why the heck I've got poly carbonate (PC). )
Ps I've got first test data saved (All case fans off, only gpu on 40% and cpu on 50% fan speeds), now I need to do the spreadsheet .

 

[Chrispy_]

The idea bout it is to study every possible combination that I would think of, and have time to set up , I know that what You showed is likely to be the best solution, but the topic is not on best solution, but to make tests , (I did not thought about 1 fan in top to be intake, and now I'm testing it with intake, thanks for showing it )

Good luck with the testing. You'll end up with the best result for your particular combination of case and components.

I'm telling you what I'm telling you because this isn't the first time someone has investigated fan layout for cooling before. I've personally read or watched at least a dozen in-depth articles or videos on various fan positions/quantities/orientations over the last three decades. They're all slightly different but they all basically agree that air goes where it's pushed and the faster you can get hot air out of the case, the better.

I was one of those nutters who was making ducting out of plumbing pipe and cutting massive 120mm fan holes in their side/front panels back when cases were beige and only had room for one case fan. There's almost no (sensible) variation of cooling I haven't tried, and I've even dabbled with stupid stuff like Peltier coolers. IMO, modern cases like the H5 flow and LanCool207 with intakes directly below the graphics card are exactly what the ageing ATX standard needs, because for at least 15 years now, graphics cards have typically drawn more power and needed more cooling than CPUs. Play to their strengths, that's what your case is designed and tested for.

 

[Kwadratowicz]

I'm telling you what I'm telling you because this isn't the first time someone has investigated fan layout for cooling before. I've personally read or watched at least a dozen in-depth articles or videos on various fan positions/quantities/orientations over the last three decades. They're all slightly different but they all basically agree that air goes where it's pushed and the faster you can get hot air out of the case, the better.

I'm doing it cause i like spreadsheet stats and tinker Ing with pc, lately playing games is not as fun as before, but playing with components has always been fun for me
I know that topic is vast, and I'm not the Best there to do it, but it is fun to See what I can learn and where I will get with that
Even if its dumb its fun to watch like it fails (Talking about different fan configurations )

Edit:::
For example the 20 minute stress test with steel nomad was to short with only fans on cpu and gpu to stabilize the cpu temperature, it was raising
and now I'm thinking of running 1h stress test for baseline to see where cpu temp could get

 

[Chrispy_]

I'm doing it cause i like spreadsheet stats and tinker Ing with pc, lately playing games is not as fun as before, but playing with components has always been fun for me
I know that topic is vast, and I'm not the Best there to do it, but it is fun to See what I can learn and where I will get with that
Even if its dumb its fun to watch like it fails (Talking about different fan configurations )

Edit:::
For example the 20 minute stress test with steel nomad was to short with only fans on cpu and gpu to stabilize the cpu temperature, it was raising
and now I'm thinking of running 1h stress test for baseline to see where cpu temp could get

If you want to get a constant heat load for faster testing, don't use Steel Nomad, use Furmark and CPUBurner. They are repeatable, synthetic loads that will instantly put your CPU+GPU to 100% power consumption and keep them there at a constant power draw.

You're testing temperatures, not game performance, so you don't need to waste all that time loading 3DMark and Steel Nomad, which aren't even constant power-draw tests for thermal testing.

 

[Kwadratowicz]

@Chrispy_
I'm quite wondering if furmark is good for the type of data I want to collect and compare.
Furmark if I've read correctly is not vram hungry, and is not stressing the vram like game benchmarks do, that's why I think for my specific tests to get best data for vram temp as main culprit better should be some of the benchmarks.
But after reading Your post I think steel nomad is not the go to benchmark I will test today with port royal, and see if the temperatures within stress test have stabilized (it is longer, and with raytracing it should be a little more demanding for gpu. )
Steel Nomad duration 1min, but Port Royal duration is 2 minutes, so stress test should be 40minutes instead of 20 like with steel nomad.

 

[Chrispy_]

Maybe someone else can correct me here, but VRAM temperatures are more a factor of core temperatures spreading heat to the VRAM. I've seen situations where all 24GB of an RTX 3090 are in use at max VRAM clocks, and they're running super cool because the core is only running a low load, and instances where the VRAM is roasting hot because the core is flat out at 100% max boost and only 1.5GB out of 24GB VRAM are in use.

If you want to stress the VRAM for stability, use OCCT. If you want to get the VRAM hot, Furmark should do that consistently which is what you want for temperature testing.

 

[Kwadratowicz]

ok so I need to test furmark and port royal just to be sure that You are right ^^.
I don't wanna to get the vram hot I wanna to work vram to get hot