More Radiator-Sandwich testing

*** Update ***

AndreaBZ pointed out an inconsistancy in my data from this test to previous tests (Thanks!). I found the problem: I had calculated the values for the solo TC and solo Magicool from the data from two thermal probes in the water and all the other values from all four probes in the water. This was purely an error in calculating.

I corrected the values and uploaded new graphs. There aren't any dramatic changes. The greatest change is that now, at 1200rpm, the Rad-Rad-Fan option is slightly better than the solo Magicool (as opposed to slightly worse).

New diagram:

http://www.abload.de/img/newdiagramcopyuqw1.jpg


About KaptChrunch's suggestion (triangular setup instead of stacking): That would certainly have the advantage that the rads wouldn't heat each other up. Or at least, not as much as in a stacked setup. I imagine that it would be problematic for the airflow though, since the fins on the radiators would no longer be aligned with the direction the fans are blowing in. Don't know how much of an issue that would be though.

Nice clear graph, good work.

Quote:

---

Originally Posted by KaptCrunch

the static pressure will flow air through the fins anyway best use a caged fan for pressure ;)

---

So you'd have to make an airtight construction, right? Or almost airtight, anyway.

I love your graph!:up:

That graph is kickass wtf. THe attention to detail (the lil arrows showing heat rising/cooling inside the rads) is amazing. Great work. What are you using to make the diagrams? Just simple excel?

Thanks, guys, I'm glad you like it.

soulless: The diagrams are from excel and the pics are done in photoshop (I know, illustrator would be the right piece of software for that kind of thing, but I don't have it).
I screenshoot the diagram in excel and then cobble together the final thing in PS.

Quote:

---

Originally Posted by HESmelaugh

Such silent fans are used by many people in Liquid cooling setups since for many, the main benefit of liquid cooling is reduced noise.

---

Reduced noise/near silent is definitely half my battle when it comes to watercooling.

Quote:

---

Originally Posted by Taylor01

Yes! Exactlly my point who cares if you can run them with 200cfm delta's. I don't want a 747 in my living room!

---

That is exactly what my first high-performance PC sounded like. It was a P4 Alienware with 5 high performance 80mm fans, it seriously sounded like an airplane.

Quote:

---

Originally Posted by Waterlogged

Actually, yes I did read your conclusion, but it seems most ppl aren't, or at least they aren't comprehending it.
When compared to ZM-F3's or S-Flex E, F or G, yeah, those are very poor quality fans. If I had to guess, I'd put them somewhere around the Noctua in this test as far as rad usage. Those numbers look about as made up as SilenX's. You can still have reasonable dB's and cooling power with any of the fans I mentioned.

---

I currently use S-Flex E and F's in my current main PC and I don't think I could ask for more quiet than they are. How much better are the S-Flex's in comparison to the fans used in this test? And would they really skew the number in this test significantly? The increase presure will help the sandwiched radiators, but it would also help the stand alone radiators somewhat as well, enough so to make the difference made to the sandwiched radiators insignificant?

Quote:

---

Originally Posted by HESmelaugh

Unfortunately, I have not yet compared nanoxias to S-FLEXs on a radiator, but on a TRUE, the nanoxias perform slightly better at equal rpms.
Can't be all bad.

---

You've already done enough testing for all of us to appreciate but dont hesitate to do more with different fans if you get a chance. I'll look forward to future tests from you.

Love the new graph! Very easy to understand at a quick glance.

If you're curious what things to watch for in radiator testing, these are some observations I learned to watch for:

Here is an example of a test I ran for over a hour thinking that was Plenty of time to get up to temperature. It was a TFC120 with 150 watts and a yate loon D12SL12 at 1000RPM. It was the first test, so the rad was completely cold. Furthermore the room I was testing in wasn't warmed up and holding temperature either. Both of these made this test no good.

Here is the test logged and plotted over time:
http://img513.imageshack.us/img513/8161/radtestbad.png
It appeared I was approaching around an 11C water delta. I also noticed my air out sensors weren't placed right.

But retesting, this is what I got:
http://img513.imageshack.us/img513/8043/radtestok.png

There's a couple of things you can see here. First the ambient temperature during the test has to remain constant. You can see how a very small climb or reduction in ambient throws off the water/air delta. You wouldn't see this if you were sort of at a constant climb or constant lowering of temperature, it would simply give you a result that was high or low by a degree or two depending on the difference.

These are the things I tried to do after seeing some of these things:

-Run your first test for a couple of hours until both the radiator and the room have leveled off in temperature.
-Don't test when your ambient temperatures can change and don't test with a forced air heater or any thermostatically controlled unit turning on and off, it'll screw up your results.
-Log log log, and plot your test results including a running water/core delta over a period of time to ensure everything was right. You can't see this unless you log the results and graph it.
-Log RPM of the fan, this will also change as the fan warms up.
-Also be VERY careful about air movement in a room. A simple thing like turning a ceiling fan on or having a window cracked WILL give you different results.

Overall the biggest thing is warming everything up and testing for very long periods of time when everything is nice and stable and logging everything. Fans have to be warm, the radiator has to be warm, and the room has to be holding constant.

You can also really tighten up the results by testing multiple times at different heat loads.
You can see some of the error that averaged out for me in this TFC480 test. The trendlines split the difference making my 1C error maybe only .5C of error. Also notice how my results are much better and more consistent between the two tests with the higher speed fans and lower deltas. The low speed fan tests are EXTREMELY touchy and prone to error.
http://martin.skinneelabs.com/img/TFC480-Thermal2.png

It sucks though because that means testing for potentially 10 good 1 hour tests, and likely a handfull of bad tests, and if ambient temperatures won't hold...you can forget it. In the end I found it was pretty typical to require the better part of a weekend to do one full set of tests on one radiator.

Then I find I might get $30 out of selling one of these rads like a 120 sized radiator on ebay later. So I if I had 20 hours of testing into one radiator, I get $30 out of it in the end, I've made $30/20hr...or about $1.50 an hour for my time...whoohoo! After about the umpteenth time making this $1 an hour, I was done.

So anyway, Kudos to the great testing. Hopefully a few of these problems I've seen may be helpful for you to watch if you're battling any inconsistencies, and furthermore I sure appreciate the time you're putting into this testing. It is truely a donation of your time, samples don't mean jack when you pencil out the time involved and it's very touchy doing radiator testing.

I'm looking forward to seeing more of this. Keep up the awesome work and the graphing/presentation is perfect!!

Martin, thanks for your input. I appreciate that very much! :)

I've also noticed, that Radiator, room and fans need to be warmed up first, to be able to get repeatable results. The fans seem to warm up very quickly but it takes a while for the whole loop to warm up.
With radiator testing, there are still some things I don't have enough control over. For one thing, I can't log fan speeds and neither can I log the Wattage of the heaters. Although the latter seem to be very constant. But as one can see from taking a close look at my data and as some have pointed out, there are still minor inconsitencies to be found. I certainly can't (and don't mean to) claim that my results are spot on, down to the decimals. But I think my testing is good enough to provide a rough overview. I can say what's clearly better and what's clearly worse, but maybe not what's marginally better or worse.

For testing CPU-blocks, I'm much happier with my results (some wil be up in a few weeks). I've managed to tune my methods to a point where it's not rare to see a total deviation of less than .2K accross three mounts so I'm very happy with that.
Which reminds me: I tested a few different CPU loading programs (Prime95, IntelBurn, CoreDamage, MaxPerf) to see which is most constant in temperatures and which produces the most heat. I could post some data on that here too, if anyone cares to see it.

Anyway, back to Martin's post.
I grabbed a random three hours out of the logs of my radiator tests:

http://www.abload.de/img/tempverlaufg0h5.jpg

Lowest temperature during this time was 20.8°, highest temperature was 21.7°. It looks like there might be a slight upwards trend here. Would you say that this is a tolerable kind of fluctuation for valid testing?

As for the monetary reward for testing: I couldn't agree more with you, Martin. It's not worth it in any way. It's only curiosity that gets me testing these things. If I just wanted a bunch of rads or waterblocks for myself, I go the cheap route and buy 'em. ^^

Cheers,
Shane

Stacked rads are fine with high RPM fans. Have I tested it? No. It is just plain physics. Once a certain airflow rate is reached, two stacked rads will approach the performance of two seperate rads to the point of infinity.

They just don't work for low airflow because the feedback vs. airflow ratio has not yet been overcome. IMO this makes them worthless for me because I am silent type.

Thanks for the excellent test though.

Quote:

---

Originally Posted by HESmelaugh

Martin, thanks for your input. I appreciate that very much! :)

I've also noticed, that Radiator, room and fans need to be warmed up first, to be able to get repeatable results. The fans seem to warm up very quickly but it takes a while for the whole loop to warm up.
With radiator testing, there are still some things I don't have enough control over. For one thing, I can't log fan speeds and neither can I log the Wattage of the heaters. Although the latter seem to be very constant. But as one can see from taking a close look at my data and as some have pointed out, there are still minor inconsitencies to be found. I certainly can't (and don't mean to) claim that my results are spot on, down to the decimals. But I think my testing is good enough to provide a rough overview. I can say what's clearly better and what's clearly worse, but maybe not what's marginally better or worse.

For testing CPU-blocks, I'm much happier with my results (some wil be up in a few weeks). I've managed to tune my methods to a point where it's not rare to see a total deviation of less than .2K accross three mounts so I'm very happy with that.
Which reminds me: I tested a few different CPU loading programs (Prime95, IntelBurn, CoreDamage, MaxPerf) to see which is most constant in temperatures and which produces the most heat. I could post some data on that here too, if anyone cares to see it.

Anyway, back to Martin's post.
I grabbed a random three hours out of the logs of my radiator tests:

http://www.abload.de/img/tempverlaufg0h5.jpg

Lowest temperature during this time was 20.8°, highest temperature was 21.7°. It looks like there might be a slight upwards trend here. Would you say that this is a tolerable kind of fluctuation for valid testing?

As for the monetary reward for testing: I couldn't agree more with you, Martin. It's not worth it in any way. It's only curiosity that gets me testing these things. If I just wanted a bunch of rads or waterblocks for myself, I go the cheap route and buy 'em. ^^

Cheers,
Shane

---

That's great!

I think I was generally pretty happy when my ambients and everything was holding withing about .5C over a 30minute period.

I couldn't log wattage either automatically, so I just manually recorded the watts used on about 15minute intervals and averaged them. You're household power must be better than mine. I could see pretty big fluctuations in voltage and I think that was part of my fan RPM variation as well. Trying to pin down say a 300 watt heat load would usually mean a fluctuation from 290-305 watts or so, so my power grid is a bit more up and down. My fan RPM would also generally go up or down +- 25-40RPM or so, but the crystalfontz logged it all, so I just averaged the recordings.

In the end, I think the hardest part for me when radiator testing was the testing of low speed fans, they were much more difficult to get good repeatability for me than the medium/high speed fans. CPU thermal testing is even worse for me.

The rad testing is very interesting. I appreciate you doing this sandwich work, that had to take you a huge amount of time. I wish I still had the passion for it, but even after 4 months of vacation I still don't even want to look at the testing equipment especially in regards to thermal testing. The hydraulics side of things is more interesting to me for some reason, too bad flow rate impacts are so insignificant....oh well.

Hopefully Swiftech will provide some additional testing specific to the MCR series. Even if it's only a 10% gain, that would still provide about the same gain a double thickness type rad would. It's not going to be huge by any means, but at least it's an option if you only have the space for one or two stacked.:up: I think the results here are spot on, stacking rads just isn't going to do much of anything for really low speed fan use. Gains will only be measurable with medium to high speed fans.

Carry on with the great science and work for the good of the whole:up:

Right, the quality of the power grid factors in as well. I hadn't even thought of that yet.

I'll try logging the wattage manually as well. Once every 15 minutes seems manageable.

For me, the hydraulics is the elusive part. I still don't have a manometer (on the fence about spending the necessary cash for one) and I wouldn't immediatly konw how to use one, either. With the high heat loads of the i7, I have been seeing some pretty decent gains with increased flowrates. Not in the regions you guys like your flow, I guess, but up to around 1,5GPM. I guess starting with pressure drop/flowrate measurements would be the next step forward in my testing.

Anyway, thanks for the encouragement and input, Martin. :)

@HESmelaugh
Wouldn´t it be absolutely gorgeous if martin forgets his "retirement" and you two meet somewhere to do some testing for us??? There is still so much to do :D

@martin
:stick: ?

Hey guys,

The Swiftech stack concept works very well in practice with just a medium speed fan in between. I'll post my CoreTemp image when I get home.

The pictures are in my completed first phase worklog entiteld "Genetics." You will see that there were not very many solutions for such cramped quarters.

Quote:

---

Originally Posted by scamps

@HESmelaugh
Wouldn´t it be absolutely gorgeous if martin forgets his "retirement" and you two meet somewhere to do some testing for us??? There is still so much to do :D

@martin
:stick: ?

---

Between all the great testers already on here, there's no need.

I'm learning here just like everyone else too..:up:

I'm getting the 240's stacked variant today and planned to do some testing on my own. I don't have any fancy equipment, but tell me what you think about this:

Log cpu temperatures over time with real temp, look at max min deviation of air in over time (cannot log). Take max cpu temp - max air in for comparison?

Planning to do this with UK 2000 and YL M @ 5, 7 and 12V with both single and stacked.

Great that you will do the testing!

The method you describe sounds solid enogh. Depending on what the room you test in is like, you might not see much change in ambient temperature.
You can calculate maxCPU-air and avgCPU-air for both stacked and non-stacked and see which option gives you the better results.

If you want to test how good your method is without using fancy equipment, simply repeat the test-run after a day or two. If you get similar results, that means your method must be pretty solid. :)

Looking forward to your results. :up:

Thank you for the input HES :) Good suggestion with repeating after a couple of days :up:

Nice to know that I was thinking along the right path. Also, I will look at RPM in free air, single and stacked to see how much difference it makes at various voltages. I think from your testing it seems crucial that higher rpm fans are needed to make it work as intended.

Yes, that would be very interesting. Unfortunately, I can't log fan voltages and rpm, so I would also be really interested in seeing the differences in free air, single and stacked. :)

Wow amazing thread!

I am putting together a TJ07 with a PA120.3 and a PA120.2 in the lower compartment and was wondering which is the best configuration.

Originally I thought seperate 5 fans, which this thread confirmes will be the best cooling! this saves me a headache in going down the sandwich route, THANKYOU!

Quote:

---

Originally Posted by Bond Number

I don’t know if anyone has tried this configuration, but it should work better with lower speed fans. It would require a little more work and slightly greater volume, but should fit into 5 ¼ inch bay. The center baffle could be beveled to direct flow out towards the sides or the ends of the radiators. One could also add small fans (25mm to 40mm) along the edges of the center baffle to lower the pressure between the radiators.

---

This is almost how I intend to use a PA120.3 and a PA120.2 in the TJ07 lower compartment.

The only difference is that both sides are the outside of the case and the middle hot air section would exhaust out the rear grill next to the PSU.

For it to work correctly the lower compartment needs to be sealed to maintain high positive air pressure.

Nice test just a quick question in scenario 3 how far apart where the fans for each rad when you tested this?

The reason I ask is if there was only a small channel between the fans (like in a dual rad setup in the bottom of a TJ07) would the results still outperform the other scenarios?

what about a double decker sandwhich


Gentle typhoon - RX240 - Gentle Typhoon - RX240 - Gentle Typhoon

gonna squeeze something like that into the bottom of my tj07

Quote:

---

Originally Posted by Economist

Nice test just a quick question in scenario 3 how far apart where the fans for each rad when you tested this?

The reason I ask is if there was only a small channel between the fans (like in a dual rad setup in the bottom of a TJ07) would the results still outperform the other scenarios?

---

The radiators were actuall at a 90° angle and quite a bit apart in scenario 3. It's just that I couldn't illustrate that neatly like the rest of the scenarios.

You'll definitely need to make sure that there is enough air being fed into the space between two radiators, if you're going to set them up like that. For a TJ07 I would personally add a grill to or completely remove the panel separating the top from the bottom part of the case.