Here are the results of two days worth of testing a Ninja for the best fan direction/setup. As you can see the board is on a table, using only one low-speed 80mm Yate Loon. I tested with the board horizontal, with the board vertical, with/without the small duct, with the fan sucking/blowing in different locations. These tests were run at 29C ambient, with the fan always at 12V (about 1925rpms). I used CPUBurn to stress the P4-2.4 NW.



This is the horizontal orientation.



This is the vertical orientation.



This is the best setup.....vertical orientation, fan on top blowing upward, using the duct.



And here's the same setup with the fan blowing downward.

Here's the whole list of the maximum temps with the fan and board at different orientations.

1. 47C....board horizontal, using duct, fan on LS sucking.

2. 46C...board horizontal, using duct, fan on LS blowing

3. 48C...board horizontal, no duct, fan on LS sucking.

4. 47C...board horizontal, no duct, fan on LS blowing.

5. >58C...board horizontal, no fan.

6. 47C...board vertical, using duct, fan on bottom blowing up.

7. 47C...board vertical, using duct, fan on LS blowing level.

8. 48C...board Vertical, using duct, fan on top blowing down.

9. 45C...board vertical, using duct, fan on top blowing up.

10. 47C...board vertical, no duct, fan on top blowing down.

11. 48C...board vertical, no duct, fan on top blowing up.

12. >58C...board vertical, no fan.

The results are fairly consistent .......The best temps used the duct. Normally blowing through the Ninja gave better temps than sucking. The best temps had the board (and the ninja fins) in a vertical orientation.

The only reading that stood out......45C. This had the board vertical, using the duct, with the fan on the top sucking (blowing upward).

Conclusions to all this.....Use the board in a tower case Vertical. Construct a small duct that causes the airflow to contact the fins for a relatively longer time. A short duct (covering about 2/3 of the fins) seems to work better than a duct that completely covers the fins, with only the one side open. Make use of convection currents....a fan on the top blowing upward seems to work best.

I did try using two fans in a push/pull arrangement. This usually lowered the max temp only 1C, with much more noise. The absolute temps in this test are meaningless. The relative temps are what to consider. If you use a bigger fan, or more or less CFM, the absolute temps will change.....but the relative differences remain fairly constant. In a case with different configurations, your results may differ from this test. But it seems that an upward airflow through a vertical Ninja gives the best temperatures. A duct certainly helps out.....as does a fan direction that makes use of convection.

Oh....what I learned for sure: this Ninja will not run fanless with a P4-2.4 no matter how it is arranged, using CPUBurn to max it out.

Thats some good testing Bluefront, roughly whats the airflow in CFM of that YL fan at 1925rpm.?

Also, something else to try would be to put something all the way down the hole in the middle of all of the fins and re-test (only one test of course). This would of course reduce airflow, but force the air nearer to the outside of the H/S where the fins are hotter. Different sizes and shapes could be used if you have any success. For the first and easiest test a simple sheet of paper rolled up and poked into the hole would be great as it will try and unroll itself thus ocupying all of the space. The major drawback to this test that I can see is that the heatpipes on the opposite side to the fan will get hardly any airflow, but again if you have success you could always experiment with different sized ducts.


Andy

nice work

my 2 cents though, would you consider your set up with a passive cooled video card and north bridge (something like hr-05), i'm wondering in that case, the hot air from video card/nb would bring down the advantage of top fan blowing up.

also, would you consider the set up inside a case with the case fan extractinig out? (with or without duct on ninja), i believe ninja would do fairly well fanless (on itself) with the case fan kicking.

That YL (Jabtech) is rated at 25CFM at 12V. I only tried a few different ducts. That one is very simple. A more complicated duct with much experimentation, would probably return slightly better results.

This test shows which are the better airflow setups. But of course if you attempt to add stuff to the configuration like hot cards, or a different NB heatsink, the temp readings would probably be affected. I took NB temp readings also during the testing. The best NB temp was with the fan on the bottom blowing upward. I didn't try the tests with the fan on the bottom blowing downward.....that no doubt would reduce the NB temps.

There are countless possible configurations one could think up. My best case setup has the rear case fan blowing inward at the Ninja, using a similar short duct. But this sort of test removes most of the variables, and focuses on the CPU temp only. I'm going to duplicate some of this test using another similar heatsink......stay tuned.

that's a very thorough piece of work!

looking at the no.s and graphs I would hesitate to draw any firm conclusions from the data. All of the fan results are within 3C (~7.5%) of each other and the graphs show considerable overlap (best graph is 44-47C, other is 45-49C) each with considerable variation and the resolution of the thermal sensor appears to be 0.5C aswell. If you repeated each test 3x and did a statistical analysis (boring I know), I'd guess that the within sample variation would be great than the between sample variation (the effect of the variable being measured).

Would it be worth repeating with a lower airflow fan? That might pull out bigger differences in orientation. Would be interesting to see the best and worst set up tested inside a case too, to see (if possible) how being in a closed case affects the results. Good work though!

Boring series of tests for sure.........and I did repeat most of the tests a few times to be certain. I was mostly interested in the effects of convection on the test results. The testing confirms (to me anyway) that the best CPU temps make use of convection currents, rather that fight convection.

That is a Biostar board that I've never had in a case, that I got very cheap. If I hurt it during all this testing, I wouldn't much care. This particular board has some unusual sensor readings, with many peaks (variance) during a graphical read-out. Some boards will show a graph that is almost a thin line going up/down. The program doing this graphical test is Sensor's View. Speedfan cannot read any temps for some reason. I had the program set to record a temp reading every three seconds (max speed).

I know that the temp results don't vary much between the various configurations with this series of tests, using this particular fan/speed. But....if you are using a setup with much less airflow, and the resulting higher temperatures, it would certainly be a benefit to use the best possible airflow configuration. That is what I am trying to root out with this test. I think Felger Carbon calls it "data mining".

And when you stuff everything in a pre-made case......your own results will vary. I'm going to attempt to build another custom case that makes the best use of these and other test results. I'm hoping there is that "perfect" configuration out there, just waiting to be built.

Well I did a little more testing with the Ninja......using the best configuration. I did block off the center hole in the Ninja, using a stiff paper tube. Presumably this would force the airflow more toward the outside edges(hottest part) of the fin pack. Didn't make any temperature difference I could detect.

Then I added a full-sized AGP card (fanless Radeon AIW 7500). This card is two inches below the Ninja. Presumably this would somewhat restrict the airflow into the bottom of the Ninja, and pick up some heat off the back side of the card. Again I could measure no temp difference. This particular setup with the fan on top blowing upward is working well.

Next I tried using an NCU1000 on the same board......



I set this board vertical and ran the setup fanless. Since the Ninja wouldn't run 100%, I didn't think this setup would either. So I loaded the CPU 25% using SiSoft Sandra. Running like this the Ninja maxes out at 49C. The NCU-1000 runs at the same temp like this. This is with a heatsink with maybe 1/2 the fin area of the Ninja. But it was designed to run fanless....and so it does at a lower CPU usage, like the Ninja. This heatsink took five minutes longer to reach 49C than the Ninja, so maybe it is cooling a little better.

Is the Ninja the best heatsink for running fanless? Maybe not......

What about testing with partial duct vs. full duct

I am curious as to why you used an 80mm fan w/ a duct instead of a 120mm fan?

I recently moved from a side blowing in configuration to a bottom blowing up to give myself access to another RAM slot. If there was a temp difference it was +/- 2 or less. I also have a power supply fan (intake) near the top of my Ninja.

I am real tempted to do some ducting now.... it totally makes sense.

Couple of things here.....this test is only using a P4-2.4. Even at 100% usage, it obviously doesn't need a bunch of airflow to keep it cool. A ducted very quiet 80mm Yate Loon is all that's necessary.

Also it's been my opinion since I got my first Ninja, that much of the airflow from a 120mm fan mounted directly on a Ninja.....is wasted, because of the over-lap. I am under the opinion that a properly setup 80 or 92mm fan is all that's needed. Makes for a less cramped installation.....no blocked ram slots, etc.

Anyway....about 25CFM at 100% is enough. Provided of course your setup is tweaked for the best airflow.

What do you mean by a full duct? I first tried a longer duct that covered the entire sides. This shorter duct worked better. Tweaking this particular duct might gain a little cooling....not much IMHO.

Shameless self-promotion: I posted a Scythe 100mm fan mounted on a Ninja using round shoe-laces. Lotsa close-up pics! Search "kitchen", because I used materials commonly found around the house, and I keep my shoelaces in a kitchen drawer.

Or search Ninja, shoelace, Scythe, 100mm with all terms needed for a match. I think "kitchen" was in the subject heading.

What I really need is a half-length duct that uses say a 92mm or 100mm fan and also holds the fan a small distance away from the fans.. that would be super quiet and efficient.

The duct probably has the effect of increasing velocity through the fins and also you don't lose the air out the sides. I'd imagine that at the far side on the fins you've already lost perhaps as much as 50% of your air out the sides.

This also might speak to the effectiveness of the Thermalright Ultra 120, it's wide but not very deep. Also perhaps the overall quality of the Thermalright is a little bit better, except perhaps the base.

I wonder if those CPU fans with the plastic ducts on them could be improved by hacking them up a little?

There are very few heatsinks that I know about, that have the ability to run fanless. The Ninja and that NCU-1000 are the two I own that will fit a 478 board, and can be oriented for the best airflow. That's why I tried these two....that and I already had them.

Having an ability to run fanless, also gives them an ability to run better in a very low-airflow situation. Other heatsinks no doubt would cool better, but only in higher airflow setups.....read louder. In order to make a valid test result, I have to make the different experiments as equal as possible.....same board, same CPU, same ambient, same fan/speed, etc. I started out trying to find out what was the best airflow setup for a ninja. After I was satisfied with the best setup/result, I applied it to a very similar heatsink, the NCU-1000. Amazingly, the overall temperature results were almost identical.

Next step in this little experiment....try to tweak that little duct to knock off a degree or two. I'll use the same 80mm fan, so I'm not entering another variable into the mix. Moving the fan slightly away from the fins will probably do better, so that's the next step.

Felger Carbon's 100mm fan would probably fit this Ninja better than the 80mm fan I'm using, but IMHO, the relative test results would have been the same. Look.....this is an easy experiment to set up, if you have the right pieces, and the time. I'm curious if someone else could come up with a different conclusion about this Ninja and the best airflow.

One day (dont hold your breath) I will try some ducting myself, but I am very lazy.

I am running my Ninja with a Nexus 120mm blowing to the back of the case through the front, which is being controlled by ASUS Q-Fan via the BIOS, and the only case fan is another Nexus 120mm at the back of the case (blowing out of course) which is also controlled by Q-Fan, and also has to cool my X1950 PRO with an AC Accelero S1.

The top fan hole is blocked off, the top HDD cage has been removed and I have a few slot covers (spines) missing for Graphics card airflow, which in my case is a low priority.

There is a lot for me to play with, but I think a simple duct will take up to 5C off of my CPU temps.

I have in the past tried my Nexus on my ninja blowing up through the bottom , which didnt work well for me (I now have a different mobo and Graphics card but I still dont want a top case fan so it will still be pants as the air will have to take a sharp 90 degree turn to exit the case). Everyone has a different rig so will have different results.


Andy

I haven't done any formal testing myself with the Ninja. But I can say I ran a Ninja with no fan attached to it on an x2 4400+ with a 120mm YL exhaust in a p180 case. The highest temp after hours of load was about 49C. I also have a 2.8ghz P4 Northwood using a Ninja with no fan attached and just a Nexus 120mm for exhaust in a Sonata case. Temp for this doesn't break 53. Ambient in the house is usually around 24-26C.

I never messed with ducts with the p4 setup, but in the p180 with the 4400+ I did make a duct from the back exhaust to the Ninja. I didn't see the duct change cpu temps, but I did notice a 3-4C drop in PWM temps. With the Abit board I had, these modules were located between the socket and back of the case. In the p180, I also had the top fan hole blocked.

Ok...here are the first exclusive photos of the Bluefront engineered "Almost Silent+X Super-Duct".









It's made to be adjustable. You can install the fan to either blow or suck. The one outer flange of the Yate Loon was cut off. The hole in the duct was sized so the fan is a tight fit. You can raise or lower the fan about 15mm (for tweaking purposes). If you mount the duct on the video-card side of the Ninja, with the fan blowing into the Ninja, the NB heatsink gets a fair amount of airflow. Plus the back-side of the video card gets even more airflow. The fan in this orientation is about 25mm from the card (on this MB anyway. I made the sides of the duct shorter than the first duct in the original photos. This gives you the ability to experiment with the length of the duct (by closing off some fins by various means....like tape). The whole thing just slides off/on.....and is held in place by the three heatpipes that poke through the one side of the duct. This thing mounts the fan blades about 5-20mm from the Ninja fins......and it does quiet the airflow somewhat. Performance gains? Not much over the original duct, just like I suspected. Maybe one degree. This duct seems to work a little better with the fan on the bottom blowing upward.

Duct making is easy.....finding the best duct design is not.

Hello Carl,

Looks good! Any difference in the temps with this duct; with the fan in either direction?

I haven't tried this out in all the various configurations like I did the first duct, with that long series of tests. I took the best results of the first tests with the fan on the top blowing upward, and the fan on the bottom blowing upward.....and tried to get better temps using the same fan @12v. I raised and lowered the fan through the whole adjustment range, and experimented with closing off some fins with strips of tape. (you can see the tape in the photos).

Unfortunately.....I got very little improvement in any orientation over the first hastily made cardboard duct, which is the same design as this one, less the easy adjustability. This duct gives a little clearance between the fan and the fins, which the first duct did not. I got a little improvement of maybe one degree.

Keep in mind that I'm basically only concerned with CPU temperature with these test results. The other board temperatures are also affected by these airflow tests. So when you finally setup your computer, you have to make many decisions about temps and airflow, the temps of everything, not just the CPU. But you need a place to start. This sort of test on the very common, very good Ninja, gives you much to consider.

Here's the next set of tests. I was interested in what a slight airflow increase would do using the same duct. This particular duct can be enlarged at the top to hold a 92mm fan......exactly like the 80mm. So I tried a 92mm Nexus, which is rated at 27CFM @12V. I also tried a 92mm medium-speed Yate Loon, rated at 42CFM @12V. A 100mm fan would not fit in this duct.....so this is the last major change to the test.



Here's the results:

1. 45C....80mm YL on top, blowing upward.
2. 44C....92mm Nexus on top, blowing upward.
3. 45C....92mm Nexus on top, blowing downward.
4. 44C....92mm Nexus on the bottom, blowing upward.
5. 43C....92mm YL on top, blowing upward.
6. 43C....92mm YL on top, blowing downward.

Conclusions......difficult to figure. The Nexus lowered the max temp about one degree. Understandable because the CFM only raised 25>27CFM. But the 92mm YL only lowered the temp two degrees. This is with a a CFM rise of 25>42CFM over the 80mm YL.

This can mean several things.....the cooling potential of the setup is limited by the design of the duct. Or.....the cooling potential of the setup is limited by the design of the Ninja itself. In any case, this certainly proves there is a "sweet spot" with the airflow through a Ninja. And if you go past that spot by an airflow increase, you get an increase in the noise level, but very little temperature decrease. The law of diminishing returns.....

If you remove the duct and put that 42CFM fan on it, the answer should make itself known...

This might be a stupid question, but do you happen to have the stock 120mm fan from scythe than runs at 800 rpm, and could you possibly test that w/ the stock mounting?

Also, how does the heat output of this CPU compare to a Core 2 duo?

Lastly, what kind of noise were you getting from the 92mm nexus and 92mm Yate Loon? I'd imagine the yate loon at 50% more airflow would be much louder, so hard to say if its really better overall compared to the quiet nexus.

I got two Ninjas with the OEM fan.....this Scythe 120mm fan ran about 1200rpms. I don't have these fans any longer to test them in this setup. But I did just do three more tests with a fan sitting on the top of the Ninja, without the duct.

Yate Loon 80mm.....25cfm.
Yate Loon 92mm......42cfm
Yate Loon 120mm.....70cfm.

Amazing results......here are the temp changes with the three different fans. Sucking or blowing returned the same results in this configuration today,

1. 80mm YL on top blowing up. no duct.
2. 92mm YL on top blowing up, no duct. -1C.
3. 120mm YL on top blowing up, no duct -2C.

Almost three times the airflow using a medium speed 120mm YL, only lowered the temperature two degrees over the low speed 80mm YL. Since the 120mm fan is slightly larger than the side of the fin pack, some of it's airflow is wasted. And IMHO, as you increase the airflow, the benefits of convection are greatly reduced. More to think about.....

Oh....these mid-speed Yate Loons are unacceptably noisy at 12V. The low-speed 80mm YL is ok, as is the 92mm Nexus. But that should be obvious. You can easily hear that result.

Good work Bluefront! Now what does this tell us? There is a bottleneck somewhere.

If you increase the heat source, does the "sweet spot" move?
If you decrease the heat source, does the "sweet spot" move?

This will tell us whether you have reached the limits of the heatpipes or of the fins. If you can add more heat, and have the sweet spot stay the same, then the heatpipes have additional headroom and the fins are the limiting factor. Though I don't think this is the case, as more airflow should cause the fins to be able to dissipate more energy. I suspect it is the pipes. In this case, a lower heat source should give much better efficiency.

could you overclock the CPU or underclock to change the thermal load?

Well it looks like I've reached the end of the easy part of these tests. I could do all these tests over with a hotter CPU, but I suspect the relative differences in the various configurations would be very similar. I also think the fin area is not the problem. That NCU-1000 with 1/2 the fin area performed very similar to the Ninja. What we'd need to test would be a heatsink with a better heatpipe setup, and a similar fin arrangement. Even then, any results would be suspect.

Looks like the airflow "sweet-spot" on a Ninja is right at 25-30CFM. But I'll have to run some more tests at a lower known CFM. I might be able to find a fan to do this......how about a 70mm YL. ?

Oh...this particular CPU in the testing (P4-2.4) draws about 60W.......and I have several machines running with these things. I've always found them relatively easy to cool silently.

There are no adjustments on this board to do any over/under clocking...so that's out.

Nobody in his right mind would run a 70x15 YL on a Ninja. It's rated at 22CFM at about 3000rpm. But this airflow is squeezed down into a 70mm circle, and no doubt has a different effect on a heatsink. Anyway.....I think this thing is blowing more than 22CFM. On the top of the duct, blowing upward, this fan runs the CPU at the same temp as the 92mm Nexus, which is rated at 27CFM. Can't be right. It feels like it's blowing more air than the 80mm YL. Sounds like a Banshee..... Scratch this test.

Here's another set of tests running the Ninja with various levels of CPU usage (using the SiSoft Sandra benchmarking program). I ran the tests at 12V and at 5V. The ambient temperature dropped to 25C, so don't compare the absolute numbers to the previous tests. The temp differences in this set of tests are what to consider. Here are the fans used...

70x15 YL...........D70SH-12B......................22CFM
80x15 Zalman....RDM8015B........................about 22CFM
80x25 YL............D80SL-12.........................25CFM
92x25 YL............D90BM12..........................42CFM
92x25 Nexus.......DF1209SL-3......................27CFM
120x25 YL...........D12SL12-C.......................45CFM
120X25 YL...........D12SM-12........................70CFM

12V test.............Idle....50%....75%....100%

70 YL.................30C....35.......38.......40
80 Zalman..........31......36.......38.......41
80 YL..................31......36.......38.......40
92 Nexus.............30......36.......38.......41
92 YL...................30......34.......36......38
120 L YL...............30......35.......37......40
120 M YL...............30......34.......36.....38

5V test.................Idle....50%....75%....100%

70 YL....................31......38......41........42
80 Zalman.............32......37.....40.........41
80 YL.....................34......40.....42........45
92 Nexus................32......39.....41.......44
92 YL......................30......36.....38.......41
120 L YL..................30......36.....39.......41
120 M YL.................30......35......37......40

All these tests were done with the fans just sitting on the top of the Ninja, blowing upward. At the 5V setting I tested the 92 Nexus, and the 80 YL, with the wood duct shown in the last photo. The duct consistently dropped the 100% temps of each fan about 2 C. At 12V the effects of the duct were slightly less.

My conclusions from these tests......A 92mm fan seems to be the ideal size on a Ninja. The temp differences between the same fan at 5V or 12V is minimal. It seems useless to hang a 120mm fan on the Ninja.....perhaps because of the overlap. Using a duct helps the temps slightly....more so at low voltage. When picking a fan for a Ninja, the quietest 80 or 92mm is your best bet.

Higher ambient temperatures might change the relative results slightly....but not much. The 120 medium speed YL started every time at 5V, but just barely. The sweet spot for a fan at 12V......about 27-40CFM. Take your pick. At 5V they all were quiet, except the 70MM YL, and the 80MM Zalman.

My favorite setup for this P4-2.4/Ninja ........the 92mm Nexus using the duct, the last photo. It looks like this setup could easily handle ambients at least to 30C. What voltage to run....depends on the ambient, and how hot you care to run the CPU. I'd probably use a NoiseMagic fan controller that goes from 5-12V.....acceptable noise levels, acceptable temperatures. If your MB fan control can do this, so much the better. Under normal usage, almost any quiet fan @5V would be ok.

YMMV..........

Good stuff. Thanks Bluefront!

Thanks.....I certainly hope anyone reading these results can make use of them. These results partially explain why I have had so much trouble judging different case airflow setups when using a Ninja. This heatsink does not need much airflow to remain in about the same temperature range.

It's strange to see an 80mm low-speed Yate Loon, which hardly blows enough air to feel with your hand, maintaining almost the same temps as the medium speed 120mm YL.......that fan blows a bunch of air. I suspect Scythe bundles a Ninja with a 120mm fan for sales appeal only. It certainly doesn't need that much airflow.

Oh....one more test at 25C ambient. No fan at all, using convection only.

No fan.......Idle.....25%.....50%.....75%.....100% CPU usage.
.................36C....43C......47C......52C......56C...................

Don't try this last test with any heatsink......except a Ninja.

Sweet. How did the air feel coming off the top of the heatsink? Could you feel the air rising?

It feels like there is a little fan on the bottom blowing upward. I should attach some strips of paper or something to show the airflow.....