Good to see that Nexus still keeps up with the other contenders.

The testing rig seems pretty advanced but hopefully it will produce more reproducible results.

Is lower temp-fans more suitable for heat-sink-cooling and high FPM more suitable for intake/exhaust?

We don't differentiate between heatsink and case fans. In both applications, you want the lowest noise for a given degree of cooling. This should mean that the Temp Rise vs SPL figure we're going to be focused on will apply in both apps.

btw, we need some help with a bit of coding for our fan articles. Please see this: http://procooling.com/index.php?func=articles&disp=131

That type of interactive chart would be wonderful for us to use for measured fan data. Can anyone help us develop a similar function?

Did Thermalright send a review sample of the new Thermalright Venomous X (the TRUE successor)?

Great explanation Mike, thank you !! and a very informative read too !

Same question as bozar. Is it true to say that :
Higher FPM at same SPL is better for a case fan (Noctua in lead)
Lower °C rise at same SPL is better for heatsink fan (Nexus in lead)

right ?

Wow! I was just toying with the idea of swapping the Slipstream for Nexus, but looks like it bests Nexus at the speed I'm currently using (700RPM) by a tiny margin! And they're practically equal at other speeds Great news, thanks!

PS.
It is really hard to find Nexus products in Poland, whereas Scythe is widely available and relatively cheap.

Nexus fans still rock. Too bad I can't buy them here
At least quietpc.com still sells them.

Correct me if I'm wrong, but did you mean logarithmic? We're talking about diminishing returns, right?

Nomnomnom great article.
Looking forward to more fan reviews

Interesting bit on heatsink selection. Do less tightly spaced fins lead to the same cooling performance as tightly spaced fins, but with more "fan selection" leeway?

I previously thought that maybe the tighter spaced fins would lead to better cooling with higher rpms/pressure fans, but maybe that is not the case.

afaik, logarithmic and exponential functions are essentially the same -- the former is a special way to express the latter. The curve I refer to is exponential if you look at what happens to the airflow -- keeps going up indefinitely while the temperature barely moves.

Hm...okay. I don't pretend to know a lot, but isn't an exponential function the inverse of a logarithmic function? I guess if you look at it one way, yes they're the "same" so to speak, but they gave them separate names for a reason, right?
Your graph seems to look like a log graph to me: it asymptotes as you increase CFM, with temperature dropping ever slightly as you go on.

I'd like to also see S-Flex, Noiseblockers, Noctua P12 (in order of descending priority).

The article is more of a showing of the new fan testing methodology using common quiet computing fans, rather than an article of fan comparisons. I'm sure they'll get around to testing other fans soon.

Huge improvement!!

One comment though: Given your new methodology, I don't think testing at standard (and carefully calibrated) fan RPMs is all that important. I'd rather you chose noise level or temperature rise as the independent variable.

Then your tables would directly address the critical question: for a given temperature rise, what fan is the quietest? Or alternatively, for a given noise level, what fan cools the system most efficiently?

Wow...

Thats all I have to say, really. Kudos to Mike for this amazing article. So short, but yet so good

Agreed -- simpler is better. All the work you don't see is what was *not* used -- like the sculptor's chips.

A benchmark achievement, Mike!

Great test system and article Mike.

I agree with the post above about using spl values or temp rises maybe (delta T/SPL) as the independent variable. Other than that, great to see the FPM/CFM numbers, this helps balance out fans in a case. As you don't really want too much in the way of pressure in a case, too much negitive, and you pull in dust, too much positive and your fans don't move air well. There are very few small DC fans that handle high(few mm's of water, 1-3 inches are common in HVAC) pressures well. The other thing to point out about fans is that larger doesn't mean quieter even if the RPM goes down for the same amount of air. The tip speed(speed of the edges of the fan blades) may be higher, and the tips are what causes most of the noise, along with the trailing edges.

I wish heatsink manufacturers would provide pressure drop formula/curves, and Fan manufacturers would provide fan curves(rpm/power/cfm/pressure drop) then in your case it would be a relatively simple mater to test a heatsink. As a CPU is basically an electric resistance heater. You know the power in, you know the starting air conditions(temp/pressure/humidity), and the ending air conditions, so it's pretty simple to see how much energy you transferred to the air. So if you vary the input wattage, and measure the amount of wattage disapated by the heatsink at a constant CFM, you can find out the maximum amount of energy that a heat sink can dissipate at that input air temp and CFM. This isn't very helpful unless you run the tests over a range of inlet temps, and CFMs. with enough points, you can figure out how an airflow change will effect cpu temp. It's way more work than is reasonable to expect, and i bet manufacturer already has this info. Getting it is a whole other matter.

Could you do FPM for PSU fans as well, preferably with the PSU case in place?

As for:


I would say yes and no, really if you are going for quiet, then you want the lowest SPL that will move enough air in to dissipate the energy you need. You can think of a computer case as an electric resistive heater. You have an allowed temp rise,max operating temp of the components, with the worst case intake temp, so you need a set amount of air into the case to remove that heat. Now that you know how much air you need, you can find the lowest SPL that will move at least that much air, more air isn't a huge concern as that leads to lower temps. You will want some margin in there, as some of the air will simply be exhausted in most cases without cooling much of anything. So a higher FPM/SPL ratio isn't always better if you can get the FPM you need with a quieter fan. 30000FPM/100SPL isn't as good all the time as a 2000/50SPL, if the extra FPM is "wasted".

Sorry for the wall of text, the engineer in me is coming out. The big problem with most of the above is simply not having the data on the parts around, fan curves, pressure drop of the case. Thanks to SPCR we know roughly the efficiency of heatsinks(standard fan, and CPU/heater) so that helps.

I totally agree with nicke2323 - I think hitting a specific rpm value is less useful. I suppose with enough data points we can get the data we seek by regression, but I think that data presented this way will be easier to draw useful conclusions about.

That being said, with so many of us using the 5-7-12 v trick, what about measuring the values at each of those voltages - that way we can get an idea about what our system sounds like. Additionally, it is easier to just do a 5v mod then to buy a fan controller and precisely turn our fan to 1080. I think more people would find this additional information useful (but you obviously need to keep the other data for an apples-to-apples comparison).

Oh, and while I am asking for the world, what about turning down the wattage on the block down to levels that many people use in their systems here? Something closer to 40-60 watts - would this have any effect on the data? Might be an interesting experiment, and once you have the fan all rigged up, it doesn't seem like it would be hard to adjust the heater.

I have to say though, that MikeC and the rest of SPCR do an amazing job; this revision is no exception. It is probably one of the most educational and well thought out "review" sites around. Kudos. /flattery

Adam

What they're measuring is fan noise at a given controlled load, not HSF performance.

Great article, thanks.
Does anyone have any theory about why higher FPM can actually result in higher temperature rise (as seen for Noctua)? I really struggle to find any plausible explanation for that.
Do you think the measured higher FPM can acutally mean lower total volume of air? This would require rather uneven flow of course... I am wondering if speed readings would be different if the vane was connected directly to the fan meaning all the air would have to pass trought the vane... or at least if the vane would be insde tube connected to the fan. This would avoid error in speed reading due to different intake airflow "speed field" shape...
I think I would have to draw pictures to explain what I mean.

EDIT: Behold in awe at my mad mspaint skills!

The cuved lines mean to represent some given speed of airflow, X FPM. I am trying to say some fan may draw air from larger angle affecting FPM measuremen. I am not sure if this is actually possible.

If you forced all air to go through the vane, perhaps through a funnel of sorts, I imagine it would create artificial impedance.

Thanks for all the feedback everyone, both positive and critical. The piece was a long time in the making -- the thinking through, the experimentation, the writing, the revisions.... many cycles of these things, and it's still probably not quite right.

Taking measurements at specific temp rise points was considered but this is no easy task. The temp cannot be obtained by setting a dial; the fan speed has to be roughly adjusted, then closely monitored, tweaked continuously to reach a given point. This process will probably double the time need to test fans compared to setting RPM, which can be done in under 1 minute. But we can give it a try and see if repetition makes us more efficient at it. We want to keep the lab time needed for each fan as short as possible -- we've had issues with slow editorial production from day one.

The other method of using SPL points is much quicker -- just turn on the digital sound measurement system and tweak the fan controller. But this means all the testing must be done in the anechoic chamber, and there is an issue with heat building up. While our use of temp rise takes into account variations in ambient temperature, for consistency's sake, we think it best to keep the ambient in a narrow range. Again, this is something we can play with to see if some efficient testing procedure can be devised.

Finally, I'd like to repeat my request for help with an interactive chart like this: http://procooling.com/index.php?func=articles&disp=131 If you haven't gone and taken a close look at it, you'll see why it could be so useful for us. The above discussion of exactly where the data points are taken will not matter if we can generate a SPL/temp rise curve for each fan and post them all on one graph. (at least one graph for each fan size)

In theory, I suppose extremely high air volumes could actually add heat to the system through friction. It's apparently an issue in supersonic aircraft. But I'd imagine a fan blowing that hard would blow the setup to pieces before it actually reached speeds that added heat that way. Anybody willing to give it a try with, say, some delta fans in a push-pull configuration?

Yes it would. I think it shouldn't be a problem as the heatsing resistance is also arbitrary - depends on chosen heatsink. Only issue I can think of is this would create resistance BEFORE the fan. This is something you will probably not encounter on real system and it may affect the performance I guess.

Very nice, Mike. I look forward to the next round of results.

Do you intend to do any testing at different airflow impedances? Even 1 review with 3-4 fans would provide useful information for people considering unusually open or restrictive heatsinks.

Yes, it does. This was one of the many things tried that were not included in the final article. Neil Blanchard hinted at this in his post; he helped critique an early revision of the article.

A semi-flexible cone was made that had one end which fit tightly around a 120mm fan frame, and the other end tightly around the anemometer vane. The FPM measurement fell by 20-30% or even more at almost every speed with every fan.

The inconsistent relationship between measured FPM and temp rise for various fans showed up not just with the Ultra 120E heatsink, but also with the Megahelems. I do not know how to explain it... but it reconfirms my experience that fan airflow and cooling don't have a very predictable relationship -- at least not at these fine detail levels.

Good initiative Mike. The only thing I'm concerned is the TRUE choice for the reference heatsink. Its design is a bit outdated since they got to rev.C by now and they released a new high end model, the Venomous X.

Secondly the TRUE might give false results because of the small depth and some high pressure fans would underperform. I'm not talking about the Ultra Kazes because I know they are crap but I'm really curious to see how a Delta will fare on this platform.

I would not be surprised a bit if higher FPM resulted in no decrease in temperature rise. But for it to result in actual increase... this seems really strange. I am not trying to criticize the test setup or anything; I am just puzzled about how this is possible.

EDIT: I hope I am not trolling too much. Stop me if I am.

You must remember that when any fan is examined by itself, cooling follows FPM more or less predictably. It's when you compare FPM vs cooling in different fans that it stops being as predictable. ie, 285 FPM in Noctua gives worse cooling than 180FPM in Nexus. This suggests something is different about the way the two fans move the air -- one gives rise to higher velocity reading in our anemometer but doesn't actually provide as good cooling, which suggests lower real airflow. (That Noctua gave high readings with every anemometer we've tried)