Great points made on the philosophy of SPCR tests. I think I made a similar point somewhere in the forums when someone was wondering why SPCR removed the supposedly better stock TIM and used Ceramique. Is this new Arctic Silver Lumière something new or just for test labs? I know that many if not most of the older reviews all used Ceramique . . . does switching to Lumière change things much?

SPCR seems to be one of the only sites that test heatsink fans and acutally try to control variables such as the stock fan and thermal interface material constant for better test results. However, for most builders what other sites do will probably be adequate since most people will keep the stock fan and TIM.

Hi, nice article explaining SPCR's point of view on heatsink testing and methodology. One significant benefit of using a reference fan is that you easily compare performance at a given noise level, which in fact to us is the most interesting comparison. A Nexus fan will sound similar at a given speed on whichever heatsink, of course the size differences doesn't help and there one needs to look at the dB(A) readings and listen to the sound samples.
I've be researching my next heatsink and ended up making a table of °C/W at each noise level from 16 to 24 dB(A) to help me see how various ones compared despite using various fan sizes and some of them not using Nexus fan at all. My Table clearly shows the Ninja to be significantly superior but I only included it as a reference as I'm looking for a heatsink <130mm tall. (Neither the Ninja nor any other 120mm fan tower is) Interestingly it suggested the "Butterfly" heatsink to be the best 92mm fan'd heatsink, although it wasn't tested with Nexus fan so noise quality would need to be looked into and the fan is very non-trivial to swap should it fail / be too loud. I've more or less settled on Zalman 9500 as it comes a close second (in my table) and my preferred supplier is doing a very good price on it and I like it as an elegant design
Guess it brings me back to the beginning, there is no one "best" heatsink, all the factors need to be considered and depending on each individuals weighting will depend on their "best" option.
Thanks to SPCR there is data that we can use, not to mention the pool of knowledge in the forums, to make informed decisions from.
Seb

The article claims that the 92 mm Nexus does 1380 rpm at 12V, which I find hard to believe. I use(d) three 92 mm Nexus fans, one in my previous system on XP-90 and two in the front of my current P150, and at 12V they do ~1600 rpm (according to motherboard sensors) and are noisy. Under 1000 rpm however, they still move some air and are quiet.

If i can believe what my mobo reports, my 92mm nexus runs at 1150rpm. No it seems to be rising very slowly... I've had it laying on its side for weeks without using it at all. Now after running for a bit more than five minutes it spins at 1370rpm.

This table is data gathered an old SPCR review:


The new article gives us 0.21 °C/W vs 0.22 °C/W in the old article for the Ninja. Not much of a difference.

The new article gives us 0.33 °C/W vs 0.27 °C/W in the old article for the XP-120. This is a much more noticeable difference.

Note the data is from two articles not three. In other words the "old" number for each HS is from the same test bed.

Is there an error in the calculation in the old or new article or am I misinterpreting of the comparison of measured °C/W on differing test beds?

Or is this just sample variance for the XP-120?

As far as I know, Lumière is an OEM only product, which makes it hard for the general public to obtain. It's advantage is (supposedly) that is cures in half an hour, making it easier and faster to obtain good results. I'm told (by Arctic Cooling) that it performs similarly to Ceramique, but we didn't do any detailed comparisons. I very much doubt it will change things more than the switch to a new processor & MB combo.

Why do you find our RPM measurement hard to believe? That's what we measured, and it's within the 10% variance that we allow for fans. FWIW, the Nexus 92mm is rated for 1500 RPM, though I can see them reaching 1600 RPM if the motherboard supplies a high 12V output. I'm sure this is just sample variance rearing it's ugly head... but I'll know more when I get a chance to test a few more samples.

I suspect that the difference between the Ninja and the XP-120 is the test bed that you're comparing. IIRC, the Ninja was originally tested on our P4-478 board, while the XP-120 was tested twice: Once on a test bed away from SPCR, and once (with the Nexus) on our P4-775 board.

On the other hand, I'm still not 100% happy with the consistency for the Ninja & Ultra-120... I may try re-testing them to see if things change.

so maybe another test with Ninja, Infinity, Ultra-120 and Noctua NH-U12, to clarify this issue?

I admit that your measurements could be 100% accurate, due to sample variance and a low 12V. But shouldn't SPCR try to test a sample that's as close as possible to the 1500 rpm specification? IMO my 1600 rpm is loud-unacceptable, while 1380 rpm is loud-acceptable, for people that don't control the fan speed (I do), so the difference could be significant when deciding to buy such a fan.

My guess would be that the XP-120 kept up decently with the 68W (TDP) P4 used in the old test rig, but this 78W (max measured in use) Pentium D might be a little too much for it to handle (without extra airflow, anyway), while the Ninja, being built for 100W+ Preshotts, can just breeze along.

floffe, I point you to http://www.silentpcreview.com/article646-page3.html

which has this content from 2006

Test Platform

* Intel 520 processor (P4-2.8 Prescott, 1 MB cache, 800 MHz FSB in 775 casing). The Thermal Design Power is 84.0W. The Maximum Power, as calculated by CPUHeat & CPUMSR, is 100.76W.




Are you saying that article misquoted the numbers from a previous article on the 478 socket platform?

As far as I can tell from the OLDEST article (that I have not linked to as it is irrelevant to our discussion) the XP-120 was not originally tested with a what is now considered to be a reference fan nor was it tested at 9 volts. The world was a different place back in 2004.

Is the TDP difference between the Pentium D 950 and the Pentium 4 520 that big that these numbers are so out of wack? Is it sample variance? Was the 1°C difference in ambient temps between the articles that significant? Is it something else?

I'm just trying to make sense of it all. Either there is something I don't understand or an error was made in a previous article. I'll take either answer so long as I learn from it...

If I understood the results correctly - the Zalman CNPS is the best heatsink ATM? If so - just wanted to remind you to insert it in the recommended heatsinks article.

Quoted from the end of the article:

My sympathies are with you - that is indeed a noticeable diffrunce. But every fan spec I've seen gives at least a +/-10% tolerance on the fan RPM. And that's just for sea level, my 4100', or Denver's 5000', or Santa Fe's 7100' produce faster speeds. Fans aren't precise devices!

Yes, thanks - somehow missed that.

IMHO, there is little sense comparing heatsinks on different voltage levels when the fans are not the same/similar brand - actualy equal noise level & equal CFM levels would be much more telling.. the table on the last page is thus a bit misleading concerning Zalman pure _heatsink_ performance but more as a fan/heatsink combo.

Other thing I noticed:

Nexus 120mm:
12V 47 CFM

and the Zalman has 92mm fan by design - so I doubt if airflow remark

really "holds water".

It would be great to see Zalman tested modded with a Nexus fan.

To clarify my interest in Zalman just in case anyone would be interested - the results sound like it has very good cooling/noise performance per gram of weght (ofcourse, "noise" with nexus-modding in mind)

I'd prefer to avoid mounting almost 2x max. AMD recommended heatsink weight (ie. Ultra-120), if possible, to be able to freely lug the computer around.

Please keep in mind one relevant difference in how °C/W is now calculated: Previously, we used a calculated value to estimate how much power was consumed by the CPU. This time, we've measured the power it consumes; that 78W figure is not TDP — a theoretical design number. It's the actual amount of power consumed by the CPU & VRMs through the AUX12V connector.

@speedy: The Nexus 120mm runs at ~1000 RPM at 12V. The Zalman fan runs at ~2600 RPM. I can assure you that, not only does the 92mm Zalman fan produce more airflow, but the higher RPM translates into higher air pressure that is better for cooling. If you want to see how the Zalman performs with a (92mm) Nexus, read this article: Modding the Zalman 9500 Heatsink.

In any case, the Zalman 9500 will never get an SPCR recommendation. No matter how good a cooler it is, it simply isn't quiet enough.

this testing method should be taken one step further and all heatsink bases should be lapped with a good finish until 1000grit sandpaper, and all heatsink should also have a similar mounting method so pressure will be equal amongst all heatsinks. True, this would be changing the manufacturer's intents, but many people lap the bases of their heatsinks to gain those extra 2c, and/or mount their scythe innfinity's with the same pressure as the ultra 120 and gain an astounding 5c. I figure this website is for "hardcore" enthusiasts and im guessing they wouldnt mind doing such modifications to their heatsinks IMHO. In a way, I think testing the potential of a heatsink is more important than the "out of the box" performance.

dhanson: my bad, but if you'd pointed to that article I'd probably have seen it. As it was I looked back for an earlier version of the heatsink testing platform, but when I didn't find one I used the numbers that I remembered. And yeah, this CPU has probably still beats the crap out of the earlier one (130W TDP vs 84WTDP, 100W max). The 15W difference in TDP between the P4 i quoted and the one you did is a lot less than the 46W difference between the hotter P4 and the PD they used.

Good

One niggle.. Although mentioned that the zalman 9500 uses a faster fan, I guess many people will just look at the table and see it is 'best', whereas its really not that good!

The Zalman fan runs at 2600 RPM at 12v. The Nexus 92mm runs at 1500 RPM at 12v.

Per the article Devonavar referenced http://www.silentpcreview.com/article284-page2.html

The nexus is less effective at 12v than the stock fan is at 5v. To me that means you would be better off modding it with a Scythe 92mm or some other 92mm fan that would give you the option of running above and below 1500 rpm depending on the voltage. The Scythe DFS922512L is 2000 rpm at 12v and would be both cheaper than the Nexus and quieter than the stock Zalman fan.

You could still let PWM or hardwired voltage keep the fan on the quiet side but especially with Motherboard or other fan controllers you would have head room for hotter processors/environments.

I appreciate the info. Apparently the TDP comparison was 84W to 130W but doesn't tell the whole story. I understand why with TDP definitions being changed and fudged so often you went to using measured vs specified.

Back to the concept of °C/W, essentially you are saying that each test rigs °C/W numbers are not directly comparable to another test rigs numbers. If a HS/Fan combo is robust enough to handle both configurations the numbers will be similar. If a HS/Fan is unable to handle one or both configurations the numbers might vary wildly.

Does that about sum it up?

Even a 92mm Nexus at 12V is way too loud for anyone interested in a quiet computer and who deosn't live in a noisy environment, so basically the 9500 cant cool a normal CPU quietly under load.

I changed my nexus modded 9500 for a Ninja and it works a lot better. The Nexus fan on the 9500 simply did not cool well enough at 700-900rpm where it was inaudible, it cooled somewhat good at 1400rpm but not good enough to have my CPU stable overclocked and undervolted. The Ninja does that just fine and it is also a lot more quiet.

Yes, that more or less sums it up. In theory, °C/W numbers should be universal, but in practice there seems to be a lot of variance. We're always working to provide more reliable numbers, but the issue is quite complex, as you can see.

Given that desktop CPUs are being sold in 3 classes of TDPs now, wouldn't it make sense to have a High and Medium TDP test bed, and list numbers from each testbed for each heat sink?

Say the common desktop TDPs are:

<45 W
46-90 W
>91 W

In an Ideal world we'd have 3 test beds and test the same HS/F on all 3 configs.

In the real world every test bed added would cost time, money, and make it harder to write the article/charts in a clear manner.

Given that, I'd say two test beds are more reasonable than three.

One targeting 50W measured
One targeting 100W measured (with the realization that you just take whatever the hottest processor is and use it)

Maybe the better alternative from a managability standpoint is to split the recommended list of Heatsinks like you do for Power Supplies.

Label the sections somehow to separate low cost vs high cost or low TDP vs High TDP.

I know I'm always asking for more from SPCR so let me just say this:

Thanks to all the SPCR staff and regulars for all the hard work and honest presentation. I do appreciate it...

isn't the Nexus 92mm 1500rpm at full blast?

the article states 1380rpm, which I believe is incorrect.

Sample variancv, as i said ealruyer my 93mm necux only runs 1400RPMM. I forgot my second poitn.

The XP-120 is clearly outclassed, and was beaten soundly by the rest of the heatsinks. Its day is past...

A bit confusing to me, as the XP-120 is still listed as #3 on the Recommended List right after the Ninja (a Q of 8, vs an 8.5 for the Ninja).

Perhaps the Recommended List needs to be updated?

My previous post is quite wrong

The Xp-120 is still a good option for HTPCs because it is so low, a lot of other high performance heatsinks dont fit in most of the low desktop type cases.

Currently working at a high-end projector manufacturer, I was recently confronted with an aspect that did not come to my mind when thinking of cooling technology: the influence on cooling capacity of the atmospheric pressure.

We installed an 6-projector system in Mexico City, which kept on shutting itself down at temperatures way below what we had measured in the lab. Only after a while we realised it was the 2200m altitude of the setup that was a crucial factor in this problem. And thus we started testing at 700 mbar atmospheric pressure and indeed noticed that at these circumstances cooling was far from sufficient.

The reason why I post this here, is because this is a variable that isn't checked during your reviews, but does affect the heatsinks performance significantly. Some theoretical work lead me to the fact that at the same rpm of the HSF, the heatsink performance scales linearly with the atmosferic pressure.

A normal atmospheric pressure is 1013 mbar, but due to weather differences, this can vary from about 900 to 1100 mbar, indicating a 20% difference in cooling performance!

Though it's nearly impossible to take this into account when calculating 'a' performance number for a heatsink, it might be worth mentioning in the article, as it might put things into perspective.

I think this can only add to the correctness of your comparisons, and to the already very sound reputation of this website. Keep up the good work!