No, you want to minimise the overall noise of the system. How fast the air inside is moving is only one factor in that but in our scale of thinking (keeping things quiet) it is the fans that make the noise, not the draft of air. We're not blowing a hurricane here.


Have you even thought what purpose the graphics card fans will have in this? Drawing air in from the rest of the system, driving it through a heatsink and expelling it out of the back? That combined with a positive pressure set of case fans can be considered a push pull system. You are very confident of your claims and perhaps you do have general scientific and engineering knowledge but for computer systems there is a lot that is specific so please do consider what we are all saying and bare in mind that we might also have the same scientific/engineering background knowledge that you also have, we just also happen to know stuff about this field in particular.


Take a look at the SPCR review of the FT02:
http://www.silentpcreview.com/article1250-page1.html
It was given an editors choice and it was noted that adding a second graphics card made no real difference to temperatures or noise. This is a good sign if you are aiming for 3 graphics cards. Perhaps this system is lower powered but adding on the possible power of additional components I only get to 900W AC even under stress testing so again this confirms that a 1000W PSU would be fine.


Again, be careful with noise measurements. Unless you are going to be testing in exactly the same way as someone else you can not say that something is louder or quieter than something else that you haven't heard yourself. That SPCR system in the FT02 was tested at 30dB at load and might be a bit louder with the components you are talking about putting in it but it is possible it could be a similar range.

Interesting discussion.

Regarding whether a system with a GTX680 can be made to run quietly, here's one that's pretty good, done without resorting to extreme methods:
http://www.silentpcreview.com/AVADirect_Quiet_Gaming_PC_GTX680

Well yes, and the way to make fans quieter is to slow them down, but you still have to move enough air to carry away all the heat. So you want to maximize the heat transport you get for a given volume of air moved. This is where turbulence and mixing come in. The reason you want to minimize turbulence is to maximize heat transfer to the moving air. The reason you want good mixing is to make sure that all of the moving air is carrying away heat, not just the part closest to the heat sinks. These two factors allow you to carry away the heat with a larger volume of slower-moving air. The way to get that is to use larger, slower-spinning fans. And that's where the noise reduction comes from.
For standard GPU designs, yes, but they run a fair bit hotter than the DirectCU II 4GB, and they're much louder. Take a look at the mounting plate of the DirectCU II 4GB:

It can't be moving much air out of the system at all through that tiny little port, so it must be dumping most of it right back into the interior of the case. And that's why I'm so concerned with air flow in the case as a whole.That's very interesting, thank you. I must say, the reviews on this site are very thorough and informative. I will let you know my thoughts when I've read it through; hopefully, later today.Yes, I understand. The best we can hope for out of these kinds of calculations is a rough estimate. How rough? Well, maybe the answer to the question I asked MikeC below will give us some idea.That configuration is different from the one we're considering here, but not so different that we can't make a rough comparison. When I reduce the number of GPUs and case fans to match the system you linked, substitute a Caviar Green for the Barracuda XT, and so on, my calculations come to 29dbA at idle an 39dbA at full load. But the review measured 16dbA and 26dBA. The difference matches but the level is off by 13dbA. Of course, I haven't taken into account any noise damping effects of the case.

Could it be that this is causing me to overestimate noise production by as much as 13dbA?

First of all, your method of caculating combined SPL is way off. Referring back to this...

Not sure where you got the SPL numbers from, but assuming they are accurate, your SPL "addition" is way off. The way SPL adds is like this: Assume identical noise sources, say 15 [email protected] (cited for your GPU). Two of them close to each other (and still 1m away from you) will measure 18 dBA. Four of them will add up to 21 dBA.

But when you have multiple noise sources will making different levels of noise, all with different frequency spectrum balances, then it becomes almost impossible to "calculate".

We can, however extrapolate from the data I measured for the AVADirect single GTX680 system:
1) assume that the 2011 CPU you chose can be kept adequately with a HSF that makes no more noise than the one in the AVAD system (it can, I guarantee you)
2) The single loudest noise source in the system is the video card. The HDDs, etc are all way below the noise level of the video card, and contribute virtually nothing to the overall SPL. My guess is that if all other noise sources in the AVAD system I tested were eliminated, the overall SPL might drop by 1-2 dBA. And pulling out the video card would bring the overall noise down to perhaps 20-22 dBA at full load. Odd, isn't it? But that's the way it works -- and measures & sounds.
3) So with your 3-GPU system (which strikes me as an expensive way to get marginal performance benefits, but that's my opin & it's not my work)... if all the video cards could be cooled equally well with the same fan speed as in the single GTX680 AVAD system, then in a even with a couple more case fans, the overall SPL would be under 32 [email protected] at full load. At idle, I expect it might be 20~21 dBA (assuming good soft mounting for the HDDs and intelligent selection of quiet case fans).

But the biggest challenge with 3 GTX680s is how to prevent one or more of them from running way hotter & louder. The pics of the Silverstone case system posted earlier show the 3 cards jammed up against each other; assuredly, these cards' fans will not all run at the same speed, my guess is that the one in the middle would run way faster (and still be much hotter) than the one at the end whose fan has open access. Are there any motherboards w/3 or more PCIe slots that have two slot spaces between them? Without that, you'd have to mod all the cards with massive passive aftermarket coolers and some way of getting large amounts of outside air directly onto them w/ fairly high pressure and without making a huge racket. Big challenge, but this would be quieter than leaving any stock GTx680 coolers to fend for themselves in a tight 6-card-slot space.

I don't know, MikeC, but the following sites:

https://ccrma.stanford.edu/CCRMA/Courses/152/SPL.html
http://www.sfu.ca/sonic-studio/handbook ... Level.html
http://www.engineeringtoolbox.com/sound ... d_711.html
http://en.wikipedia.org/wiki/Sound_pressure

all give the formula for SPL in dB as

SPL = 20 * log(P / P_ref)

where P_ref = 20 uPa is the reference pressure, so doubling P adds 6dB to SPL, not 3.

Are you saying this formula is wrong?

Assuming we agree that this is the right formula, we can invert it by writing

P = 10^(SPL / 20) * P_ref

Now, unless there are some really bad tonal peaks, we can assume -- as a first approximation -- that all fans have roughly the same (pinkish) frequency spectra. Now, I know this isn't actually true, but if we choose our fans carefully, we can make it close enough for rough calculations.

Then, if we want to know the total sound pressure level of a number of sources given in dB, we have to convert each one to sound pressure before adding. Then we convert the sum back to dB. So,

1. convert each SPL in dB to pressure in uPa

P_1 = 10^(SPL_1 / 20) * P_ref
P_2 = 10^(SPL_2 / 20) * P_ref
...
P_n = 10^(SPL_n / 20) * P_ref

2. add the pressures

P_sum = P_1 + P_2 + ... + P_n

3. convert the sum back to SPL in dB

SPL_sum = 20 * log(P_sum / P_ref)

Putting it all together in one formula, we get

SPL_sum = 20 * log(10^(SPL_1 / 20) + 10^(SPL_2 / 20) + ... + 10^(SPL_n / 20))

This is the formula I've been using. Again, are you saying this formula is wrong?

On the page you linked -- https://ccrma.stanford.edu/CCRMA/Courses/152/SPL.html -- there is a heading "Multiple Sources" which repeats exactly what I said -- and I've confirmed this with umpteen hands-on measurements for many years.

Where you seem to be going wrong with that formula is assuming that a second identical sound source equals a doubling of pressure -- it clearly does not. Find a physics lab in a nearby university (or are you at one?) and try SPL measurements with a SLM and 2 identical fans. Or just use 2 identical speakers, an amp and a tone generator.

Here's another example of how multiple identical sound sources add up (in SPL) -- in this case, many hard drives in a system measured in my anechoic chamber: http://www.silentpcreview.com/Nine_Susp ... nd_Testing -- 6 HDDs, 16.85 dBA; 9 HDDs just 18.62 dBA (less than 2 dBA difference).

On the second page you linked, the last 2 paragraphs and the scale on the bottom explain clearly how 2 sound sources add in SPL (assuming they are identical except in level). http://www.sfu.ca/sonic-studio/handbook ... Level.html

That sentence is a figure caption, and the figure is inconsistent with the formula they (and all the other references) give. In addition, there is a link to a table right after the sentence you quoted. And that table gives the number as 6 db, so their table is consistent with the formula, but their figure is inconsistent with everything else.

The only possible conclusion is that the figure is wrong.

You're reading the table wrong. It's the Power or Intensity Ratio column that's relevant.

Listen, I'm trying to give you the benefit of over a decade of listening to, thinking & reading about, testing and measuring the acoustics of computer gear - and SPCR has the most systematic, consistent, practical test measurements and data on computer acoustics you can find on the web. Forget the formulas, look at the measured data -- that's what counts; what you hear is the only thing that counts more in the end.

Put 2 identical fans together, you will get 3 dB more -- period.

Or your thinking is wrong.

I admire Mike for his patience. But that's one of the reasons i chose to be a part of this community, the spcr-style we do have in our forums.

As Mike hinted at: You won't catch up on our knowledge with 5 minutes of googlish reading. Most of the people contributing regularly have more than 5 years of experience with silent computing or even more, in addition a lot of our members make a living in the IT business. So don't be disappointed if we are unable to enlighten you in 5 posts in this forums.

I'd say we've reached a splitting point: You take our input and just believe it or you get to study all that stuff you do not understand fully until you do. Then come back and convince us of something we haven't thought of or we were seeing the wrong way.

20*LOG(100) = 40.0 dB

20*LOG(200) = 46.0 dB

difference = 6.0 dB

What am I doing wrong?

Yes, don't worry, we all know that.

Nice load of theory there and I don't doubt that you have understanding of it however we all do too. The trouble is that when you talk of turbulence you are missing the major point of the Silverstone FT02 design: all of the air moves in one direction through the case and this minimalises turbulence. When trying to move a lot of air this is a good thing.


Assuming that a formula actually matches what is seen in practice. The formula might be suitable for many cases but what is seen in practice may be a special case. Experiments on SPCR have shown that adding fans doesn't really do what you expect noise wise and 3dB is the measured increase in SPL.

It's probably better that you give up on sound calculations. They are not neccesasarily correct, you're not going to measure it in real life anyway, the actual SPL doesn't matter in real use - how loud it is to YOUR ears does - and can you even calculate SPL when the fan speeds are not going to be all a constant or increasing to any given speed in any preset conditions?

Just get on and build the system with the components you want and if it ends up being too loud, THEN start looking at what you can do to make it quieter. Stop trying to silence something that is only theoretically loud!

Thats because

L(p) != L(w)

But both are meassured in dB.

http://en.wikipedia.org/wiki/Sound_pressure

http://en.wikipedia.org/wiki/Sound_power

Could it be, then, that when we talk about "SPL", we're actually referring to Sound Power Level and not Sound Pressure Level?

If so, then all those formulas above would have 10 in them instead of 20, and a factor of 2 would indeed correspond to 3 dB.

I'm not trying to be difficult, just trying to get to the bottom of how such a complete disconnect can come out of something so elementary.

Actually, that makes total sense if SPL stands for Sound Power Level, as conjectured above.

Going from 6 to 9 disks is a factor of 1.5, and 10*log(1.5) is 1.76. Well, the difference between 16.85 and 18.62 is 1.77 -- so the measured and calculated numbers agree to within a half a percent.

Seems like the apparent inconsistnecy is resolved.

Well, assuming the suggestion in my last two posts is correct, my corrected noise estimates come out to the following:

noise power at idle
CPU: 14 dbA x 1 = 14 dbA
GPU: 15 dbA x 3 = 20 dbA
PSU: 10 dbA x 1 = 10 dbA
disk: 17 dbA x 1 = 17 dbA
fans: 10 dbA x 7 = 18 dbA
total: 24 dbA

noise power at full load
CPU: 28 dbA x 1 = 28 dbA
GPU: 28 dbA x 3 = 33 dbA
PSU: 25 dbA x 1 = 25 dbA
disk: 19 dbA x 1 = 19 dbA
fans: 22 dbA x 7 = 30 dbA
total: 36 dbA

This has come a long way from my original estimate of 50 dbA at fulll load, due partially to correcting the formulas and partially to configuration changes -- both resulting from the good advice given in some very helpful posts by knowledgable people in this thread.

Many thanks to all who contributed!

Of course, I'm still open to further suggestions...

How did you get the noise-level for your harddisks down from 24 dBA to only 17 dBA ?
Are you still planning to use WD disks ? Green or Black ?
I just checked the WD website. And all WD Green disks seem to be specced at 24-26 dBA at idle.

Or is it that 9-hdd-review that said their disks only produce 17 dBA ?

I'm trying to make my own machine silent. Every step has improved my noise-level. But the HDDs are now the biggest problem. I have 3 sources of noise left. 5 Case fans, who are really silent now. A waterpump (Laing DDC-1T). With a new fan-controller (coming in a few days), I hopefully can undervolt my pump. (My pump seems to need 10-11V starting voltage. A fan-controller with "spinup control" should do the trick).

But the last source of noise are my 2 HDDs. 2x Samsung HD204UI. I have put them in elastic suspension. That makes a huge difference for the seek noises. But not for the "spinning sound". That noise is the same at idle as at during seeking. And it's rather loud. (Now that most other sources of noise are gone). Much much louder than my fans. I suspect (wild guess) that it is indeed around 25 dBA. I was thinking of buying 2 "silent boxes" for them. But 1) I've read that those help mostly for reducing seeking noises. (Which are gone from my system). and 2) I don't have room to hang 2x 5.25" cases in my FD Arc Midi.
I ordered a bigger SSD. Hopefully I can then power-down my 2x HDDs (manually with script, or with 5min Windows timeout setting).
A last resort would be to buy a NAS, and put the 2x HDDs in the NAS, and run only an SSD in my PC.

So if there is a trick to reduce HDD idle spinning noise from regular ~25 dBA on most disks, to just 17 dBA, that sounds very interesting.
Edit: I just saw that the WD Green WD10EZRX does only 21 dBA. But it is only a 1TB disk.

No, I didn't come up with some magic trick to reduce the noise output of a hard disk by 7 dB. I changed hard disks from the WD Caviar Black 2TB to the Seagate Barracuda XT 3TB. The noise measurement comes from this review, right here on SPCR.

By the way, have you thought about decoupling your water pump from the case? Martin's Liquid Lab found that suspending the pump by wrapping it in foam rubber reduced noise production hugely.

It's not clear exactly how WD measures SPL for its drives, but when a HDD is suspended with rubber cord and in a case, it is much quieter than sitting atop a piece of foam on a benchtop (which is how we measure them in the anechoic chamber, at 1m). Also, for whatever reason, our SPL measurements for the WD Greens are easily 7-8 dBA lower than what WD specifies... which again begs the question of how they measure; we know how we measure.

Well, don't keep us in suspense, Mike. How do they measure?

If Mike'd really really know how, he'd probably be working for them and would make tons of money, repaying him all those years of tinkering and learning.

Oops, I misread that as "we know how they measure."

OK. I also saw this list: http://www.silentpcreview.com/article1285-page6.html
From that list, it seems that the Samsung HD204UI are the quietest. I happen to have 2 of them. And I do hear them from 1m away. (My HDDs say HD204UI on their labels. However, I can not see if that means they are also EcoGreen F4. Maybe there are different HD204UI disks ?) But I wish there would be even quieter disks. I also wonder if I should buy a dBA-meter. Right now all my measuring is subjective. I bought a power-meter (only 15 euro) to look objectively at my PC's power usage (good idea). Unfortunately the dBA-meters that can go under 20 dBA seem very expensive.
Yes.
I sawed a 10cm hole in the back of the cupboard of my desk, which is next to my PC. And placed my pump in there. I put it in 3-5 layers of noise insulating foam. Unfortunately, the foam hardly blocks noise. (I learned that the foam is to deaden reflected noise, and not stop it going through. It stops only 20-30% of noise going through). I also suspended the pump just like my HDDs. I took a piece of concrete-iron, cut it in the right size, wrapped it around in the form of a cage. Then took a plastic skipping rope, and tied it in the cage to create an elastic bed. I made one for my 2 HDDs, and one for my waterpump. The effect on seeking noise of the HDDs was spectacular. Spinning noise seems unaffected. The suspension of my waterpump also hardly seemed to help, compared to just placing it on the noise insulating foam. (There is a benefit: the underside of the pump does not get warm anymore at all).

I have 2 tricks left for the pump. 1) Attach it to my new fan controller (Kaze Master Flat KM06-BK). It has "spin up" voltage. My pump needs that. With a regular controller, my pump has problems starting at 10V already. With this new controller, I hope I can run my pump at 7V (at which it runs, but does not start). At 7V, the pump is a lot more silent.
2) The second trick is to replace the noise insulating foam with cotton-based noise insulation. That material seems to be much better. Unfortunately it seems hard to find in my country. (60 euros cost + 30 euros shipping seems a bit over the top. But then, the amounts of money I spent on ridiculous stuff for this project keeps going up and up anyway).

I'm sorry to derail your thread. I was planning to ask more about my plans for HDDs in the storage forum. I might do that after I get my larger SSD (and fan controller) and can do some more testing first.

Edit: I took some crappy pictures (with my tablet).
HDDs:

Waterpump:

It's likely that they use the ISO 7779 standard, which places the IT noise source on a standard table & the mic 1/2 meter away, a bit above it. Not sure whether any padding is allowed underneath the drive. Anyway, the 0.5m distance compared to the 1m distance we use should not account for more than a 1.5~2.0 dBA difference. The fact that we use a low cut filter on the mic input to cut the rumble from day street traffic 1.5 blocks away could have some impact, though with the A-weighting, the effect is should be minor. I may have to test that to confirm.

For now, I have to admit again that I don't know why there's so much difference between their spec and our measurements. Until I have a chance to speak with a WD engineer and/or visit their test lab, this will probably remain a mystery.

OK, I have to correct myself, after a few minutes of checking SPL with a 3TB WD Red.

1) The 3TB WD Red is pec'd at 23 dBA.
2) Lawrence measured 13.65 [email protected] in his review. With the HDD on a piece of foam atop the table, I measured 13.6 [email protected] this morn. (nice to see the consistency!)
3) Decreasing the measurement distance to a half meter (keeping the mic at the same ~30 deg angle above the HDD), increased the SPL to 18.5 dBA, this being nearly +5 dBA from the 1m distance.
4) As expected, turning the low freq filter on/off made less than a 0.2 dBA difference at either distance.

18.5 dBA is closer to the 23 dBA WD spec than 13.65... but it is still a significant 4.5 dBA difference. The only way I could get my sample to measure anywhere near this high was to place it directly on the table, which then caused some vibration-induced resonance in the table top: the 1m reading went up to 16 dBA, and the 1/2 meter reading climbed to just under 21 dBA.

The table we use doesn't adhere to the table spec'd for ISO 7779, which could explain the 2 dBA difference.

I think my little experiments suggest that WD probably derives the acoustic spec for its drives by placing the drive directly on an ISO 7779 table and uses the "seated user" mic position (about 0.5m) to measure SPL in dBA. This is fair enough, as they expect most users to hard mount their drives in a chassis, unlike SPCR readers who mostly suspend or otherwise soft-mount their drives.

Anyway, mystery more or less explained -- tho until I get confirmation from WD engineering, all this is still conjecture.

Thanks for taking the trouble to do that, Mike! It provides some really interesting insight into the whole business of comparing noise measurements.