Stren's 2012 CPU Water Block Roundup

[snowyowl]

Impressive work Stren ! One wish would be to see picture of each taken appart....i know it's a lot of work but the engineer in me had to ask....can't wait to see my block against to whole pack have a great vacation

[Six]

Just one thing. I dont understand why you give to the Alphacool block a gold.
It has disadvantages of quality, flowrate, temperature variation of different mounts.
Nice work. Many thanks.

[BuzzNL]

Here's some info for everyone who wants to buy the Alphacool block now (including myself):
I sent an E-mail to Alphacool asking why the block was called the 'Light' version and they replied to me that there will be an all-metal black-chrome version of that block early next month.
A few more weeks waiting I guess

[st0ned]

Amazing testing ! And not less interesting results, I find it funny the "volte face" that happened on socket 2011 vs the results I got out of socket 1155,this only shows that both sockets need to be tested and we cant assume the results taken with one of the to be of any relevance on the other.


kepp up the good work

[rge]

Though socket/type of cpu matters, you cant even make conclusions based on same socket/cpu type. Several testers tested 2 samples of same block on same system and got up to 1C or more difference, so just water block variance alone could put 2C difference in testing same block which would make a top block in one test be near middle of pack in another, if got best sample of one block and worst of another. Then add in IHS differences plus socket/cpu type differences, and no wonder results dont show the blocks performing in same order. I looked at data from 9 testers, including yours, it was skinnee, Stren, St0ned, hw max, martin1, martin2, jayhall, cartago, scamps, and mine. Only used testers that measure water temps.

Made a chart showing avg degrees C each block was from the best in each test and the standard deviation. The difference was just as pronounced looking at blocks tested on same socket, both martin and stren used 3930K on some testing, other testers used same cpu type, again with variable results. So I wouldnt even expect testing same socket would show any better correlation. Variability in individual IHS differences and waterblock base/channel differences are more than enough to prevent same block from testing same across multiple tests, even if all tested using same socket, but add in socket cpu type differences, and again no wonder of differences.

alphacool nexus only been tested twice, and is only one marginal test away from joining rest of pack, so cant make any conclusions, may simply not have been tested enough yet. Not scientific for showing what block tests best, but looking at SD vs difference, you can see the issue.

[Church]

So, if variable of block samples is at play, and it's impossible required costs and available testing time wise to test large enough amount of blocks .. it's up to - "take any from these top blocks performing within degree choosing by any other factors but cooling efficiency", and just "drop from selection choices these clear outsiders performing noticeably worse".

[st0ned]

Well I've nexxxos here sadly summer got in the middle of my tests but I'll get to it. Though I should confess I'm admired with the differences shown between nexxxos and phobya, because looking at them side by side they look a lot alike, taking out the part that the nexxxos seems to be a more restrict block due to a more restrict inj plate

[rge]

I would be interested in seeing more testing in either of those, Im getting a nexxxos myself, just to test on my system for my own curiosity. Doubt will end up using it, unless I like the looks better than what I see in pic.

The nexxxos base has 90 degree cross-cuts/ie pins that are 2.5mm high and .5 x .5mm wide, more similar base to swiftech. The phobya uc-1 base has 68 channels, each 0.2mm wide and ?depth. So bases, tops, injection are different.

But I am of same opinion as Churchy, need multiple testers to weed out the blocks that consistently dont perform as well, then of remaining ones choose based on personal preference/aesthetics/mounting/construction quality/etc.

I am probably going to end up with a few more blocks just out of curiosity, but will be using the one I like looks/quality of most, not the one I get .5 or 1C better temps with...that I wouldnt be able to see or get any benefit from.

[musicfan]

I agree with rge & churchy we want to "weed out" blocks "performing noticeably worse". But the question is how? rge your post #80 table is interesting. How many tests does each row represent? And what URL may I find scamps tests please? Thanks. PM if you wish.

Looking at the same tests (I missed scamps), it's unusual for any sample to fall 2 SD beyond the mean of that population in the test. In stren's MX2 5-mount test; N=15, Mean is 41.8°C, SD is 1.19 or population SD 1.15. The Rasa falls 2 SD worse than mean. The M6 is barely 1 SD worse but is the only one (other than the Rasa). The XP3 is 1 SD better than mean and the only one. The rest huddle +/- 1 SD around the mean.

Does the Rasa deserve outlier status? Weeding out the Rasa may be an easy call since XSPC promotes their newer Raystorm over their older Rasa. If we use 2 or maybe even 1 SD (that would add the M6) as a limit for our weeding, can't we say, "if all else is equal, newer CPU blocks test within 2 to 3°C of each other"?

[rge]

Scamps tested 3 blocks here, and then later in another post added how the supremecy did in comparison.

My original graph also had how many tests of each, but deleted for simplicity sake, figured no one would be interested...should have kept it in. But most blocks tested in 3-5x, mips and nexxxos 2x, and supreme hf and raystorm tested near all. Though alphacool nexxxos im testing tom or sat, have block now. Actually mine doesnt look bad, the acetal is somewhat shiny and the gold name is in an indented area and on straight. Though the mounting looks cheap and ugly. If I do use it for 24/7 will be using EK's newer black mounting system which I like best in quality/looks. Luckily their mounting system was too small to stamp circles all over.

And yep, I would call the Rasa an outlier, it is a generation back, and consistently tests near bottom when against any of newer ones including like you mentioned XSPC raystorm.

[musicfan]

@rge. Thanks so much for the link. It's fun to read German by Google translate. Makes me want to clean up my grammar. Now that I know your rows had multiple results in each row, what exactly was the SD describing? the worst case or an average? I am only seeking to the math and how it might help us. For example, several testers testing several blocks gives greater reliability. Thanks for the help.

@XS...On to flow...and the CPU Water Block Flow Rate (MCP35x2@39%PWM) graph. Stren makes some gorgeous bar graphs. When I count the bars, there are 15 so N = 15. When I put the numbers in the calculator and round to 3 significant figures, the mean is 1.68 gpm and median 1.72 gpm. SD is 0.176 and population (using all 15) SD is .170 gpm. The only ones below -1 SD are the Apogee HD, Rasa, and UC-1 LT, which are moderately restrictive relative to the pack. The MIPS IceForce HF is probably the High Flow champ but barely flirts with +1 SD.

If I was using this to help guide parts selection, I would encourage those who purchase the UC1-LT to have a good pump, like a Laing, if it is going to part of big loop. Relatively speaking it's slightly more restrictive than Swiftech's Apogee HD, which helps place it. As always, please correct me if I erred. Just my opinion. Again, many thanks to stren for sharing his work.

[rge]

What was alphacool thinking with the nexxxos block threads. No way am I testing, let alone using 24/7 a 50% threaded water block. Nearly 10 years of watercooling, I have seen some shortcuts but really? See pic of out "threads"

Got to love a review like here, when the block in both pics is turned the only way you can not tell it is only 50% threaded.

Since no way I would use it, using my fingers I tightened it slowly just to see if tightened ok...it gets a little snug but still could easily rotate fitting by gently twisting tubing, then with tightening with fingers a little more, it easily went loose, ie stripped. I have never stripped a block in 10 years with countless blocks, finger tightening all the way, then 1/8 turn with wrench. This easily stripped with finger tightening.

What a waste of whatever I paid for it, and more irritating the time I spent preparing to test it....this one is going in the trash.

alphacool.jpg

Edit: @musicfan, it was an excel calculated SD from mean.

[paulbagz]

Tbh, I haven't touched Alphacool since I bought one of their 9800GTX blocks.

I see I haven't missed much lol

Thanks for the head up rge!

-PB

[BuzzNL]

What was alphacool thinking with the nexxxos block threads. No way am I testing, let alone using 24/7 a 50% threaded water block. Nearly 10 years of watercooling, I have seen some shortcuts but really? See pic of out "threads"

Thanks for the warning indeed.
I almost ordered this block but when it's only half threaded there's no way I will buy it.
Looks like I have to look at the test results again to see what my new choice is going to be.

[musicfan]

Thanks rge. Looks like there is more than one way to "weed out" a block. The quality control will not be satisfactory for many on that Alphacool. Great picture by the way.

Again nice review by stren. I really like reading the thumb nuts of the Swiftech and EK mounting bottom out to make it easier to get the unit to perform as intended.

What are opinions about the CPU Waterblock Mount Variation (SD of 5 mounts)? Others looked at loose parts or number of steps necessary as a proxy for ease of mounting but the temperature variation of the 5 mounts seems like a very good proxy.

In this the mean is .44 and the SD +/- .26. The best are borderline 1 SD better than mean. The M6 is 1 SD worse than mean and the Heatkiller 3.0 and Rasa are nearly 2 SD worse than mean.

Will this "weed out" any blocks for anyone? We have already weeded out the Rasa. Would this remove the Heatkiller 3.0 for some as a new product recommendation? How about the M6? Thanks.

[jogshy]

A question about the Alphacool XP3 .... does it admit two 13/19mm compression fittings? I have the impression the water I/O holes are too close together.

[Martinm210]

Great job..you have been busy!!

[stren]

Back from vacation so it's time to read through all the replies!

Impressive work Stren ! One wish would be to see picture of each taken appart....i know it's a lot of work but the engineer in me had to ask....can't wait to see my block against to whole pack have a great vacation

Thanks snowy - that was always the plan after I was done testing. Sadly I didn't have time to take them before my vacation and I didn't want to hold up the results. Now that the alphacool thread issues has been found I very much regret this as it would have caught the problem. More on that later.

Just one thing. I dont understand why you give to the Alphacool block a gold.
It has disadvantages of quality, flowrate, temperature variation of different mounts.
Nice work. Many thanks.

Yes it has lower flowrate and lower quality. But flow rate data is presented and none of them are so low that I would consider it a problem. Given that fixed pump power is used for testing the extra restrictiveness is taken into account with thermal performance. In terms of performance the alphacool block does top the roundup. This doesn't mean that I would necessarily buy it myself as I favor a certain amount of quality and aesthetics myself. But those are things that are very hard to scientifically test and become quite personal. In the review I was trying to focus on scientifically measurable criteria for differentiation.

Here's some info for everyone who wants to buy the Alphacool block now (including myself):
I sent an E-mail to Alphacool asking why the block was called the 'Light' version and they replied to me that there will be an all-metal black-chrome version of that block early next month.
A few more weeks waiting I guess

Yes I saw they also claimed 2C better performance with the all metal block and this sounds less than likely.

Amazing testing ! And not less interesting results, I find it funny the "volte face" that happened on socket 2011 vs the results I got out of socket 1155,this only shows that both sockets need to be tested and we cant assume the results taken with one of the to be of any relevance on the other.

kepp up the good work

Thanks st0ned - yes it's hard to know. I'm hoping to retest at some point with another socket and the same blocks to weed out what is socket variation from block-block variation. I also have double of some of the blocks so if I get really bored I might just run those too!

Though socket/type of cpu matters, you cant even make conclusions based on same socket/cpu type. Several testers tested 2 samples of same block on same system and got up to 1C or more difference, so just water block variance alone could put 2C difference in testing same block which would make a top block in one test be near middle of pack in another, if got best sample of one block and worst of another. Then add in IHS differences plus socket/cpu type differences, and no wonder results dont show the blocks performing in same order. I looked at data from 9 testers, including yours, it was skinnee, Stren, St0ned, hw max, martin1, martin2, jayhall, cartago, scamps, and mine. Only used testers that measure water temps.

Made a chart showing avg degrees C each block was from the best in each test and the standard deviation. The difference was just as pronounced looking at blocks tested on same socket, both martin and stren used 3930K on some testing, other testers used same cpu type, again with variable results. So I wouldnt even expect testing same socket would show any better correlation. Variability in individual IHS differences and waterblock base/channel differences are more than enough to prevent same block from testing same across multiple tests, even if all tested using same socket, but add in socket cpu type differences, and again no wonder of differences.

alphacool nexus only been tested twice, and is only one marginal test away from joining rest of pack, so cant make any conclusions, may simply not have been tested enough yet. Not scientific for showing what block tests best, but looking at SD vs difference, you can see the issue.

Thanks rge - I haven't seen the data on 2 samples yet. I do believe a quite significant change from socket to socket though. SB-E is really so large and flat, while IB really isn't. My data and Martin's were both on 3930K and showed fairly large differences though so it may be the case.

So, if variable of block samples is at play, and it's impossible required costs and available testing time wise to test large enough amount of blocks .. it's up to - "take any from these top blocks performing within degree choosing by any other factors but cooling efficiency", and just "drop from selection choices these clear outsiders performing noticeably worse".

Yeah pretty much - I think given how close things are you may as well choose based on the secondary characteristics of the top few blocks (hence so many silver awards)

I agree with rge & churchy we want to "weed out" blocks "performing noticeably worse". But the question is how? rge your post #80 table is interesting. How many tests does each row represent? And what URL may I find scamps tests please? Thanks. PM if you wish.

Looking at the same tests (I missed scamps), it's unusual for any sample to fall 2 SD beyond the mean of that population in the test. In stren's MX2 5-mount test; N=15, Mean is 41.8°C, SD is 1.19 or population SD 1.15. The Rasa falls 2 SD worse than mean. The M6 is barely 1 SD worse but is the only one (other than the Rasa). The XP3 is 1 SD better than mean and the only one. The rest huddle +/- 1 SD around the mean.

Does the Rasa deserve outlier status? Weeding out the Rasa may be an easy call since XSPC promotes their newer Raystorm over their older Rasa. If we use 2 or maybe even 1 SD (that would add the M6) as a limit for our weeding, can't we say, "if all else is equal, newer CPU blocks test within 2 to 3°C of each other"?

Yes I would agree that the rasa is really an outlier. It's an older generation block anyway. I only included it because I had it and because it's a useful comparison reference for older tests as it's never been revised (to my knowledge).

Thanks rge. Looks like there is more than one way to "weed out" a block. The quality control will not be satisfactory for many on that Alphacool. Great picture by the way.

Again nice review by stren. I really like reading the thumb nuts of the Swiftech and EK mounting bottom out to make it easier to get the unit to perform as intended.

What are opinions about the CPU Waterblock Mount Variation (SD of 5 mounts)? Others looked at loose parts or number of steps necessary as a proxy for ease of mounting but the temperature variation of the 5 mounts seems like a very good proxy.

In this the mean is .44 and the SD +/- .26. The best are borderline 1 SD better than mean. The M6 is 1 SD worse than mean and the Heatkiller 3.0 and Rasa are nearly 2 SD worse than mean.

Will this "weed out" any blocks for anyone? We have already weeded out the Rasa. Would this remove the Heatkiller 3.0 for some as a new product recommendation? How about the M6? Thanks.

Thanks Yes I didn't particularly like the number of steps as a useful gauge of mounting difficulty. For most blocks on 2011 the steps are few anyway. My main dislike was those blocks that required access from behind the board or had excessively long threads. The problem with the mount variation is how much the variation can be skewed by a particularly bad data point. 5 data points really is too few for this kind of metric I feel. But more than that is just too much time when looking at so many blocks.

A question about the Alphacool XP3 .... does it admit two 13/19mm compression fittings? I have the impression the water I/O holes are too close together.

Not sure if I have any handy but the port spacing is about 31/32" of an inch center to center. Make sure you read the info about the threads on the alphacool before buying. All of the blocks I could manage to fit my VL4N QDC's on except for the heatkiller.

Great job..you have been busy!!

Thanks!

[Church]

Thanks st0ned - yes it's hard to know. I'm hoping to retest at some point with another socket and the same blocks to weed out what is socket variation from block-block variation. I also have double of some of the blocks so if I get really bored I might just run those too!

If you will test on other socket too, imho better with Ivy Bridge, not Sandy, as Ivy due even smaller core, thus higher heat throughput per area, and worse stock TIM1 below IHS, potentially being even more of a "different" platform potentially favouring even more bow/thermal contact.

[bds71]

...and worse stock TIM1 below IHS...

given the TIM issue under the IHS, would it not be wiser (from a consistancy perspective) to stay with sandy bridge? doesn't ivy bridge (stock) just introduce another variable the way they incorporated the IHS? or is it pretty consistant (though definately not the best option)?

[stren]

If you will test on other socket too, imho better with Ivy Bridge, not Sandy, as Ivy due even smaller core, thus higher heat throughput per area, and worse stock TIM1 below IHS, potentially being even more of a "different" platform potentially favouring even more bow/thermal contact.

given the TIM issue under the IHS, would it not be wiser (from a consistancy perspective) to stay with sandy bridge? doesn't ivy bridge (stock) just introduce another variable the way they incorporated the IHS? or is it pretty consistant (though definately not the best option)?

While I'd like to test ivy by the time I'm done with the SB-E testing I think haswell will just be released. I'll most likely be picking one up for my wife in matx format to replace her e8500 so I imagine I'll steal it for a couple of months for testing on that. Hopefully Intel will have gone back to welding the IHS down by then. If haswell is late I might get a 3570K for HTPC duties lol in which case Ivy may be tested.

BTW feedback from alphacool:

I can confirm you that technically the thread is 100% ok, there is no
problem with mounting fittings, neither of leaking or of compability.
Attached I have got a photo with a fitting mounted that has a 5mm long
thread.
As you know most actual threads are even shorter.

So Alphacool designed the waterblock to maximize the performance that is why
they need a maximum waterbackchannel, and if you use threads with <5mm
length (for example the new Bitspower or Alphacool fittings) then that is
also even better for the waterflow.

Because there is an O-Ring under the thread, it is technically not possible
to cut the “half-thread” of the other side away, so that may look a little
strange, but is just a result of the maximum performance that this
waterblock has.

I can understand that the users in the forums may think that this I a
construction fault, but if you think a little about this, you may also come
to the conclusion that the thread is 100% good and working and there is
nothing to worry about. I would be grateful if you could explain that to the
forum users, so that this little misinterpretation does not lead to a
mocking of the waterblock.

Have a nice day!

Given that a fitting has some amount of thread anyway it's ridiculous to assume that this gives better performance as the fitting will block the flow anyway. I didn't see a leak, but it's a silly reason to risk it, and it's not like this block is spectacularly high flow anyway. I'll be following up to ask how much torque did it take for a fitting to rip out and whether the full metal version will be the same. It sounds like they are defending the design rather than fixing it. Shame.

[Martinm210]

For what it's worth, I was curious about a more scientific answer about thread engagement length and did some searching.

Most machinist handbook suggest anywhere from 1.5 to 3 X the bolt diameter assuming like material tensile strengths. The intent and general good design results in the screw or bolt failing before the female thread.

This link provides some calculators and additional design information on thread engagement:
http://www.engineersedge.com/thread_...engagement.htm

It also notes the above assumes the tensile strength of the bolt/hole materials are equal. Unfortunately in regard to brass/acetal we don't have that, so you should adjust it even longer by the stress ratio.

Yield Tensile Strength of Acetal = 8-10,000psi
Yield Tensile Strength of Brass = 19,600psi

So...technically you should actually increase that thread length by about a factor of 2

Regardless I have yet to find something to suggest much less than about 1.5X the barb diameter and we have a worst case scenario mixing acetal and brass where the female thread actually has about 1/2 the material strength of the brass barb. Looking at a typical barb, I see many with only a measly 2-3 threads or about .4X. Granted the barb itself is also compromised with it's reduced x-sectional area being drilled out to extremes, but the numbers seems to go far below typical machining standards.

I personally have stripped a block or two in my day of wrenching on standard barbs so it does appear even with the full engagement of the barb that we are already dealing with what most machinists would call a bad threading design in our barbs thread lengths. It has been a copied and accepted compromise made to reduce materials costs and why most users limit torque to hand tightening methods to mitigate the poor design. Unfortunately most barbs still today do not provide any torque guidance so stripping has been and will continue to be a problem for new users even with plenty of block threads to match that of the substandard barb thread depth.

Bottom line, barbs themselves are already poorly designed to handle normal torque limits possible using tools and should include some sort of torque guidance. If a block design utilizes even fewer threads they should seriously consider providing some torque guidance. Similar to that of a spark plug or as Swiftech has done, tell users they should tighten to "X" turns after o-ring contact. Without that critical guidance, it's anybody's best guess as to what is enough or too much...my 2c...never liked accepting hand tightening of barbs as it is and one of the reasons I've always favored brass tops where you get twice the strength.

As far as testing on multiple chips goes, don't burn yourself out. Every chip out there will produce a slightly different heat signature and produce completely different results, but nobody is a machine so cut it off clean with your current platform and let someone else slave away at it later..

[blackpanther26]

good review. And yes I think Ivy would be a good challenge to the water blocks as ivy is a hot chip.

[stren]

For what it's worth, I was curious about a more scientific answer about thread engagement length and did some searching.

Most machinist handbook suggest anywhere from 1.5 to 3 X the bolt diameter assuming like material tensile strengths. The intent and general good design results in the screw or bolt failing before the female thread.

This link provides some calculators and additional design information on thread engagement:
http://www.engineersedge.com/thread_...engagement.htm

It also notes the above assumes the tensile strength of the bolt/hole materials are equal. Unfortunately in regard to brass/acetal we don't have that, so you should adjust it even longer by the stress ratio.

Yield Tensile Strength of Acetal = 8-10,000psi
Yield Tensile Strength of Brass = 19,600psi

So...technically you should actually increase that thread length by about a factor of 2

Regardless I have yet to find something to suggest much less than about 1.5X the barb diameter and we have a worst case scenario mixing acetal and brass where the female thread actually has about 1/2 the material strength of the brass barb. Looking at a typical barb, I see many with only a measly 2-3 threads or about .4X. Granted the barb itself is also compromised with it's reduced x-sectional area being drilled out to extremes, but the numbers seems to go far below typical machining standards.

I personally have stripped a block or two in my day of wrenching on standard barbs so it does appear even with the full engagement of the barb that we are already dealing with what most machinists would call a bad threading design in our barbs thread lengths. It has been a copied and accepted compromise made to reduce materials costs and why most users limit torque to hand tightening methods to mitigate the poor design. Unfortunately most barbs still today do not provide any torque guidance so stripping has been and will continue to be a problem for new users even with plenty of block threads to match that of the substandard barb thread depth.

Bottom line, barbs themselves are already poorly designed to handle normal torque limits possible using tools and should include some sort of torque guidance. If a block design utilizes even fewer threads they should seriously consider providing some torque guidance. Similar to that of a spark plug or as Swiftech has done, tell users they should tighten to "X" turns after o-ring contact. Without that critical guidance, it's anybody's best guess as to what is enough or too much...my 2c...never liked accepting hand tightening of barbs as it is and one of the reasons I've always favored brass tops where you get twice the strength.

As far as testing on multiple chips goes, don't burn yourself out. Every chip out there will produce a slightly different heat signature and produce completely different results, but nobody is a machine so cut it off clean with your current platform and let someone else slave away at it later..

Thanks Martin - excellent post as always! Interesting to know this, I've often felt the threads were too short on some blocks, but it's good to hear just how much they are undersized!

You are right about burnout. I'll try and listen to how I'm feeling. Chances are reviewing something else may be more exciting than repeating with the same blocks. We'll see how it goes

[rge]

Look at the picture of alphacools poorly made block with the fitting screwed in. By having no threads on 1/2, it does not help flow at all. I used a bitspower fitting which already pushes the envelope with very short threads, and any amount of threads is going to block some of that "hole", so no increased flow will exist.

So looking at that pic, why is there not at least some threads on 100% to at least equal the amount that even short threaded fitting sticks out? NO possible reason other than cost. Or was I supposed to cut 1/2 threads off my fitting as well, and then match up the holes?

Looking at the design to make threads 100% might cost a little extra. But unless water magically flows through metal fitting screwed in there, it wont help flow to have no threads there at all.

As for how much torque, mine was not finger tight and easily stripped.

alphacool_bpfitting.jpg