Stren's 2012 CPU Water Block Roundup

[.::NeO::.]

Hi everyone,

I see in this review that the Watercool Heatkiller 3.0 is still one of the best performer, on par with the top performer out there. And because I like a lot the Ni-Bl version (especially with the new Bitspower white compression fittings) I want to use it in my new loop.

The loop:

1x Watercool Heatkiller 3.0 Ni-Bl
2x Watercool GTX680 Fullcover
4x Alphacool UT60 480mm
2x D5 Vario + EK Dual top single loop

My question is, since the Heatkiller is a little more restrictive, will 2x D5 Vario @ speed setting 5 be enough for this kind of loop, or should I split the loop in two parts? (I would like to have one big loop for pump redundancy)

thank-you in advance

[Church]

.::NeO::.: imho you, like many others, overstate importance of flow. To me it seems that you won't see significant temp drop even with one D5 at speed #3.

[bds71]

The koolance 380 and Phobya through most users pump settings would be a performance tie. The "GPM flow chart" is actually an "unfair test" since to achieve equivalent flow in the more restrictive Phobya, you are actually adding more pumping power to that block than Koolance...nobody in real world sets their pump speed a notch or 2 higher on one block vs another to equal flow through a more restrictive block...the restriction is part of its performance.

while it may not be a fair comparison using overall flow, it does show what the block is capable of, at those flow rates. i use dual pumps in my set-up (dual PMP400's - with an option to add a third) so flow is more relevant to my set-up. but i do understand your point. even still, at a given set point the two are dead even (and, if i'm not mistaken, the phobya is considerably cheaper?) - yet the phobya was still not recommended - in fact, it didn't even receive an honorable mention (or any mention at all to be specific).....is this only because of the poor MX2 results?

side note: the Rasa and the HeatKiller also had similarly high deltas between IX and MX2 - i'm wondering if perhaps something in the design of the base leads to poor mounting with this particular mounting set-up? maybe they just need more pressure applied (since you limit the pressure to 40 ppsi for fairness)? as it stands now, i would consider them "outliers" rather than doing a lot of retesting to find out why they stand apart from the rest (3/18 is only 17%). anyway - thanks again for all your great work in testing all these blocks Stren - i certainly don't want you to think this is an "attack" on you or anything. just observations

[rge]

From Stren's post 18 in this thread which states the results, he gave the phobya the silver award...see quote of that post below. The phobya has good internals evidenced by indigo results. But bow hence contact is not optimized as evidenced by mx2 results.

And I use 2 D5 pumps myself, so interested in high end of curve on pump% settings test...but still no interest in gpm test since I dont turn my pumps down when using a less restrictive block to equalize flow.

Gold Award

The Koolance CPU-380 performed exceptionally besting every block on MX2 (nearly all of them by 1+C) while still being at the top of the heap with Indigo Xtreme. The quality and feel of the block is up to scratch too. Mount variability is good and the flow is decent too and I would heartily recommend it.

Silver Awards

The following blocks also performed very well and have their own niche uses:
– Swiftech Apogee HD because it performs very well and has a very nice mounting system
- MIPS IceForce performed very well and has very high flow
- the Phobya UC1 performs very well with Indigo Xtreme
- the XSPC raystorm performs well and wins the value for money award!

Until Alphacool has fixed the single port threading issue I can not recommend the block!

[stren]

Ah I'm glad rge looked that up. My memory is worse than I thought! haha.

An update on the rotation:

Originally when I started testing I kept the channels orientated the same way for the first few blocks because I thought that that as many variables as possible should be minimized. Then I changed my mind as I figured if I had the orientation incorrect that it was disproportionately unfair to blocks without channels. I thought that as I was testing with manufacturer's mounts and following their installation instructions then I should also rotate the block as they intended. I meant to go back and retest those blocks. However I forgot and that's why the Phobya MX2 data was bad. I apologize for this. It's only when I saw some of the funkier IX to MX2 deltas that I remembered this and had to go back and re run data points. Rerunning with the correct orientation on the Phobya produced results that beat the Koolance CPU-380 in normal orientation (as predicted by the IX results). However to be fair I have to retest the other blocks on mx-2 that have bad data (DD M6, alphacool nexXxos, BP Summit). Note that I believe this is only constrained to MX2 and not IX. So I wouldn't expect a major divergence from the IX results. In addition I want to retest the CPU-380 when rotated as it may still equal or best the Phobya block on MX2.

Hopefully I'll finish that this week and this will mean that all the blocks will have an IX-MX2 delta of ~1-3C with the exception of the Rasa and the Heatkiller 3.0. I'm not sure why those two are so bad. I ran another 5 MX2 runs on the HK and got similar results so it's repeatablly bad with MX2 over 10 runs. The Rasa performance could be explained by bow, however the heatkiller is pretty flat already which should suit a 3930K. So this confuses me, and I don't like to be confused by the data as it can point to flaws in my procedure!

If I have time I'll also re run a couple of other blocks rotated (Supremacy/Sniper/HK3 come to mind). After that I'll be selling off the 3930K rig and then I'll take the flow measurements finally with my shiny new manometer. If I test cpu blocks again in 2013 it'll most likely be on Haswell, but this test took about 6 months for me to finsih with a lot of learning on the way, so whatever I do next time will be smaller in scope and hopefully quicker as I have learned more!

So my expectations are that the two leading blocks for the 3930K/3960X/3970X will be the CPU-380 when rotated and the Phobya UC1 in normal orientation for both IX/MX2. Between the two I have to admit I prefer the look/feel/quality/flow and mount of the CPU-380 to the phobya. However the phobya has the price advantage as it's $20 cheaper.

[stren]

Here's some updated data from the retests:

Note that based on the MX2 to IX delta that I still don't trust the MX-2 data on the following blocks and still need to re run those to be sure:
- Alphacool NexXxos (might be rotated on MX2)
- Danger Den M6 (might be rotated on MX2)

I think I'll also retest the Bitspower out of paranoia, and hopefully test the supremacy and sniper at 90 degree rotation too. Typically it seems to be making 2-3C difference average with MX2.

[Church]

I suggest always updating data&charts in initial posts aswell, as many users going for quick result lookup may be lazy and won't page through all theme pages to get latest data very probably drown among other posts by time.

But hmm, indeed it seems worth to retest orientation with other popular blocks. Heck, it's two degrees from nothing, a lot for modern blocks with as close results. Vendors not checking performance of their own blocks and not releasing suggested best orientation mount instructions?

[Martinm210]

Love the work and detail you've put into this Stren. I've said it before, but you've raised the bar again...

My hat is off to you..

[m_jones_]

Can't thank you enough for putting so much work into the watercooling community. Never expected the order to be this and rotating the block to make such an difference.

[rge]

Stren, you may have this some where, but when you say 90 degree rotation, can you clarify barbs in horizontal or barbs vertical. I assume the cpu 380 was originally mounted with barbs horizontal?, and by 90 degree rotation you mean barbs vertical, or is it other way around. Thanks.

[stren]

I suggest always updating data&charts in initial posts aswell, as many users going for quick result lookup may be lazy and won't page through all theme pages to get latest data very probably drown among other posts by time.

But hmm, indeed it seems worth to retest orientation with other popular blocks. Heck, it's two degrees from nothing, a lot for modern blocks with as close results. Vendors not checking performance of their own blocks and not releasing suggested best orientation mount instructions?

Thanks Churchy - yes I've been lazy - I was hoping to get a few more done this week and then update all the pics lol. I was hoping no one would notice As for vendors - yes I was surprised by this too!

Love the work and detail you've put into this Stren. I've said it before, but you've raised the bar again...

My hat is off to you..

Thanks again Martin!

Can't thank you enough for putting so much work into the watercooling community. Never expected the order to be this and rotating the block to make such an difference.

No problem yes I was surprised by the rotation too!

Stren, you may have this some where, but when you say 90 degree rotation, can you clarify barbs in horizontal or barbs vertical. I assume the cpu 380 was originally mounted with barbs horizontal?, and by 90 degree rotation you mean barbs vertical, or is it other way around. Thanks.

It depends. By "normal" rotation I mean if you had the block on a board in a normal ATX tower case then the text on the block would be "normal" i.e. horizontal. For the phobya that means vertical ports, for the CPU-380 that means horizontal. Ports don't always align with the waterblock channel direction either lol.

[stren]

So this is more of a "rambling and thinking out loud" kind of post rather than a real "here's some awesome data that's very conclusive". This makes me confused.

After rotating the CPU-380 and Phobya blocks and seeing a 2-3C difference between vertical channels horizontal, I expected this would be similar for all blocks. Then I re ran the Danger Den M6 as this was one of the blocks with poor MX2 performance relative to Indigo Xtreme.

First off I was lazy so I used the CPU-380 mount because it's much quicker than taking the board off and putting screws through the back of the socket. So I retook the data with the "normally orientated block". I saw an improvement of 2-3C and thought "Wow that's a lot for the mount". I bet the rotation will get me even more of an improvement. But I didn't. Instead the data stayed the same. So questions that are on my mind:

- Why did the mount (bolts/springs/nuts) have such a difference?
- Why didn't the rotation matter this time?
- Is there any reason to suspect anything funky?

Here's the data - first off - the latest version of MX2 vs Indigo Xtreme for each block:



I was working through retesting the blocks with poor (4+C deltas), the HK3 was repeatable - I did ten mounts and they look similarly bad. The rasa I haven't tested and probably won't. The DD M6 moves to right where it should be when using the CPU-380 bolts/nuts/springs though.

Here's the hard data on the M6 runs, ignore run #6 I really think I must have got a setting in prime wrong (e.g. default to blend not my normal custom run):



There's virtually no difference here when rotated at all! If we fold this into the overall averages then we get this:



Let me know your thoughts! Stuff like this throws me for a loop and I always like to hear the opinions of those who have more experience! The data seems repeatable, but I just don't feel like it makes sense!

[Martinm210]

I think it depends on a few things:

IHS shape, seemed like my 2600k liked the really sharp bow both directions.

Some blocks bow both directions while others only bow one way and which works best depends a little on the individual IHS shape.

Bow flattening is also influenced by mounting pressure, some all metal blocks are fairly rigid and can retain their bow under high pressure while others flex and can do worse with more. I can recall a few blocks that didn't like too much pressure.

The M6 is one that needed much more bow on my 2600k than factory and I saw huge improvements with shims on that processor to induce more bowing. Not sure on the 3930k though it doesn't like nearly as much bow.

It all comes down to mechanical contact and TIM spread from what I saw in MX-2 mounts which unfortunately is influenced by the individual processor IHS shape a bit too.

[Church]

- Why did the mount (bolts/springs/nuts) have such a difference?

Because apart from bow mount pressure is another important point of quality of thermal contact, at least with TIM. I'm guessing that with IX you would see less of improvement. I recall one comment/thought that EK Supreme HF might had won back then over Apogee XT because even while it's repeatability/mount consistency was worse, it's mounting sys did allow to screw in block tighter, with more pressure.

- Why didn't the rotation matter this time?

Who knows? I'm guessing clarifying might take armed with good knowledge of hydro and termo dynamics study in depth of this position agnostic block internals design, comparison to internals of blocks where improvement took place. I'm among those that lack such knowledge

However things are, i still feel somewhat dissapointed about vendors not testing/catching such big (relatively to close performance of top blocks out there) improvement about their products and not passing that info for customers. They don't want their products to win?
Whatever, still nice of you to test all possibilities and catch this interesting piece of nfo.

[stren]

I think it depends on a few things:

IHS shape, seemed like my 2600k liked the really sharp bow both directions.

Some blocks bow both directions while others only bow one way and which works best depends a little on the individual IHS shape.

Bow flattening is also influenced by mounting pressure, some all metal blocks are fairly rigid and can retain their bow under high pressure while others flex and can do worse with more. I can recall a few blocks that didn't like too much pressure.

The M6 is one that needed much more bow on my 2600k than factory and I saw huge improvements with shims on that processor to induce more bowing. Not sure on the 3930k though it doesn't like nearly as much bow.

It all comes down to mechanical contact and TIM spread from what I saw in MX-2 mounts which unfortunately is influenced by the individual processor IHS shape a bit too.

Thanks Martin - I guess the direction of the bow makes a lot of sense. I did notice the MX2-IX Delta was different for the CPU-380's rotation. It does make me think how useful the cpu block testing is when so much can depend on the luck of the draw is with the CPU IHS also. If we test with IX we can get repeatable results that take the mount out of the equation but that also seems unfair. I guess I like clear definitive answers, and the more I learn and the more I test the less clear the answer is!

- Why did the mount (bolts/springs/nuts) have such a difference?

Because apart from bow mount pressure is another important point of quality of thermal contact, at least with TIM. I'm guessing that with IX you would see less of improvement. I recall one comment/thought that EK Supreme HF might had won back then over Apogee XT because even while it's repeatability/mount consistency was worse, it's mounting sys did allow to screw in block tighter, with more pressure.

- Why didn't the rotation matter this time?

Who knows? I'm guessing clarifying might take armed with good knowledge of hydro and termo dynamics study in depth of this position agnostic block internals design, comparison to internals of blocks where improvement took place. I'm among those that lack such knowledge

However things are, i still feel somewhat dissapointed about vendors not testing/catching such big (relatively to close performance of top blocks out there) improvement about their products and not passing that info for customers. They don't want their products to win?
Whatever, still nice of you to test all possibilities and catch this interesting piece of nfo.

Thanks Churchy - I wonder if rotation is also affected by the thickness of the copper base of the block. Presumably the thicker it is, the less the rotation matters?

For now I'm moving on to the retesting the final few blocks that I was suspicious about my data on!

[bds71]

If we test with IX we can get repeatable results that take the mount out of the equation but that also seems unfair.

i don't know - i guess it depends on what results you're after. once i retire (in a couple years) i have always planned on trying to get into the testing arena. i have already been giving thought to certain things i would like to do. for one, i would love to do all my testing with IX - but as has already been stated, it's just not cost effective. so i thought instead of how to have *equal* testing with paste (since, with paste, so many more things can influence the end result - such as mount design/mounting pressure, bad mount (not properly centered), bow shape, amount of bow, differences in batches/bad batches). i think i may have come up with a solution - but that solution still takes the factory mount (partially/mostly) out of the equation and puts all the tested blocks on more or less equal footing. to me, i think that is a more fare (farer???) way to test blocks. but, what i want is to test the effectiveness of the block itself; to see which *block* is better/best rather than which *system* is best (or what company has the best current mounting system). if i understand what you are trying to do: you want to test the "whole package" which includes the mounting system *and* the block (?) personally, i don't think either method is fare or unfair - i just think they are designed to test different aspects of the same item.

[stren]

i don't know - i guess it depends on what results you're after. once i retire (in a couple years) i have always planned on trying to get into the testing arena. i have already been giving thought to certain things i would like to do. for one, i would love to do all my testing with IX - but as has already been stated, it's just not cost effective. so i thought instead of how to have *equal* testing with paste (since, with paste, so many more things can influence the end result - such as mount design/mounting pressure, bad mount (not properly centered), bow shape, amount of bow, differences in batches/bad batches). i think i may have come up with a solution - but that solution still takes the factory mount (partially/mostly) out of the equation and puts all the tested blocks on more or less equal footing. to me, i think that is a more fare (farer???) way to test blocks. but, what i want is to test the effectiveness of the block itself; to see which *block* is better/best rather than which *system* is best (or what company has the best current mounting system). if i understand what you are trying to do: you want to test the "whole package" which includes the mounting system *and* the block (?) personally, i don't think either method is fare or unfair - i just think they are designed to test different aspects of the same item.

Yeah I think Martin was testing with a standardized mount in order to take that variable out. I might be wrong though, maybe he can comment. There are so many variables, and it's impossible to remove them all to get a balanced playing field. My theme has been to try and test the block as it would be used by an end user which means more variables, but hopefully it keeps the data closer to what an end user might expect. However from the data you can see that in the end so many blocks are so close in performance that they are within the margin of error. I guess this is a function of how standardized the designs have become. Luckily at least two blocks stood out in the front. However would they stick out on Ivy Bridge with a different bow? Would they stick out on AMD? I also hoped to be able to find patterns to trace performance back to i.e. link performance to number of channels, or channel width, or base thickness, or channel surface area etc. I.E. How do the manufacturer's choose what parameters to use and are we now as optimized as we can be with the standard channel/jetplate architecture?

[Church]

Imho most reasonably possible scenario for average joe user - stock mounting sys with maybe non stock TIM. 2nd most possible - have parts from very old block new one was purchased to replace, with probably no interchangeability and/or new socket support. 3rd - having several blocks of current generation imho is too rare to count on to have. Mounting sys is what comes with block kit as a whole, and imho it's not enough to judge only usability/repeatability of mounting sys, but test block performance with other, as that is not what majority of users will be able to reproduce unless they are very few having big pile of blocks. At most what can be tried - tighten screws over manufacturer suggested limit in mounting instructions if it's not limited by construction.
So i'd leave to each block mounting sys they have, but test with same TIM. It's cheap enough and can be bought by many separately or come from old cooling solutions, including air ones.

[stren]

Imho most reasonably possible scenario for average joe user - stock mounting sys with maybe non stock TIM. 2nd most possible - have parts from very old block new one was purchased to replace, with probably no interchangeability and/or new socket support. 3rd - having several blocks of current generation imho is too rare to count on to have. Mounting sys is what comes with block kit as a whole, and imho it's not enough to judge only usability/repeatability of mounting sys, but test block performance with other, as that is not what majority of users will be able to reproduce unless they are very few having big pile of blocks. At most what can be tried - tighten screws over manufacturer suggested limit in mounting instructions if it's not limited by construction.
So i'd leave to each block mounting sys they have, but test with same TIM. It's cheap enough and can be bought by many separately or come from old cooling solutions, including air ones.

I also like this because anyone with a decent coolant temperature sensor that can be logged can buy some MX2 and run their own tests and see if they correlate. Even though correlation is almost impossible I'm still a big fan of the idea of people being able to try and reproduce data cheaply even if not easily

[Church]

I meant result replication not so much as actual retesting by users but rather that if testing way/testbed is made more similar to most common real life setups for majority of users, then your test results will become more useful and for wider share of LC users to actually guide WB purchase choice. If uncommon setup is tested, performed work will be less useful, more like theoretic exercise that still leaves out lot to guesswork/speculative estimation to do purchase choices.

[stren]

I meant result replication not so much as actual retesting by users but rather that if testing way/testbed is made more similar to most common real life setups for majority of users, then your test results will become more useful and for wider share of LC users to actually guide WB purchase choice. If uncommon setup is tested, performed work will be less useful, more like theoretic exercise that still leaves out lot to guesswork/speculative estimation to do purchase choices.

Yup this is very true also.

Running the BP summit right now, original data was for the block when rotated as suspected, the "normal" orientation is doing about 2C better placing it in the middle of the pack.

[stren]

Updated with Bitspower Summit data - original data was @90 rotation, new data is "normal" rotation, delta is ~1.3C between averages:



and just for fun - the best mount of every group of runs I've done:



Last spot check to do - the alphacool NexXxos, then on to checking some rotations!

[Church]

Funny to see 380 & UC1 on mx2 outperforming many blocks on IX
And HK3 doing that good? Have there been some silent revisions for this veteran block? :/

[stren]

Funny to see 380 & UC1 on mx2 outperforming many blocks on IX
And HK3 doing that good? Have there been some silent revisions for this veteran block? :/

Haha yeah, but bear in mind that's the best MX2 run, not the average of the MX2 runs But yes that's how good those two blocks are. Yes the HK3 still does very well with IX, but the MX2 performance is much worse. It may even improve further by rotating it 90 degrees.

[stren]

Updated with EK Supremacy rotated - nothing surprising - the rotation improves the performance but only 1C.

It pushes it to the top of the "middle" group but still doesn't threaten the CPU-380 or the Phobya block:



I like the confidence plot because it really gives you a feel for how close most of the blocks are even if there isn't really enough data points to support this as an accurate plot:



I think the conclusion we can draw from this is that two blocks perform above average and one performs below average. The rest are similarish.