Stren's 2012 CPU Water Block Roundup

[Martinm210]

+2 Flow rate info is nice to have and can be a value in plotting dT vs flow rate to understand flow sensetivity differences, but for stand block testing with a fixed typical condition, fixed pumping power and allowing flow rates to vary as they do in real use is much more desirable. Lower restriction blocks will net higher flow rates as they should. There is a very minor error included in the form of ignored pump heat dump with that but mounting errors far exceed heat dump anyhow so a fixed pump method is the more preferred.

Even if you go to the extreme in testing multiple pump conditons a dT vs flow rate chart still skews the results in favor of the more restrictive blocks. dT vs hydraulic pumping power or a simple low/med/high is a better comparison. Again it comes back to what gear and info most users have on hand. Pumping power using simplified units like Low=single D5 setting 1, Med = 35x 100%, High = 35x2 100% are understood where most users don't have a flow meter so flow rate doesn't speak to them very well.

Anyhow, I would not consider comparing at fixed flow rates if possible. It is useful for creating estimation data, but poor for most block comparison other than just supplimental info.

[tiborrr]

Agreed, testing at fixed flow rates is not okay since more restrictive blocks will have advantage. Test with a pump at it's highest setting (e.g. MCP 355 or D5 Strong) and that's it.

[stren]

+4 Yeah I'm not testing at fixed flow rates but at fixed pump powers.

For the test vs flow I'll be using IX like skinnee/vapor did. I'm using 7 PWM settings of the mcp35x2 to get the range. Data would be presented like this except with all the blocks (I'm only showing one because I don't want to show more than one yet)



For the regular testing vs 5 mounts using MX2 I'll just use one pump power which will either be 39% on the mcp35x2, or if swiftech want me to run the APD2 also then I'll swap out the mcp35x2 for a mcp35x for a more apples to apples comparison.

[Martinm210]

Nice! Maybe as a suppliment create a complimentary pumping power based chart where the X axis is pump setting instead of flow rate. I like seeing the flow rate too for estimating purposes, but for comparison prefer leaving X in some sort of pump power unit. Using flow rate only will make the more restrictive blocka look better since the curves stay to the left and lower than lower restriction blocks. That could be useful in comparing amore complex loop, but its also very possible to have a lower dT with the low restriction block produce a curve further right and most users will interperet lower being better at like flow rates essetially fixing the flow in their head and completely missing the pump power data points being shifted along the X axis. The problem is that comparison is two different pumping powers. That's the problem I have with comparison of dT vs flow rate charts. They have all the data correct but hard for most users to read without automatically comparing at equal flow rates. I think that is why Vapor produced both types of charts...one where X is flow and another where x is pump power. I have also seen Cathar convert pump power to energy as another means to fix this problem.

Anyhow, very cool. My hat is off to you for having the patience to evaluate pumping/flow sensitivity. This is one thing that is much easier to do with IX as trying to do 5 mounts using paste over many flow rates is time prohibitive...at least for my level of patience which gets smaller and smaller as I get older..lol

Great work!!

[stren]

Nice! Maybe as a suppliment create a complimentary pumping power based chart where the X axis is pump setting instead of flow rate. I like seeing the flow rate too for estimating purposes, but for comparison prefer leaving X in some sort of pump power unit. Using flow rate only will make the more restrictive blocka look better since the curves stay to the left and lower than lower restriction blocks. That could be useful in comparing amore complex loop, but its also very possible to have a lower dT with the low restriction block produce a curve further right and most users will interperet lower being better at like flow rates essetially fixing the flow in their head and completely missing the pump power data points being shifted along the X axis. The problem is that comparison is two different pumping powers. That's the problem I have with comparison of dT vs flow rate charts. They have all the data correct but hard for most users to read without automatically comparing at equal flow rates. I think that is why Vapor produced both types of charts...one where X is flow and another where x is pump power. I have also seen Cathar convert pump power to energy as another means to fix this problem.

Anyhow, very cool. My hat is off to you for having the patience to evaluate pumping/flow sensitivity. This is one thing that is much easier to do with IX as trying to do 5 mounts using paste over many flow rates is time prohibitive...at least for my level of patience which gets smaller and smaller as I get older..lol

Great work!!

Sure I can do that- I guess with PWM settings it's somewhat less intuitive than the 5 settings on a D5 Vario but yes it may help the understanding

[snowyowl]

Can't wait to see the results.....subscribed

Would you consider including a custom cnc waterblock in your roundup? (sent you a Pm on this)

[stren]

Replied

[OCAUfedexpress]

I voted for ek 370 cos it is sexy and i have one lol!

apparently this guy tested

1, XSPC RayStorm
2, Koolance 370
3, EK Supreme HF

and the 370 won! :P

http://www.overclockers.com/forums/s...d.php?t=709078

[jogshy]

I'm very interestered in this test because I must buy a block soon
Btw, why are you using Indigo Extreme and not Coollaboratory Liquid Pro/Ultra? Problems with corrosion? And... are you going to use a lapped CPU?

Some advices:
1. Test with the block in the four different orientations.
2. It would be fantastic if you can test them also using a standard ceramic TIM ( applied in 4 different configurations: X, |, full spread, dot ).
3. I've seen you use Prime to test in youtube. I would better use the latest version of LinX compiled with AVX because it runs much hotter.

pd: voted for EK... because I'm sure it will show its... supremacy

[rge]

He is using Indigo, and mentioned he may use standard tim afterwards. But four different orientations of blocks would quadruple the already massive time investment in testing that many blocks. The manufacturer should know best orientation and supply it and test done with that orientation. Testing 4 different tim applications is unnecessary and again quadruples the time for testing...you can get good tim coverage with any of the methods, you just want to use the same method that you can reproducibly and reliably get good coverage. Indigo obviates that anyways, though realize your were speaking after his indigo test with regular tim.

Linx is a variable load program, and will give less accurate results than prime fixed at same ffts. If your using accurate temp probes, running prime (same ffts over and over) you can get repeatable results withing 0.1 to 0.2C with same mount, and no more than 1/2 degree between different mounts (throwing out all obviously bad mounts and controlling mount pressure with hard stop point/etc). Using linx that accuracy decreases as the load/power varies during testing, hence linx is not a good testing program for cpu blocks that only differ by 1-2C. Also using 64 bit prime v27.7 with avx and ivy cpu, my max cpu temps and power draw are similar to linx.

[stren]

I voted for ek 370 cos it is sexy and i have one lol!

apparently this guy tested

1, XSPC RayStorm
2, Koolance 370
3, EK Supreme HF

and the 370 won! :P

http://www.overclockers.com/forums/s...d.php?t=709078

Interesting!

I'm very interestered in this test because I must buy a block soon
Btw, why are you using Indigo Extreme and not Coollaboratory Liquid Pro/Ultra? Problems with corrosion? And... are you going to use a lapped CPU?

Some advices:
1. Test with the block in the four different orientations.
2. It would be fantastic if you can test them also using a standard ceramic TIM ( applied in 4 different configurations: X, |, full spread, dot ).
3. I've seen you use Prime to test in youtube. I would better use the latest version of LinX compiled with AVX because it runs much hotter.

pd: voted for EK... because I'm sure it will show its... supremacy

Yes prime is the latest

IX because I had experience with it previously. The lack of curing time is very useful for taking data vs flow rate. Vapor had also used it succesfully for his testing. MX-2 will be the TIM for regular TIM tests. I put a lot of weight in the way martin/skinnee ran their tests and don't like to reinvent the wheel in order to discover the same mistakes lol.

No - no lapped CPU, most people don't lap any more so the normal TIM test will be more relevant. Plus most blocks have some amount of bow that is worth seeing. If I lap the CPU I should lap the blocks too. There is a chance of course that my block has "unusually" large or small amounts of bow. But I think we'll have to take that risk. For IX of course it wouldn't have mattered.

Four orientations is over the top as most blocks are symmetrical anyway - If I had less blocks I would have done two, but this is just too many to do that with.

Same with the TIM application, I have 15 blocks - I need to take 7 data points on each IX application, plus 5 mounts with regular TIM. If I then add in 4 different TIM applications it's going to get out of hand and take all year.

I'm now finally taking Martin's suggestion of doing MX2 first - this way I have a baseline to compare to IX such that I know quickly whether my IX mount is bad rather than waiting two days to complete a data set before I find out.

Family size pack of MX-2 that arrived today:



Also a pic of the test setup - I was running out of room so used the guest bathroom as a temporary desk through the addition of a plywood sheet:

[musicfan]

Nice pictures stren. Looks great. Thanks for keeping us informed. Please keep it fun. I commandeer the guest bathroom every 6 months for PC draining & leak testing too. Wonder how many guest bathrooms become workshops for water-coolers?

[SonDa5]

Looking forward to the results.

[stren]

Thanks guys - testing is still in progress. I switched to MX2 so that I had a baseline to compare IX to so that I knew I had a good mount before spending 2 days running tests vs pump settings. I'm on the 3rd block out of 15 (5/15 done on IX). But if I'm on top of things I might be able to do 2 blocks a dayand then release the MX2 results. Then I can go back to IX and finish up

[Xphobe]

Well, I have to buy a water block at latest tomorrow to finish my build. Seeing as how I cannot wait any longer for this roundup to get done, at least according to THIS review, the XPSC's Rasa has the best performance to flow ratio. Would it be possible to confirm that the Rasa is one of the top performers from the testing done thus far?

[stren]

Well, I have to buy a water block at latest tomorrow to finish my build. Seeing as how I cannot wait any longer for this roundup to get done, at least according to THIS review, the XPSC's Rasa has the best performance to flow ratio. Would it be possible to confirm that the Rasa is one of the top performers from the testing done thus far?

Nearly all of the blocks in that review are pretty old now, and some of them like the supreme HF have had new jetplates released that improve things. I'd expect the rasa to have among the worst performance in this roundup. I haven't tested the rasa yet but if you're going for XSPC I would suggest to at least get the raystorm. Martin has reviewed some of the blocks in this review already so I would go check out his last reviews if you can't wait for my results.

[M.Beier]

Good work Simon, I hope you have solved your IX TIM reflow issues. I would also recommend testing the blocks using a single pump as this is the most common application.

Agree..

Pointless to use multiple pumps for the verdict..
Maybe use a D5 and notify performance at each step? or 3 of them.

[stren]

Results of MX-2 and some IX on the first page now:

http://www.xtremesystems.org/forums/...=1#post5123416

I'll be on vacation for the next two weeks so apologies for any slow replies to questions.

[musicfan]

Wow. Thank you very much stren. It is fun looking at your results. Have a great vacation.

[Church]

Woah, thanks for work & congrats on finishing .
Enormous work & nice result representation. I hope that you didn't get tired from testing and won't go in retirement so that we can see more from you in future
lol, yeah, nor did i expect nexxxos as winner, nor supremacy as not among 1-2 best ones. Surprises.
P.S.
Any plans on testing on IB@lga1155? I suspect blocks with more distinctive bow (or with tunable bow) might gain there a bit.

[stren]

Wow. Thank you very much stren. It is fun looking at your results. Have a great vacation.

We will

Woah, thanks for work & congrats on finishing .
Enormous work & nice result representation. I hope that you didn't get tired from testing and won't go in retirement so that we can see more from you in future
lol, yeah, nor did i expect nexxxos as winner, nor supremacy as not among 1-2 best ones. Surprises.
P.S.
Any plans on testing on IB@lga1155? I suspect blocks with more distinctive bow (or with tunable bow) might gain there a bit.

Thanks churchy! Honestly I may skip IB, by the time I'm done with the retests on this, as well as the DT Direct and IX testing it'll probably be time for the replacement CPU for IB (is it haswell next?) to be out and maybe I might test these blocks again on that! We'll see though - I don't want to get burned out or annoy my wife too much either! One thing this testing has taught me is enormous respect for the guys who have gone before me!

[Church]

BTW, has anyone ever tested 2 or more blocks of same type? As in if variation between several exemplars of same block/revision? IMHO few of discrepancies between different results of different testers might be explained by lucky or bad particular block as well. - If that variation can reach 1 degree, and when we see top blocks within 0.5-1 degree? Pity that it might allow some play with vendors sending in "hand-picked" blocks and about the only way to workaround it is to not take in directly any block, only money for which those can be ordered at some common shop. But then again nothing will leave out pure luck. Testing several pieces of all of the blocks will rise required time/work to unacceptable levels though. :/
Hehe, maybe i'm trying to think too deep

[stren]

BTW, has anyone ever tested 2 or more blocks of same type? As in if variation between several exemplars of same block/revision? IMHO few of discrepancies between different results of different testers might be explained by lucky or bad particular block as well. - If that variation can reach 1 degree, and when we see top blocks within 0.5-1 degree? Pity that it might allow some play with vendors sending in "hand-picked" blocks and about the only way to workaround it is to not take in directly any block, only money for which those can be ordered at some common shop. But then again nothing will leave out pure luck. Testing several pieces of all of the blocks will rise required time/work to unacceptable levels though. :/
Hehe, maybe i'm trying to think too deep

I do have two of each of DT's blocks for my build, and two raystorms laying about too. I too worry about manufacturers cherry picking samples, which is why I do like it when a vendor sends me multiple blocks from their shelves like PPCS did

[Church]

Yes, you have thought on that too .. but now multiply all tests for 15 blocks with 5+1 mounts by extra two or three times with 2-3 blocks of each kind. You see where i'm going to? .
Hats off to you already for taking all the other suggestions to testing methodology and such from us, despite it giving LOT of extra work.

[scamps]

Thanks for this pile of work!
Surprising results that I can´t reproduce on my socket 3770K (socket 1155) at all.