Stren's 2012 CPU Water Block Roundup

[stren]

Another interesting fact is that 100% of all color based charts discriminate against color blind people

The MIPS won the flow rate challenge.

Yeah this is why I do all the charts I can as single color (blue too rather than red or green). The busy charts are ones that I can't do much about, but those are the ones I post process and show you more useful versions because I know they're hard to read.

My boss is very color blind so I've been getting used to trying to work around it as best I can. When I update things I'll try and improve those as best I can.

 

[stren]

I think you forgot to include the sample size when calculating the confidence intervals. Confidence intervals are calculated from the standard error. The formula is

X +- 1.96*s / sqrt(n)

where n is the sample size, X the sample mean, and s the sample standard deviation.
The row should be

DT Sniper 41.40  0.25 0.25

Note: I am ignoring the fact that 5 is a very small sample size (statistically speaking) and thus should probably use the t-distribution, so we should preferably use 2.77 instead of 1.96.

Finally got round to adding the performance intervals - thanks to mdatmo and rrrhal for the info on how to do this. Maybe I made a mistake in not giving the raystorm an award, particularly given the price.

What this does show is just how close things are.

Also took some internal shots of the summit, cpu-370 and sniper. Hopefully I'll get those put up tonight.

Also compiled a price list with links, but having difficulty embedding google docs, any ideas?

<iframe width='580' height='375' frameborder='0' src='https://docs.google.com/spreadsheet/pub?key=0AhqJsdm9xVf4dFdfSktQaDBpNEtVRVZDc3F4RmlEOEE&output=html&widget=true'></iframe>

 

[stren]

Updated results [http://hardforum.com/showpost.php?p=1038985356&postcount=18]post (#18)[/URL] with

- DT Direct
- Swiftech Apogee HD under IX
- EK Supremacy Retest under MX2
- "Mystery Block" under MX2

The "Mystery Block" is a new block that is being released on monday. You may notice that it's the new top performer. Can you guess who it's made by? :thumb:

I'll release more details of it on Monday including photos!

Don't forget to like my facebook page for exclusive updates and giveaways coming soon!

 

[t_ski]

I'm out of the loop - any hints?

 

[stren]

I'm out of the loop - any hints?

Lol well it's monday so I can tell you:

Koolance CPU-380

Pics here:

http://hardforum.com/showpost.php?p=1038985345&postcount=10

I also updated the results post and all the pics with spoilers so it's a bit easier to see what's going on now!

If you want an individual review you'll have to head to my website:

http://extremerigs.blogspot.com/p/koolance-cpu-380i-cpu-water-block-review.html

Don't forget that you can follow me at www.facebook.com/extremerigs

 

[stren]

Jeez it's quiet up in here

Vanilla raystorm vs flow/pump complete, full copper now on the rig:

 

[hammerforged]

Ha it is pretty quiet over on hard these days for some reason.

Your OCN post has 30+ pages (60 pages in defualt view).

 

[stren]

Ha it is pretty quiet over on hard these days for some reason.

Your OCN post has 30+ pages (60 pages in defualt view).

Yup which is why OCN are now trying to make me pay them $30 in "sponsorship" fees. So I guess I won't be posting there anymore so maybe this thread will get more traction in future.

 

[hammerforged]

Yup which is why OCN are now trying to make me pay them $30 in "sponsorship" fees. So I guess I won't be posting there anymore so maybe this thread will get more traction in future.

Oh wow I was not aware of this. They should be helping you out You bring quite a few hits to their site. I know most the time if someone is asking about CPU blocks I just tell them to search for your roundup. Shame....

 

[FaRKle0079]

OCN seems to have a lot less visitors and useful information since they changed to a for-profit model. Signal to noise ratio there dropped heavily which is why I barely go there anymore.

 

[Neo Zuko]

So where is my 380i Indigo Extreme results?? You didn't think I was going to let you off the hook just because your Overclock thread was canned? Haha just kidding.

Honestly I used to chat here all the time but then overclock had profile pics and system build tools and rig pic uploads... But the more interesting discussion usually happens on hardforum anyway.

 

[stren]

Oh wow I was not aware of this. They should be helping you out You bring quite a few hits to their site. I know most the time if someone is asking about CPU blocks I just tell them to search for your roundup. Shame....

Yeah they approached me wanting me to pay them $29, then when I argued the plentiful case against it they said that it didn't matter because they wanted exclusivity on the content and because I post here and to XS that the content had to go. So the thread has been locked and all the information removed.

OCN seems to have a lot less visitors and useful information since they changed to a for-profit model. Signal to noise ratio there dropped heavily which is why I barely go there anymore.

Lol that's a good way of putting it.

So where is my 380i Indigo Extreme results?? You didn't think I was going to let you off the hook just because your Overclock thread was canned? Haha just kidding.

Honestly I used to chat here all the time but then overclock had profile pics and system build tools and rig pic uploads... But the more interesting discussion usually happens on hardforum anyway.

Ah I was dealing with OCN mods *all* of yesterday evening. I didn't even get the 380 on the bench yet. Hopefully today, so maybe results tonight or sunday because I'm busy all day saturday.

Here's the updated IX temps vs flow/pump results with the full copper raystorm:

You can see that the raystorm vanilla and copper basically perform the same (within the margin of error), unlike MX-2 where the full copper block is a good bit worse. This lends credence to the materials making a difference to the bow.

 

[Neo Zuko]

Yep, I don't understand why the full copper is worse, if anything it should be better.

 

[stren]

Yep, I don't understand why the full copper is worse, if anything it should be better.

With MX2 you mean? With IX it's basically the same.

Full copper tops don't add any significant performance difference because the increase in surface area is small, and it's not even attached to the CPU directly. So the heat flows through the base and gets poor thermal contact with the copper top because there is an o-ring in the way. Then it has minimal surface area to dump heat to the water with.

For MX2 it's different because the bow and mount are different. The mount plate is aluminum rather than plastic so is less flexible. The bow is different because copper has different mechanical properties to acetal or delrin or whatever the raystorm uses. Bow on a block is usually created by mechcanically stressing the base often with a jetplate. The difference in material means the bow is different and that affects the MX-2 results I believe.

This is just my theory though.

 

[rge]

Yep completely agree Stren, vapor and others have tested different material tops and come to same conclusion, the bow is simply different, tops dont improve or detract from cooling directly, but indirectly from altering bow. So metal or the acetal top could perform better on any given cpu, depending on shape of IHS, die type, etc, and which provided a more beneficial bow for a given cpu. And stren has point with brackets as well, plastic bracket bends more and will alter pressure pattern, hence contact pattern. Also one base may simply be better than another as seen when testing 2 identical raystorms (exact same tops) and seeing .5 to 1C difference.

And wth is up with OCN. I started posting there because xtreme is too heavy handed with deleting opinions, and now OCN demanding pay for testing or demanding it only be posted there as an exclusive, ie their new money making (more likely losing) policy. Not a policy that is going to grow their site or be profitable in the long run. Just will kill all the testing threads, except for small/single bias review/testing that is paid for by one single manufacturer, ie advertising, that few would care about.

So much for OCN.

 

[Neo Zuko]

Indeed, I'm deciding just where I want to do my CaseLabs Merlin Build Log...

Interesting, on the tops affecting bow, I wonder how the Apogee Drive 2 handles bow. That block/pump has a specific 2011 model though.

 

[stren]

Yeah rge I'd heard similar things before so it's good to get confirmation. The blocks were so close in testing I'm wondering really how much manufacturing variation there is. Does anyone have reliable data where the margin of error is small enough to really show how big the variation is?

CPU-380 got a good reflow with the new instructions, now running vs different flow/pump settings:

 

[Neo Zuko]

CPU-380i is my next block. Well it's my first block not counting heatsinks and the H80.

I wish you could have tested the Apogee Drive 2... All that talk about it being the only block that "Adds" rather than "Subtracts"... Like even if you have a MCP35X2, adding sounds better than subtracting.

 

[dcds1]

Well the funny thing about the poll vs the results. The top 2 blocks chosen to be the best were in the end just middle of the pack contenders. So it thus leaves a person to wonder what other recommendations given in the water cooling section is just hype.

 

[x509]

CPU-380i is my next block. Well it's my first block not counting heatsinks and the H80.

I wish you could have tested the Apogee Drive 2... All that talk about it being the only block that "Adds" rather than "Subtracts"... Like even if you have a MCP35X2, adding sounds better than subtracting.

And in plain, simple, everyday English, exactly how does a block "add" instead of "subtract," and what is the block "adding?"

 

[Neo Zuko]

It's hard to find reviewers who test on the same level as Stern. Most reviews I read are very subjective and not very usefull. Most often I'm like ok it's not a bad option but how does it compare to my other options? And that is often missing, like the product exists in a bubble world alone. Or they will make one token comparison. So its awesome when I come across a through scientific large comparison review such as this.

 

[Neo Zuko]

And in plain, simple, everyday English, exactly how does a block "add" instead of "subtract," and what is the block "adding?"

It has a built in MCP35X Pump, so it pumps/adds to the water flow instead of restricting/subtracting from the flow.

 

[stren]

CPU-380i is my next block. Well it's my first block not counting heatsinks and the H80.

I wish you could have tested the Apogee Drive 2... All that talk about it being the only block that "Adds" rather than "Subtracts"... Like even if you have a MCP35X2, adding sounds better than subtracting.

Swiftech actually asked me to review that but it's hard to do it in an apples to apples sense. Do you run it with one less pump than I have for the other blocks or do you run it with the same mcp35x2 I have in the loop, in which case that block has three pumps which seems a bit unfair.

Given that it's the same base as the HD coupled with a mcp35x then it should perform identically to the HD but at higher flows. So let's say you were comparing the Apogee HD with 1 mcp35x vs the APD2 + 1 mcp35x. The Apogee HD should perform like the Apogee HD test at 39% PWM, the APD2 should perform like the Apogee HD test at 100% PWM. In other words about 0.4C difference. I hope that helps, it's hard to compare a integrate block + pump to other blocks you see.

Well the funny thing about the poll vs the results. The top 2 blocks chosen to be the best were in the end just middle of the pack contenders. So it thus leaves a person to wonder what other recommendations given in the water cooling section is just hype.

Well different blocks perform differently on different sockets. Even so there are discrepancies that I can't explain between Martin's tests on a 3930K and mine. Particularly with regard to EK. At the end of the day so many blocks are very close that it becomes hard to distinguish reliably and accurately between them all.

 

[stren]

CPU-380 vs flow/pump results - still kicking butt, in fact kicking so much I had to readjust all my axis' to have more range:

 

[Neo Zuko]

True it may not be fair to test an Apogee Drive 2, but that in and of itself is not a reason to pass on it as a system builder. Sure there will be modest loops with it where that is the only pump, or where it counts as the second pump for redundancy. But I'm most interested in how it would perform in addition to a MCP35X2. Basically eliminating CPU restriction at the source. I asked around and everyone kept saying three pumps were unnecessary, but I think it might affect CPU and post CPU restriction if you already have another MCP35X2 in the loop.

I think the Apogee drive 2 should be tested in a separate review but I still want to read it. I don't want to build a fair system, I want to build the best system!!

 

[Tsumi]

3 pumps in series are going to act like 3 pumps in series, regardless of where they are positioned in the loop. Extra tubing increases restriction, and having a dual-top reduces that restriction. However, if the pump sits on top of the CPU block, then there's no added restriction. Additionally, restriction due to an extra foot or two of tubing is minimal. This is due to basic fluid properties of incompressible fluids. I'm sure there are other factors, but those factors are minimal.

Pumps in series do not add extra flow-rate. Only in parallel are they able to do this, and as Martin's testing showed, running pumps in parallel has unintended consequences that actually reduce flow rates. The 35x has a maximum throughput of ~3.5 GPM. Adding more pumps in series will only get you closer to 3.5, but it's an exponential decay curve. As you get closer to the max, you need significantly more power for a limited gain (i.e. from 2.0 to 2.5, you might just need one extra pump, but from 2.5 to 3.0 you would need 2-3 extra pumps). Also, as you can see from the graphs, the curves also exhibit an exponential decay. The benefits of going from 2.0 to 2.5 is minimal (~0.4 C), and 2.5 to 3.0 will be even less.

 

[Neo Zuko]

Edit

 

[Tsumi]

The fillport should go directly to the fillport of the reservoir, otherwise the reservoir won't do its job correctly.

 

[stren]

Tsumi is right, more than 2 mcp35x's won't buy you much. If you really want to go high flow for that last tenth of a degree you'll need an Iwaki RD-30, but even that could only push 3.5gpm through a 5 Noz and radiator at 29V. The benefit above 2.5gpm is small. On older blocks there was even a point where high flow made temps worse. I have yet to see that point but I'm not pushing crazy high flows in testing here, it's more a range that I thought would be common through to "normal" amounts of crazy

 

[thequn]

If anything at that point the 3rd pump with only increase the overall heat of the loop by its wattage of time, which in turn would lower its overall effectiveness as far as gain as we can already see this with 2 pumps in a loop.

 

[Jorona]

If anything at that point the 3rd pump with only increase the overall heat of the loop by its wattage of time, which in turn would lower its overall effectiveness as far as gain as we can already see this with 2 pumps in a loop.

No. The thermal impact of the pump would be inconsequential.

 

[thequn]

You are correct- for the normal people user but in strict testing everything is taken into account and factored.

 

[baldrik]

You are correct- for the normal people user but in strict testing everything is taken into account and factored.

Perhaps if you were talking D5s, but not really with MCPs.

 

[Neo Zuko]

Edit

 

[dsmwookie]

I just wanted to say, thank you! This was a nicely written article and the charts were easy to read (except the one using a bunch of colors) I ve noticed some reviewers using symbolts every 3-4 centimeters to help differentiate the different lines. Other than that it was a great read and helped a lot. Pretty sure I will end up with the Apogee HD, Koolance 380, or the Apogee Drive 2 (leaning on this one).