White Water review at oc

[AngryAlpaca]

http://www.overclockers.com/articles1262/

Based upon other reviews, I'm not sure what to think of Joe's tests anymore.

[jaydee]

I got the White Water ahead of the TDX in all my tests.

[Etacovda]

The polar flow 'i designed this using a spirograph' block is better, apparently.

Joes testing has clearly become 'interesting'.

Is this on the 1" square die? if so, its hardly surprising. He needs to be shot for cross-corrolation of data across two entirely seperate die sizes...

[Cathar]

Originally Posted by Etacovda

Is this on the 1" square die?

Could it be? :shrug:

That would explain a lot. The WW's cooling patch is ~17mm x 30mm in size, as is suitable for real CPU dies, whether they be IHS capped or not. If the data is an extrapolation from a 1" square die then that would explain the less than stellar performance.

[Ls7corvete]

Is this die size thing pissing anyone else off yet?

[pHaestus]

Those results show the MCW5002 as being superior to the whitewater design too. That doesn't correspond to any testing I've done ever. That suggests that either the die size issue is a large one, or that there's something about the Cooltechnica WW that makes it perform differently from the DTek or Little River ones.

[jaydee]

I got the exact same WWLE he tested and my numbers always show the WWLE ahead of the TDX. In fact my numbers scale pretty close to yours. I think the main difference is the jet is slightly smaller than the original LRWW and DTek WW. But that would only improve performance.

[Bundles]

I think overclockers.coms testing has went massively downhill, the tests seem to be getting more and more breif and inacurate.

If you read that, then read the Swiftech storm review, they could be copied and pasted from the same notepad txt file, and then just have the odd detail edited a bit.

[Eddy_EK]

I think that Joe doesn't give its best these days in tests.
With all respect to Joe.

[dacooltech]

Originally Posted by pHaestus

That suggests that either the die size issue is a large one, or that there's something about the Cooltechnica WW that makes it perform differently from the DTek or Little River ones.

well pH, you tested the D-Tek WW that Cooltechnica manufactured before ...
and you also have the same White Water Limited Edition with a copper top as well, so you can test it and enlighten us hopefully soon

P.S.: WoW is bad

[dacooltech]

Originally Posted by Cathar

Could it be? :shrug:

That would explain a lot. The WW's cooling patch is ~17mm x 30mm in size, as is suitable for real CPU dies, whether they be IHS capped or not. If the data is an extrapolation from a 1" square die then that would explain the less than stellar performance.

hmmm... well that still doesn't explain me why the WW clones perform better on the same die and exact same test bed though (which also doesn't correspond to any other testing that's been ever done before) :shrug:

[Cathar]

Originally Posted by dacooltech

hmmm... well that still doesn't explain me why the WW clones perform better on the same die and exact same test bed though (which also doesn't correspond to any other testing that's been ever done before) :shrug:

I think you misunderstood what Etacovda was saying. He was postulating that JoeC, as of about 5 weeks ago when he stated that his old small-die testbed had broken, is now perhaps using his newer 1" square die and extrapolating results as per the content of this recent article by JoeC => http://www.overclockers.com/articles1259/

As in, is it possible that all blocks tested more than 5 weeks ago were all done in the original small die testbed, and all newer tests in the last 5 weeks (including the Storm) are now being done on the newer 1" sq. die and extrapolated?

That is what is being proposed. I truly don't know, but I do know that the WW would not fare too well on a 1" square solid heat die, but only because a 1" sq. die is not how CPU's distrubute heat, not even IHS capped CPU's.

[dacooltech]

Originally Posted by Cathar

I think you misunderstood what Etacovda was saying. He was postulating that JoeC, as of about 5 weeks ago when he stated that his old small-die testbed had broken, is now perhaps using his newer 1" square die and extrapolating results as per the content of this recent article by JoeC => http://www.overclockers.com/articles1259/

As in, is it possible that all blocks tested more than 5 weeks ago were all done in the original small die testbed, and all newer tests in the last 5 weeks (including the Storm) are now being done on the newer 1" sq. die and extrapolated?

That is what is being proposed. I truly don't know, but I do know that the WW would not fare too well on a 1" square solid heat die, but only because a 1" sq. die is not how CPU's distrubute heat, not even IHS capped CPU's.

man if that's the case, then I definitely misunderstood what Eta was saying...
IF true, that's the craziest thing that I ever heard... that's apples to bananas comparison... how could that be possible?

[dacooltech]

JoeC said he tested the WW-LE on the small die...

[Les]

The more test data I see, the more I ignore, and the more I rely on my simulations.
Attached predictions for all WWs
Notes:
"(C/W)in" is *T(wb face) - T(water inlet))/(Watts absorbed by water)". Excludes TIM and sensor offset. Includes allowance for water temp rise in wb.
JoeC's die was considered to be 140sq mm (link) prior to Aug 2005(link)

Edit
Updated graph. ref

[Eddy_EK]

There is another rewiew of that block at OC.

I can't believe that such a small difference could make so big difference in preformings. :shrug:
Could dacooltech tell us what was the differences between those bloch to achieve that big difference... :shrug:

[Cathar]

Originally Posted by Eddy_EK

There is another rewiew of that block at OC.

I can't believe that such a small difference could make so big difference in preformings. :shrug:
Could dacooltech tell us what was the differences between those bloch to achieve that big difference... :shrug:

Indeed. The 1.2mm fins vs 0.8mm channels is the incorrect way to implement the design, especially if the channel height hadn't been changed to compensate. I hadn't realised that the designs were different. If Bruce had contacted me I could've told him that such a configuration would've been less than optimal. Still, a ~0.03C/W difference seems a little large to me - may have been some other variations present.

The WW overall is a very balanced design. Bruce/Cooltechnica acquired the machining duties for it from DTek, but it's also why every other WW Clone uses close to the exact same fin/channel dimensions that I originally established over 3 years ago. It's simply what works best. There are things that I'd change slightly, after 4 years experience with designing blocks, but I won't disclose them here.

The results shown now have me ever more convinced that something was definitely wrong with the Storm as tested. The pressure drop was close to double what it should be, and the results at 1gpm were not as good as it should be in comparison to the old WW Classic design.

[Les]

Originally Posted by Cathar

Indeed. The 1.2mm fins vs 0.8mm channels is the incorrect way to implement the design, .....

It models ok.


Edit
Updated graph. ref

[Cathar]

Les, there are effects due to the jet impingement that play out sub-optimally with the jet height of 4mm and using 0.8mm channels. Also the wall width vs wall height is a little low. These are all things that I noticed in the development of the design originally.

Overall I would not have thought it would make more than a 0.01C/W difference though, hence my statement that something else is also probably going wrong.

[jaydee]

Which version of the block do I have Bruce?

Maybe this is why I got the WWLE Closer to the TDX than pH. I still have it better than the TDX in my latest results with the new Silver TIM though.

[jaydee]

I see what the problem is. I don't think Bruce changed the design. The problem is in the machining and I see why. They are using slit saws to make the channels instead of end mills. I took mine apart and it clearly shows the slit saw marks. My trusty micrometer tells me the fins width on mine is .82mm, .88mm, .88mm, .84mm, .82mm, .84mm and .88mm. Channel width varies from .9mm to 1mm....

If I were Bruce I would be chopping machinists heads off.

[Eddy_EK]

About difference.
I see only smallest difference in chanel width.
And the 2nd block is durtier

Jaydee: I don't think that manufacturing with slit saw could made such difference. Anyway I can't see how to make such channels otrherwise than with slit saw.
Making that block with end mill would take about 30 minutes just to make those channels.

Perhaps the copper material is difference, and the worst is made out of les condustive copper :shrug:

[pHaestus]

The DTek wws were made with slit saws too. The big difference you'll see with using an end mill will be a decreased pressure drop not a change in the C/W for a particular fixed flow rate. At least that's what I saw when I tested a WW variant made with an endmill instead of slit saw

[jaydee]

The tooling should not change anything if done right. Providing the tooling has the same width on the cut.

Main problem with slit saws is any little vibration or imbalance will make the channels wider than the width of the blade. Thay can become out of balance a lot easier than end mills especially if ran hot. Probably not the case here though. That much of a change was probably some other issue. Still shouldn't change the results that much though.


I don't see it taking 30 minutes to mill a white water with a end mill. The channels are not that deep. 2 passes at 1IPM would only be about 20 minutes. A good mill should be able to run at least 5IPM conservatively with a 1mm end mill in Copper.

It would be about 16" total (just the channels) with 2 passes. 32" with 4 passes. Should be under 10 minutes on a good mill. Slit saw would probably cut that in half as you should be able to make the entire channel in one pass. Anyway Bruce or Cathar can correct me if I am way off as they both done it each way.

[Cathar]

You do a depth pass of 2.0mm with a 1.0mm end-mill in copper? Sounds pretty ambitious to me. Not saying that it can't be done, but typically faster to do 0.5mm depth cuts per pass and feed at a higher rate, or at least that's what my machinists told me. Had problems enough with the end-mills snapping as it was with 0.5mm depth passes. One end-mill bit lasted about 15-20 blocks.