Hey, now there's a good idea! ;)

Ok, everyone who wants to send me an XMas present, PM me! ;)

The cascade that they tested was not silver though. They only tested the copper version.

The problem wasnt that the casecade was the copper version. More a problem that the reviewer used a weak WEAK ehiem 1046 pump that just isnt enough for the casecade to really shine. Chances that the cascade with a 300~gph pump would pull ahead since the cascade depends alot on impingement~jets of water hitting the cups.

The murk is actually a good design. I love to see how it performs with bit stronger pump. Reminds me of an Intel white paper I seen.

Yes, I do beleive the murk would perform much better with a larger pump also. I was just meaning to say the cascade should have more to gain in performance. But that is a good point also....we dont know exactly how good the murk works with a normal 300gph pump many use.

Of course we're assuming that the testbed is producing reliable results.

Single mount testing only.

Uninsulated heating element favoring smaller footprint blocks.

Look at the results of the WCP testbed to most other trusted testing results and one can quickly see that something is strongly amiss.

What kind of gph were each of the waterblocks operating under with the ehiem 1046? What kind of mounting pressure were the blocks mounted under? And why did they use the ehiem 1046 of all things instead of a more standard 300~gph pump like a maxi-jet 1200, via 1300 and ehiem 1250?

I liked this statement:
seems like people have forgotten about that form of information printed on a thin white substance called paper. Libraries are a good source of information and usually more reliable than google.

Perhaps the most defining paper that I used to launch from for Cascade development:

http://web.mit.edu/lienhard/www/laiche.pdf

The 4xID thing isn't just a theory. There's a good deal of research behind it if one bothers to look. Things are a little bit different though due to the use of the cups. The paper also focuses on pretty high jet velocities (~45m/s) and using about 80PSI pressures.

It should be noted that I targetted the Cascade's design for 4-10LPM flow rates, or roughly 2.5->6 m/s jet nozzle velocities, and 2-10PSI pressure-drops across the block. The Cascade will continue to show good performance improvements right through that range of pressures.

If I were designing the Cascade to work well on a super low-flow testbed barely mustering 0.5PSI PD across the block, then I would have done a large number of things differently.

The test procedure followed by Phaestus, in the wake of BillA's excellent work, is a shining example of why single point testing is utterly useless. Had Phaestus decided to focus on just a 0.5GPM flow rate we'd all be left thinking that the Maze4 was a piece of crap, and the RBX is just average, and indeed this is just what the WaterCoolPlanet testbed is doing, and claiming to be able to rank the world's waterblocks. Rather interesting too is the ~2C difference between the RBX and the Cascade for Phaestus at 0.5GPM, yet just a 0.5C difference at WCP. Where'd the other 1.5C go? Oh I'm sure that it had nothing to do with mounting variances and testbed irregularities. Worse, the WCP testbed test load is quite a deal higher, so the differences should've been more like 3C, so where did the other 2.5C go that should've been there? What, though, of the RBX, Maze4 and Cascade's performance when operated within their targetted design parameters? Are such even considered, let alone explored, under such a testbed?

I've ranted enough about the WCP testbed. It just saddens me to think that people would actually choose to use it as a reference when it is such a joke.

im no crazed waterblock tester, and even i know using a 1046 to test a block like the cascade is just... silly.

well presumably when I stop mounting waterblocks like a little girl and quit using springs then I can get by with a single mount as well Cathar :)

The rbx was using the #5 nozzle plate. They showed the #1 plate to be about 1C worse. Plus pHaestus had alot of trouble mounting the rbx. I have a feeling that when Joe over at overclockers tests the WW that the difference at 1gpm between the rbx will be alot smaller than what pHaestus showed.

As I understand it, the german site used 100 watts, less an unknown secondary heatloss (say 5%) amount to pHaestus's 75ish watts as measured in the block. So thats 1.5C minus 1C for different nozzle, then add 20% equals .6C off the mark which could be explained by mounting variance.

This isn't to say that I'm validating the site's methods, just to say that the results aren't that impossible.

Bad mounts were thrown out rather than averaged in to skew performance numbers. Which was what made the outcry over my comments kinda funny I thought (OMG HE'S BIASED AGAINST DANGERDEN)

I understand that but I still have a hunch that Joe won't get as big as a difference. I base this on Joe's difference between the swiftech and rbx being the same as BillA's results between the swiftech and WW.

If we assess the relative differences between the MCW5000-A and the White Water between BillA and Phaestus, we see that they do correlate fairly closely.

If we assess the relative differences between JoeC and BillA at 1GPM, we do see a quite a serious anomaly, such as the SlitEdge vs the MCW50002, then there's the PolarFlo.

JoeC, to my understanding, also mounts just once. JoeC has faith in his mounting schema due to some tests he ran in the past obtaining repeatable results, and has dropped mounting blocks multiple times because of that. The right choice or not?

Phaestus's results with the Maze 4 and the RBX correlate very closely with JoeC's.

It would seem that for whatever similarities you can draw between the 3 test-beds, one can also find an anomaly.

Then we have Phaestus's testing of the Cascade yielding a mount that obtained a result more than 1C better than his average. Then we have the RBX variations which were dropped.

Even BillA yielded the odd mount with the White Water in his tests that obtained a result about 0.5C better than average.

What this tells me is that mounting blocks is bloody hard to do, even for those who are really taking care. WCP mounts just once. JoeC mounts just once.

Those who don't mount more than once won't catch variations. Those who do mount more than once typically throw away the extremes (high and low) and take an average.

Myself, I believe in assessing a block's performance on the basis of its best mount out of a large sequence of mounts (at least 12), with anything else less than that best to be considered in error, rather than considering any good mounts as anomalous.

Single mount testing, even for those who know what they're doing, is always subject to +/- 1.0C variations, no matter how firm their conviction that the truth is otherwise.

When designing a block I want to know what the design is capable of. Taking an average of mediocre mounts won't achieve that, because all you're doing is taking an average of your mounting "flaws", rather than measuring what the design can actually do.

IMO Phaestus should've used the results he got from the best mount over the sequence of 10 mounts. That's how I do my testing but it's not for me to impose my beliefs on others. What's to say that JoeC didn't "luck out" on his first mount of the RBX, and achieve an optimal mount, when many other mounts of other blocks were merely "average". My own testing, and other's testing, would indicate that to be the case, but I'm biased. I've also stated my position, so it's for people to take it however they want. I know in my testing I don't want to throw away the possibility of a good design because of mounting variances. Mounting variances can be fixed easily, but losing a good design and chasing up a wrong path is not so easy to "fix".

Cathar:

I believe that JoeC mounts multiple times, as shown here

In principle he should get less variance than Bill or I because of his clever method of mounting.

From a statistical standpoint, throwing out all mounts other than the best one is a bad idea. A medium to large sample and a mean + std dev is much more logical. We have different goals though; you want to see maximum possible performance from a block and I want to see minimum chance I am wrong about recommendations :)

The Polarflo that Joe tested had over 2 times greater pressure that the one BillA tested so we can't expect any comparison. The slitedge results still have me confused. Hopefully pHaestus will test it shortly.


I think you're wrong here. Joe states a variance in each test, usually around .0015C/W but different each test, so he must be doing multiple mounts. I was under the impression that he was doing 4 mounts. If he's doing only one mount, then I would agree with you.


I agree with your best mount philosophy. It's always frustrating when someone will do multiple mounts then only show an average without saying how much variance he had or what his best mount results where.

As far as the results from the german site go, they're accurate enough to make me very curious. From what I can tell from the picture, the frontrunner looks to be just a cross drilled block, with three or four holes drilled from one barb to the other.(refering to the 1A-HV2)

Edit: changed .015c/w to .0015c/w

Okay, I think I was wrong about JoeC's number of mounts.

He still has some somewhat strange data though.

Oh, I totally understand your position and I by and large agree with it for purposes of writing a review for the common purchaser. It's just not ideal when trying to separate out design performance from "typical mount" performance.

We're striving for two different goals. For me, looking at the best results with each block is more interesting to my eye.

This is quite interesting discussion going on here indeed :)
my 5 pennies:
Lokking at pictures and analyzing decriptions of aformentioned deustch testbed system I came to following conclusions (based on my limited knowledge of fluid mechanics and thermodynamics):

1. Heatload simulating element used (it's size and configuration) produces massive thermal energy density per mm^2. Such setup preffers blocks with thicker base plate (Fourier's law, isotropic heat diffusion etc.), which may in turn explain better performance of Murks block (VERY thick base plate)

2. Taking into account the above and low flow (jet impingement design is based on localized increase of coolant's density/pressure which entails increased thermal capacity maintaining the highest possible dT) it's totally expected that Murks performance is better then Cascade's (sudden 4C drop is simply impossible, unless Tom discovered how to dump excesive heat into another dimenssion or used sth like foamed graphite insert combined with laser beam micro channeling of copper ;) )

3. Results are just numbers and like with all statistics one reading it has to know very well how to interpret results or not be suprissed arriving at incorrect conclusions.

To summarize, as Cathar stated above, there are no universal designs and every piece of h'ware shows it's potential only if used within it's design perameters.
My conclusion is that (apart form changing goal post mid game i.e. heatload element swap) for given setup results are correct but they do not represent anywhere near the full potential of jet impingement designs.
Its like saying that Iwaki pumps are the best but peeps seem to forget that for an 'imaginary' system with head loss below 3 meters they are outperformed by great number of much cheaper and generally considered not on par with Iwaki babies:)

Please rephrase to an "imaginary" system with a very low restriction, since the head loss varies with flow rate. It is NOT constant.

The restriction or resistance to flow rate, is a property of a system, however, this also varies with flow rate.

The number of volts applied across an electrical circuit is not a physical property of that circuit is it. It is something we have applied, much like a head loss or pressure drop in a liquid circuit.

8-ball

Hi mate :)
Btw, check Xtreme thread ;)

WC system is like electrical system with all the 'bits' installed so we are not applyiing anything, all we do is provide different power source (diff pumps) but performance of regular electrical components is linear (guessing here since I am not that conversant with electrics/electronics) and with stronger amps or volts resistor is going to produce the same resistance (hence watts ratings) all the time (lame example?).

Anyways, at diff flows restriction posed by a system as a whole is different, absolutely right. All I am saying is that an assembled system has just one flow-to-head curve dependant only on hardware from which it was made off. Curve doesn't change, values on it do :)

Actually cooling performance (in terms of W/m²K) is not affected by the thermal density of the heat-load. The coolant flowing through the block is applying a fixed rate of cooling effect, which is proportional to the rate of thermal convection that's going on inside the block, and since the cooling area is fixed, then the amount that the heat source will warm up by is directly proportional to the heat that it emits.

i.e. a block that cools better at 50W of heat load, will still cool better at 100W or 200W, and in fact will provide increasingly better temperatures.

What a thick base-plate does is determine the amount of thermal spread of the heat by the time that the heat reaches the convectional zones within the block. Stick on a thick base-plate, and all of a sudden you have effectively more surface area for the water convection to work on (presuming that the heat source is initially far smaller than the waterblock).

What we strike here is a balancing act though. The thicker base-plate increases the thermal resistance as well. As the rate of thermal convection is improved (typically through higher flow rates) the thickness of the copper increasingly becomes a barrier to further improvements in cooling performance, as it becomes the predominant source of thermal resistance.

For thin base-plate blocks, these blocks rely on the rate of thermal convection being high enough to overcome the lack of thermal spread. Less convective surface area is available for the coolant to operate on, but the rate of thermal convection is high enough to overcome this drawback. There is the added benefit here that as the thickness of the copper is reduced, this too offers less thermal resistance.

So we have a balancing act. If designing a block for ultra-low flow rates, one would naturally choose a thicker base-plate to offset the reduction in thermal convection by increasing the effective area available to the coolant. If designing a high-flow block, then assuming it's done properly, going down to quite amazingly thin base-plates is exactly what you want.

The amount of heat load of the heat-die has nothing to do with it.

A thin based block will suffer at flow rates below its design balance point in comparison to a thicker based block. A thicker based block will not see any significant gains above its designed flow-rate balance point in comparison to a thinner based block.

Speaking of jet impingement, another important aspect here is the jet power. As flow rate is dropped, so too does the jet power and its ability to impinge on the surface. Higher jet velocities demand a wider cup-jet width ratio as the size of the primary impingement zone increases with velocity. Higher jet velocities also demand that the jet stands off from the base of the cup further, as the increased velocity allows for greater mixing/turbulation of the incoming jet stream without it losing significant power. With lower jet velocities (the result of lower flow rates and pumping pressures) the jets need to be placed closer to the base-plate to offset the loss in impingement "power" that would be experienced if one kept the same parameters as a higher-velocity setup.

i.e. for low flow setups, one would both decrease the jet/cup width ratio, and bring the jets closer to the base of the cups (within 2.5-3.5d).

However, doing this will then impact higher flow performance. One won't see quite the gains that is possible at higher flow rates.

Co-incidentally, the Murks 3.1, from what I can see of it, does just what I'm highlighting as issues for low-flow/pressure setups.

Exactly. Waterblock design is a game of trade-offs. For myself, I focus on the highest possible design performance, accepting that in doing so I am sacrificing low flow performance.

I do still believe though that the results with the Cascade at WCP are at the very least between 1-2C worse than I would have expected, but given that the block had been modified by the user with no guarantees of anything after that event, then I guess anything is possible. The Cascade as it ships is still a very powerful low-flow performer, but I happily accept that had I been focussing on lower-flow performance then it could be made to perform better in that scenario.

Changing the base-plate to 2.5-3mm thickness, dropping the jet/cup ratio to 2d (presently above that), dropping the jet standoff distance to 3d (presently significantly above that), and increasing inter-cup area ratio slightly, would yield a design that at 100W would probably pick up 1-2C or so at 1-2LPM flow rates, but we'd see much flatter performance curve beyond 4LPM at which point the present Cascade design would overtake it.

Horses for courses.

Why would increasing the baseplate thickness help when it seems that the overall jet grid area isn't being increased? Would this design have more jets as well?

Well the jetted area on the Cascade is larger than any core is presently, and this is to cater for large cores that may be covered by an IHS.

As it stands, the heat of say, a Barton die underneath the shipping Cascade really only engages about 35% of the block's jetted area. The other 65% is basically cooling nothing.

By making the base-plate thicker, the heat will spread to a wider area, engaging more of the jetted area in the act of cooling the heat. By making the base-plate thicker, the thermal resistance inherent in the copper's conduction is also increased.

As I was saying, there is a trade-off point for the base-plate thickness on the basis of the rate of convectional cooling being applied.

Wow! That makes the cascade's performance even more impressive when you consider it has a "Whole lotta nothing goin on" :D

If it were economically viable to produce a block that is explicitly tuned for a particular CPU die, and not worry about when people upgrade their CPU's, then it would be possible to do even substantially better than what the Cascade is doing.

That is another trade-off that I admittedly made.

This is another aspect to consider when people start tuning around a test-bed. It's like certain graphics companies that go around tuning their drivers for a specific benchmark, cutting corners and removing absolutely anything at all that is superfluous to the demands of the testbed/benchmark.

There can undoubtedly be a situation of something performing brilliantly under a single very tightly defined scenario and significantly less well for other test cases vs a case of performing almost as well across a wide range of scenarios.

Speaking of which - my machinists just informed me that the Cascade XXX (with customisable cooling zones) has been coded up and they're about to run it on the CNC mill pending my approval, for which I'm about to head out the door and have a look...*clicks heels*

XXX eh? Surprised it doesn't carry the "FX" or "XP" moniker :)

Sounds like an experiment! Take the Cascade and plug some of the outer holes and see what happens. Theoretically if the pump is decent it should cause an increase in velocity in the remaining unplugged jets and may cause a performance increase eh?

It's pure water-block porn. ;)

Really though - approaching 3x the jets/cups in about the same area - hence XXX.

If I got it right - it should be quite a deal better than the Cascade right across the board.

Have already made the cutouts for it a week ago. Haven't had the time to plug it in yet.