Some test results (and site news) from the road
 

Pretty busy at Argonne Lab now but I figured I'd take an evening break and finally get around to posting Cascade SS performance (attached). I verified with Cathar that he had done some tweaks with flow resistance on the SS; you can clearly see the maximum flow rate is higher. Overall a very nice block though the price premium puts it a little out of my league. Performance is actually a little better than I expected; I am guessing if I were to retest the Cascade with my current testing protocols I might see slightly better performance. In any case I figured a few of you Cascade SS owners would be interested in the results.

I have several packages at my house that bode well for the test setup. I am hoping that July and August give me time to get it running. First item will be to configure and run pressure drop tests on all wbs tested thus far.

Also DTek Customs will be a sponsor of Procooling as soon as I get back to Saskatoon and put up their ad and Pro/Deals space. I am hoping that Danny has a special deal for you guys on those new pumps we've been talking about :)

Just curious...
 

Hey Phaestus, I know its totally off-subject, but I was wondering what kind of work your doing at Argonne? I interned for the DOE via UC for a summer at LLNL, so I remember a few folks from Argonne.

I am here using the Advanced Photon Source to do x-ray absorption spectroscopy. I use the APS a few times a year for research like this and I also work at the SRC in Madison 2-3 times a year. It keeps me young to stay up all night watching experiments

Well I get that'll make Liquid3D happy now. No longer will the Cascade almost solely dominate the lower section of the comparison chart. :D

Thanks again Phaestus for all the hard testing work.

eh they aren't for sale though. Thinking I should have 3 graphs for wb results: commercial blocks, DIY blocks, and wbs of legend...

I await someone to crunch numbers on theoretical performance difference between silver and copper and point out that the design must also be somewhat different to explain results. Any estimate on how much performance difference is due to a few little tweaks and how much is due to silver?

Argonne - dim memories of some good "dislocation dynamics" work in the 1960's

"current testing protocols "
Have I missed something, and will a reread of article enlighten?

"I await someone to crunch numbers on theoretical performance difference between silver and copper"
Think only Cathar has a specialised program connecting h(conv coeff) and h(effective) for this type of wb.

Argonne Labs
Think an equation by C M Li , which I failed to derive in 10years trying, was presented at an Argonne conference.
The equation was a relation between change in "dislocation velocity" with change in "short range stress".
Never used it .Had to use a more unwieldy beermat equation.

Nice...
 

Two years ago I got a stint for a summer working on hydrodynamics in radiation dominated environs for potential NIF targets at LLNL...mucho X-band work. Actually used some APS sourced data involving high-z elements in some of our computations. I think the Z-pinch type approach is the future for ultra-intense X-ray outputs...many cantakerous debates with the NIF lovers over that! It was a lot of fun though, met some way crazy smart peeps, also got some exposure to the "Dark Side" of the nuke business...scary stuff. What are you experimenting on that needs X-band ev? It sounds interesting.

On a watercooling note, I am highly anticipating more data/discussion of the SS Cascade. Also, what of Asetek Anarctica? Its well made, posts good stats at overclockers.com, and bucks the trend of European blocks and 8mm tubing being inseperable. Have you ever come across one Phaestus?

For me, given a ~100W CPU heat load, I sorta-kinda honed in on a 0.7C improvement with the SS over the standard copper based block.

Of that:

Tied to cup porting and jet tweakage:

~0.1C as a result of cup porting giving lower pressure drop giving slightly higher flow rates
~0.2C as a result of the jet tweakage

Both of the above could be applied to the copper block by anyone who cared to, but it's very fiddly work which I was doing by hand. Somewhat at a cross-over point of if it were better to do it via the NC, or by hand.

Tied to the metal and base-plate tweaks:

~0.2C tied directly to the use of silver's higher thermal conductance
~0.2C tied to base-plate thickness tweakage to match the silver's higher thermal conductance

So basically ~0.4C due to the silver, but only by modifying the design to take advantage of the silver, as opposed to the ~0.2C that would've been gained by just using silver without changing the design.

Les, I have no real strong working models for what goes on in a Cascade. It is still somewhat of a mystery, mostly because at times when apparant theory said that things should be getting worse, they were actually getting better. I have enough of an understanding through emperical data on how to tweak it to best effect, and a rudimentary set of hypotheses of why, when tweaking one aspect, a certain result comes about.

There isn't really enough room in a Cascade cup in relation to the jet size for classic jet impingement activity to form. In fact, according to "classic" jet impingement theory, the block shouldn't perform as well as it does. I guess that's the difference between on-the-ground empericality, and theory. I don't pretend, on this one, to have enough of a clue to be able to derive theory for it. Am more interested in just making it work well through observation and experimenting.

[Edit: Said lower when I meant higher]

Yes, intelligent "sucking and seeing" is the the key to r&d.
However unraveling the "mystery" can be fun.

I did, and still do, have prototypes here whereby I implemented a more "classical" jet impingement block, meaning cups that were large enough in relation to both the jet size, and jet offset from the base. Still cup based, but the cups were much larger to a point where the equations worked out that the jet impingement activity should be forming properly, and "h" was maximised.

In order to make it all fit, there weren't that many jets/cups (~20) and over a slightly larger area than the Cascade covers. Jet velocity was much higher than for a Cascade, and the block was a lot more restrictive. Increased bp thickness (to 1.8mm) to "compensate" for variations in h that occur with the radial distance from the jet center.

End result was something that only basically matched the White Water, even at lower flow rates where the thicker bp should've helped it out.

I tried a number of variations on the "classical" impingement theme, and was basically unable to match a Cascade, and certainly not with anything like the low (in comparison) pressure drop that the Cascade/SS has.

As stated here, tend to think of in terms "vectorial change in coolant velocity" with impingement being a special case, for which Flomerics allows some fumble(do like the term Cathar) numbers to be used

Here we have the h(conv.coeff) relation with h(effective) problem .There are several ColourFul Displays(Deeppow's terminology?) in the forums. However non appear to cover a range of h{conv coeff) and the h{conv coeff) used are unspecified or(in my view) unrealistic.
With the exception of channels, where I use Kryotherm, my beermat approach is totally inadequate for this problem
May be Groth can have an impact here.

Will keep unraveling and then re tangling.
Tis Friday and nearly pub time.

Edit: Added link and colour.

Yes, I couldn't agree more. It's something that I constantly find myself beating my head against. Whilst a very high conv. ceoff. may be very attractive, sometimes it doesn't produce the highest effective h, and I believe that the Cascade in comparison to a "classic" impingement block is a very good example of this.

The Cascade offers more "useful" surface area, and despite, to my best estimates, offering a convectional h of around 80% of that of a "classic" impingement effect, achieves more effective cooling.

Something I struggled long and hard with was the inter-cup geometry. Decreasing inter-cup spacing puts more cups into the same area, but now there is reduced vertical (and to some degree lateral) heat spread, and performance goes down. Put too much inter-cup spacing, and you manage to engage more of the cup walls as part of the cooling effect, but performance goes down because the spreading resistance goes up.

I tried hard to computationally model this effect based on emperical evidence, but was unable to achieve anything that I had any confidence in. Basically I found myself falling back to established measured good ratios. I found myself in that mode that Bill disdainfully describes as "measuring the data, and then assembling the theory to fit".

In the end I gave up, and just satisfied myself with approximated guesses as to h

Anyone up to a database-driven grapher so we can select the blocks we wish to see on the chart?

I tried to get Joe to set this up but I guess there was some effort required. I also heard it HAMMERS server to generate the graphs dynamically. You have to set it up to cache all possible graphs every day I guess.

Les about the testing methods comment:

I must have somehow managed to lose a few sentences somewhere along the line of revising this article. Here's what is in the article:

"This procedure has been used throughout all testing, but my thoughts on how to best characterize variability and reproducibility have somewhat changed. My initial philosophy was to simply repeat the above tests 10 times and record an average and standard deviation. In principle, this works well enough. I also found it sensible to throw away anomalously bad results (1-2C higher than normal) as being due to user error in mounting. What I found in some cases, though, was that a single test run would give anomalously good results (that PERFECT mount is a bit like the fish that got away). "

Here's what's missing:
"To account for this, I have somewhat modified my testing approach to make the reported results closer to the best performance that I can reproducibly achieve with a given waterblock. As I repeat mounting replicates, I continue to test and mount until I have five successive mountings that are within 0.5C of my BEST performance. This is why newer test results have standard deviations that are smaller than the first waterblock tests. It is also why I rarely report 10 mountings in my test results for newer waterblocks.."

I will add that as soon as I get back home. It doesn't dramatically shift the performance numbers but it does tend to push blocks I have retested down a little closer to the bottom of their error bars.

Sorry for the confusion.

Do the graphs really need to be dynamically generated? I know you were thinking something along the lines of the late Arachnid Stats, but your data isn't changing hourly (I hope).

How about a stack of transparent gifs? The PHP to make a page with an arbitrary stack shouldn't be too tough for a coding-stud like Joe. Hell, it could probably be done client-side with java.

I dont think it takes that long to generate all the graph possibilities groth it's just something to be aware of if you're implementing the commonly available code

Use gnuplot. Can generate personalised graphs extremely quickly on the fly. Have a collection of data points, one per block. People click on a set of radio buttons to select which blocks, that gets POST'ed to a cgi, out comes a personalised on-the-fly gnuplot generated picture. Takes <0.1s of CPU time on a slow CPU to generate most graphs. Have the graphs generate in .png format, which is smaller/better than .gif, and you're done. Unsure of how much load there is on your web-server, but I'm pretty sure it's not going to be more than 10 reqs/sec for such operations.

I do like the idea of caching all possible graphs. All (2^n)-1 of them. Within the year, each review you did would require Joe to buy terabytes of more storage. :)

hehe currently just paying for hosting for the site is a primary concern at this point, terabytes of data aside :)

oooh, i'm liking the possibility of deals with D-tek. (yeah, might not happen) but any idea of a timetable on that? i've got a b-day coming up soon that would be a great excuse to get one of those pumps, especially if on sale.

Put up a Pay Pal donation link, as long as you don't pester people they will most likely give a little now and then.

I'll put up their ad and Pro/Deals by July 1. When and what they add to the Pro/Deals is up to them of course

cool, thanks for the response.