Water block Writeup with the big dogs!!!!

[webmedic]

Originally posted by gone_fishin


Glad to see your not giving up. But you should expect such challenges when you aspire to "with the big dogs" as your title
No harm intended.

By the way it would be nice to test one of your blocks when I get the BIG DOG equipment set up.

[bigben2k]

All right. I'm glad to see that this discussion hasn't digressed into agresssive name calling, and the such. WebMedic, I don't know how you stay so cool...

Since we're going to talk about errors, let's take a quick look at the categories of error, shall we?

The first one, the most obvious, is a procedure fault (like not mounting a block properly). The results will be way off, but most importantly, not consistent (i.e. they cannot be reproduced).

Second, there are the type of errors that will produce an OFFSET in the results, like what's induced by the diode reader. That's not so bad, as long as the error is consistent, the results still have a certain validity, and if the procedure is clearly explained, one could reproduce it.

Third, there's the nastiest of error types: the gradual one, or the % offset. A bad TIM joint might offset the results by a percentage. A clogged flow meter might do it too. The point is that the error margin increases (or decreases) through the range of the parameters tested. This is bad, very, very bad.

Just like a lab report, you should state the sources of error in your tests. You should do this by categorizing the error, then stating the amount of the error, then state how you add these errors to come up with the total error margin.

That way, you should be able to produce results like this:
(EXAMPLE)
Block A: CPU temp (full load) @ 2.0 gpm: 37.8 C (+/- 2.1)
Block B: CPU temp (full load) @ 2.0 gpm: 37.4 C (+/- 2.1)

As you can see, with this type of results posted, you cannot declare a winner, because the difference between the blocks falls within the margin of error.

Of course it would be nice if you knew right now what the total error margin is going to be, because that's a deciding factor as to wether or not all of this is worth the effort, or at least gives the results that you expect. In other words, if you know your error margin, and are looking to declare a winner, then you should know by now if your procedure is going to allow you to do that. If your goal is to give out a performance chart, as rough as it may be, then you'd know if it's going to be worth reading.

[webmedic]

Ok then. If anyone finds it before me we would need to know the margin for error in the xp diode and the winbond w83696HF. I guess I may be able to get it straight from epox which would be a plus. I'll be looking for this.

As for the mounting it is the only other area were and I might add the biggest area for posibilities for error to creep in. This is the reason I decided to use the mounting that comes with each block. The consumer needs to know the chance for error with the product also. Knowing that if they mount it wrong can cause 10c swings in temp I think would be benificial. After carefull consideration as to weather or not I can perform an exact test. The readings will be as exact as I can get with the equipment but all other things need to be cosidered also so the consumers can at least decide well the block is cool but it's mounting sucks. Mental note buy new mounting clips for my new block.

[bigben2k]

Just as a side note, for future block designs...

It seems to me that BillA had a good point, about hot Swiftech deliberatly puts a curve on their waterblock, because the baseplate may bend, as it's clamped down (leaving a gap in the middle, over the core).

So clamping from the top, in the middle, should be better, right? The block could then be perfectly flat.

EDIT: correction: the Intel IHS is curved, not the Swiftech.

[pHaestus]

I have talked to Bill a bit about this, and decided to go with his advice: suspend the entire motherboard from the ceiling via twine. My implementation is sort of a "hanging plant" style setup, and it is a good bit lower than the test equipment (actually it's pretty near the floor). This way, the wb inlet and outlet tubing can be pretty close to vertical and minimize the strain that the tubing puts on the block both as you clamp it down and later under use. If you have a Swiftech block, then use those springs for all blocks, as BillA has a load-defection curve here:

http://www.overclockers.com/articles608/

He told me the number of turns for proper force; I want to say 7 full turns past first contact.

If you get the same amount of even pressure and no strain from tubing, you are left with variability of the TIM joint. That's where no cleverness in setup design can help.

[Les]

Originally posted by webmedic
Ok then. If anyone finds it before me we would need to know the margin for error in the xp diode ......................... I

Bulit reports some interesting work on xp Diode calibration.
His results are briefly presented here :
http://www.ocworkbench.com/ocwbcgi/u...&f=37&t=000245

and shown here:

[BillA]

Originally posted by bigben2k
Just as a side note, for future block designs...

It seems to me that BillA had a good point, about hot Swiftech deliberatly puts a curve on their waterblock, because the baseplate may bend, as it's clamped down (leaving a gap in the middle, over the core).

. . .

erroneous "quote"
it is the Intel IHS that is 'dished'

the Swiftech is flat, FLAT, FLAT

webmedic
not too sure you will understand these but no doubt one of the "posting experts" here will expend 1000 words on what is wrong with them
(and I am indebeted to pHaestus for his guidance here)

these portray 10 installs under VERY controlled conditions;
measurements at 70W (~95W per Radiate) under steady-state conditions (after 4 to 5 hours),
with equipment 'beyond the pale' (for most, but NOT all, here)









you have a formidable task
I would suggest you scrap your timetable and start your testing lessons;
you cannot, despite all the words, imagine how difficult this is
-> to do well

[webmedic]

If I'm readaing correctly for every ten tries you got ten different results.

But bill if the results are so minute what does it matter anyway? I don't get it. If you have reduced it to a measurement that for real world performance does not matter then why? I mean the measurements are so minute that nothing matters. Who has the best cooling product doese not matter. The only thing that matters is the ease of mounting for the end user and the cost.

[BillA]

webmedic

this may sound harsh, please hear it positively

you need a technical education, you can indeed learn this stuff yourself (as I do often); but it will take some time and you will make many errors - such is the path to improvement

10 trials, 10 different results
well sure, what on earth did/do you expect ?
nothing is perfect, or the absolute same if your measurement tools have the capability of revealing the variation

this is not bad, this is what is needed to understand what the values actually are

"the results are so minute" ??
are you suggesting that a range of 1.2°C in die temps is "minute" ?
when everything has been controlled out the ying yang ?
(this range will be less at lower power levels)

why do YOU think that, despite the Maze3 (as an example) being out for a number of months, no one is up on their soapbox saying its good, bad, or whatever ?
-> because ANY tester worth their salt immediately comes to understand that they cannot obtain reproducible results indicating the superiority of any one of the "good" wbs over another

notwithstanding that comment, the low (typical ??) flow superiority of the Innovatek Rev.3 is apparent

gone_fishin has told you plainly: you are going to get blasted by your own list if you cannot support your assertions

and I'm telling you plainly that demonstrating ANY results to be "valid" will overextend your technical knowledge

start slowly, you've a very long road ahead
(how do you think that in months you will be able to do that with which I am still struggling with after 2+ years and thousands spent in equipment ?)

[bigben2k]

.AMD publication 24309 has some info on the diode, but there's little mention of error margins

[webmedic]

Actually the equations to figure the offset are in this document starting on page 75.

http://www.amd.com/us-en/assets/cont...docs/25175.pdf

[bigben2k]

It looks like someone should place an order for doc# 25443 "On-Die Thermal Diode Characterization"...

[#Rotor]

are these up to scratch?

http://www.hotektech.com/Tes945_946_950.htm


I'm not sure, but 0.4ºc System Accuracy on the 946?!?!?..... nice that the 946 can upload to a pc though.....

[BillA]

0.4ºC System Accuracy is ok, but nothing special
a $50 Fluke 2190A off eBay will do better (+/-0.22ºC to 90 days from calibration with a type "T" TC)
- but to get it to below 0.1ºC is much more expensive, different gear altogether

"that the 946 can upload to a pc" is certainly useful;
the IEEE 488 business is a PITA, and expensive for the hardware and software

some basic info on measurement uncertainty is here

getting a number is not so difficult, also with lots of zeros
understanding - and demonstrating - the validity of that number can be taxing
(for the non-statistician)

[Les]

Originally posted by unregistered





I would like to know what is going on at ~0.7gpm ?
thoughts ?

Dunno but I get this for the 462u results:



Edit: Now showing correct graph.

[Heavy_Equipment]

One thing I wanted to add in regard to Joe's waterblock roundup.

I was using a Z3 at the time he did his testing, and I have to admit, the numbers I was seeing here at home, and what Joe was reporting, were amazingly close.

I think we're getting to the point, with the current generation of water blocks, (...from established, knowledgable block makers) with regards to design experience, and the material at hand, where it's almost as simple as picking the one that is most visually appealing.

A +/- of 2C in testing is far greater than the degree of separation we could possibly see between any good blocks.

If anybody can get block A, to get impressive numbers, on an AMD running at 2 GHz, then it is a good enough measure for me to consider block A.

Now, if B, C, D, and F also appear to run at the same temp as block A, within the margin of error, while block E is running a few C hotter,(greater than 4C, @ +/- 2C) the testing is a valid way to remove the sub-par from the rest of the pack. Of course, the results would have to be reproducable.

[pHaestus]

The testing is typically done in one of two ways:

1) With a "typical" pump and no control of flow rate

2) With a powerful pump and valves (or adjustable power on the pump) to produce graphs as a function of flow rate

The first method is really nothing more than a test of flow resistance, but it can be useful to others with the same pump, rad, tubing size, and filling method

The second method can really be useful in choosing the best block for a particular application, but not very many people can interpret the results this way. Since the flow rates are all interactive, it takes a fair amount of trial and error even with the data to get a truly optimal system.

Bottom line: You could take one of the "poorer" performing blocks and carefully optimize the overall loop and get better performance than the "best" block that is not properly matched to parts.

This begs the question of where to proceed with watercooling and testing. My thought would initially be that the wb makers should do the work of tuning the overall loop and make suggestions. In reality though this doesn't work due to the time, effort, and equipment needed. Such is life.

[Les]

Oops.
My last post in this thread was showing the wrong graph(Conv.Coeff v Reynolds). Have corrected.

[Pinkster]

i doubt the validity of any waterblock shootouts these days. although it helps us determine poor designs or super outstanding designs (btw they dont exist ), i have noticed that often temps do not tell the compete story for faily closely matched blocks...

for example in one comparison of the maze 3 versus the cyclone 5 there was only between a .2 to .5 difference in temp depending on the cpu overclock, with the cyclone ahead, but the cyclone 5 was a full 20-50mhz faster stable than the maze 3. he also mounted and tested them several times and got consistant results. he was also using one of those flat external probes that touch the outside of the core.

I think if minute temperature differences such as this can have such a big effect, then perhaps to get an accurate result as to which blocks perform better you would need rediculously accurate equiptment and a lot of skill to mount blocks perfectly every time.

i think that perhaps the most accurate way to tell which blocks are better is to test there maximum overclocks at different flow rates and heat loads. sure u dont get a c/w vakue, but... who cares if its able to determine which blocks are better? and how accurate are c/w values anyway

[Balerion]

I just wanted to ask if anyone knew of any other waterblock roundups, with current waterblocks like these, that will even come close to the accuracy (low or high as it might be) that webmedics will. Some testing is better than no testing, and if they are all within the margin of error, then fine, there is no clear winner. At least well get to compare flow curves, and at least we'll see a test in a system that resembles a users system.

When I was first trying to pick out a block not so long ago, I looked at alot of reviews, roundups, etc. The only ones that were accurate weren't testing newer blocks that we KNOW perform better, and the only new reviews were always single reviews, and all had varying test methods. At least if newbies like me can see wedmedics review, we'll have SOMETHING to go on.

At least webmedic is trying. I applaud his effort.

Everyone bitches about [H] and toms not knowing shit about WCing, and posting bad reviews where the testing isnt good at all, but this is the first time since then that someone is DOING something about it. He's testing a wide variety of blocks, in a system like most peoples, and hes doing a pretty good job ensuring its accurate and fair. Thank god for Webmedic, Patron Saint of Waterblock Roundups. Think thats a lofty title? Then why doesnt someone else come and usurp him. I challenge others to make a roundup better than his. Now, I'll run away before I get shot by the newbie-haters.

[#Rotor]

Birrman54 has one , and its very close to getting published, but these shootouts are not easy to come by, especially the ones including more than just one or two manufacturers.

[Joe]

Originally posted by Balerion

Everyone bitches about [H] and toms not knowing shit about WCing, and posting bad reviews where the testing isnt good at all, but this is the first time since then that someone is DOING something about it. He's testing a wide variety of blocks, in a system like most peoples, and hes doing a pretty good job ensuring its accurate and fair. Thank god for Webmedic, Patron Saint of Waterblock Roundups. Think thats a lofty title? Then why doesnt someone else come and usurp him. I challenge others to make a roundup better than his. Now, I'll run away before I get shot by the newbie-haters.

The reason is simple. Testing waterblocks correctly is hard... DAMN hard, and takes a ton of time. I still have nightmares to the WBRU1 I did. I DO want to do another one. but I am waiting for the money and the time to get the proper gear. Thats the reason it took like 6 months to do the last roundup, I wont do a test unless I have the gear that will make me feel comfortable enough that I can get accurate and repeatable results.

I have a few other things that I want to do before I get sucked into another testing black hole, but the plan is to do a series of smaller but more frequent WBRU's in the future.

Also the scale, and the fact someone is doing a roundup doesnt mean jack. Wait for the results before you start to praise a review... Doing it the other way is like thanking the chef for the great food before you get served... and then to find out he just pooped on the plate.

[BillA]

Balerion
get a grip, you're blowin' smoke

have you read this thread ?
or are you just building your post count ?

scroll up (back) this page, several more pages
see any graphs ?
look around at some other threads
need a list ?

[pHaestus]

Some people with a conscientious streak out there doing testing with typical overclocker gear. The key is to have a clear goal and to keep in mind that there are many interrelated parts that are involved in "whole system testing" that people like to see:

http://www.overclockers.com/articles546/ Note the emphasis on mounting and subjective properties, and note that Hoot controls ambient air temp AND water temp by adjusting the voltage of his fan. Hoot is trying to test a block in a system not test the overall system.

http://forums.overclockers.com.au/sh...threadid=50674 Some good insight and advice on pitfalls to avoid in "in situ" block testing here.

http://www.coolhardware.co.uk/module...owcontent&id=9 Testing on flow rate only. Another example of the kind of stuff an o/cer can put together of use.

In all cases note the lack of mention of replicates or the possibility in errors in measurement.