Water block Writeup with the big dogs!!!!

[bigben2k]

Gotta chime in...

Gone_fishin: I understand your point of view, but I think that you're forgetting that WebMedic's testbed, as flawed as it may be, will still show the difference in temps between blocks. Any test bed will have an offset, but all testbeds should show the same temp difference, from block to block.

Overall, I see a lot of dancing around the issue about finding out which block is best, when the reality is that a lot of blocks can be made to perform equally, with different flow rates. I think that BillA has proven that already.

So if this is going to be a block roundup, then what it really is is a close-up examination of its performance curve, at different flow rates.

As for WebMedic's testbed reliability, I have a few concerns myself. I'd like to see a detailed writeup of the procedure, a writup so detailed that even a child could follow it. IMHO, labs operate the same way. I would also introduce a random element, just to attempt to validate the results: don't run 5 consecutive tests on a block, then move on to the next block: run 1 test, then change block.

The flow rate is going to be important, because each block has a performance to flow rate curve that is unique. That's the tricky part.

As for real world results, the different lenghts of tubing, rad and res selection will greatly affect the flow rate, so you can't even really test a rig with standard pumps: all you can do is report the blocks curve.

[bigben2k]

Originally posted by unregistered
VERY interesting pHaestus

question:
are ALL diode readers going to produce the same results ?

if not,
how should the diode reader be described (speced) to indicate/assure its accuracy ?

In other words, what's the accuracy of the internal diode, and more specifically, what deviation can one expect for the same CPUs? Does AMD spec that? Does the batch number matter?

Here's another silly question: Is it just me, or does anyone else believe that the two AMD cores will produce different results? Can anyone lend WebMedic a 2200+? (If he's willing to test with it)

[BillA]

goddamn, someone loan ME a 2200 !

[pHaestus]

Are all diode readers going to produce the same results?

I doubt it. As for advice on specing them, check maxim's technical document.

Keep in mind though that Andrew (Millennium Thermal) said his die simulator (same design as JoeC's) typically is about 15% higher in C/W than JoeC's. I would think that the diode readers should all be that close

I should also note that some heatsinks never performed as well for me as JoeC predicted, and others seemed to do a lot better. I found the Alpha PAL8045 to be substantially better than the Swiftech MCX462 with a variety of fans, and I found two of the Dynatron heatsinks equipped with the 38mm thick Delta fan (45CFM) to both do a good bit better than Joe saw with his simulator. For the Dynatrons, I really think I was seeing some secondary cooling due to the enormous amount of air around the socket. For the Swiftech vs. Alpha, I am chalking it up to Joe's simulator not really being designed for 4hole mounting (something about wedging a piece of wood in place that just doesn't sound good to me).

At any rate, I feel comfortable pointing people to Joe's huge table of heatsinks, as my testing and his seemed to jibe pretty well in comparative terms.

[pHaestus]

I don't think that 2 identical speed CPUs will report differences in termperatures that are substantial enough to see with the diode reader's resolution in motherboard monitor. I wasn't able to compare thermal pastes successfully with that setup and I think the spread in temps with different CPUs would be considerably smaller

[BillA]

"check maxim's technical document"
are all diode 'readers' going to be based on that chip ?
no difference due to different bios ?
(that not what I'm hearing at all)

and a means to cross-correlate ?

[pHaestus]

Oh. motherboards

All crap; I trust none of them. Not a single manufacturer is following the specs exactly.

I thought we were talking about the (somewhat common) MAX6657 hack like my older stuff or (better yet) the evaluation systems like I now have.

[bigben2k]

Originally posted by pHaestus
Oh. motherboards

All crap; I trust none of them. Not a single manufacturer is following the specs exactly.

I thought we were talking about the (somewhat common) MAX6657 hack like my older stuff or (better yet) the evaluation systems like I now have.

Ok, it's time for me to introduce a new concept (except to BillA, of course!)

Cumulative error margin.

If the diode has a +/- 0.1 error margin, and the chip that reads it has a +/- 0.5 error margin (assuming specs were followed), then we have a +/- 0.6 margin of error.

If we test perform a block, there are many, many variables, not all of which can be measured. Right now, it's too late for me to try to list them, so I'll let a younger whipper snapper give it a shot.

[pHaestus]

Scan up a bit; some comments on uncertainty from me. And I think that the uncertainty is better described by the square root of the sum of squares. Assume +/- 1C for both the diode and the reader, or sqrt 2 for the uncertainty (1.4C) . Two temps need to be about 3C apart to have any confidence that they actually different, eh?

Or with compunurses, the water temp and "CPU" temp would be +/-3C. So sqrt 18 = 4.2C

This is a simplification of the error propogation of course and as #rotor mentioned you can do some stats to make use of large sample populations and lower this somewhat. By and large, though you are doomed if you are looking for subtle temp differences with such gear.

[bigben2k]

Originally posted by pHaestus
Scan up a bit; some comments on uncertainty from me. And I think that the uncertainty is better described by the square root of the sum of squares. Assume +/- 1C for both the diode and the reader, or sqrt 2 for the uncertainty (1.4C) . Two temps need to be about 3C apart to have any confidence that they actually different, eh?

Or with compunurses, the water temp and "CPU" temp would be +/-3C. So sqrt 18 = 4.2C

This is a simplification of the error propogation of course and as #rotor mentioned you can do some stats to make use of large sample populations and lower this somewhat. By and large, though you are doomed if you are looking for subtle temp differences with such gear.

Agreed. I'm glad to see you understand this well, eh?

[pHaestus]

somewhat.

[gone_fishin]

Originally posted by bigben2k
Gotta chime in...

Gone_fishin: I understand your point of view, but I think that you're forgetting that WebMedic's testbed, as flawed as it may be, will still show the difference in temps between blocks.

Sorry but there is just no way in hell he could show the difference in individual blocks when the margin of error in his equiptment is higher than the difference in block performance. Let's stop this pretending with this proposed shoot out and focus on what it can really do, highlight the available blocks. The best that can be done (I'm having deja vu) is come up with two groups, one which SEEMS to perform well and one which SEEMS to perform less. Of course the middle of the road between these two groups could not even be well defined.

[decodeddiesel]

Originally posted by gone_fishin


Sorry but there is just no way in hell he could show the difference in individual blocks when the margin of error in his equiptment is higher than the difference in block performance. Let's stop this pretending with this proposed shoot out and focus on what it can really do, highlight the available blocks. The best that can be done (I'm having deja vu) is come up with two groups, one which SEEMS to perform well and one which SEEMS to perform less. Of course the middle of the road between these two groups could not even be well defined.

Damn It!! I wanted to stay out of this but I am confused here. G_F if in fact he could not show the difference between blocks with his proposed test set-up (whether you mean due to ACCURACY OR PRECISION I do not know) then how could he possibly seperate the tested blocks into "GO" or "NO-GO" catagories? Are you suggesting he base this purely on on asthetics? Without any testing it is simply a matter of "this one looks shitty...this one looks good" from his oppinion. I guess I am just confused as to what you are suggesting he do here??

[gone_fishin]

Originally posted by decodeddiesel


Damn It!! I wanted to stay out of this but I am confused here. G_F if in fact he could not show the difference between blocks with his proposed test set-up (whether you mean due to ACCURACY OR PRECISION I do not know) then how could he possibly seperate the tested blocks into "GO" or "NO-GO" catagories? Are you suggesting he base this purely on on asthetics? Without any testing it is simply a matter of "this one looks shitty...this one looks good" from his oppinion. I guess I am just confused as to what you are suggesting he do here??

Reread my posts. I propose he could not tell the difference to come up with an order of block performance..A, B, C. What I said was he could only validate this by having a goal which matches his testing abillities eg. These blocks over here SEEM to perform good and these blocks over here do not SEEM to perform good.
The conclusions would then be a more honest representation of the testing abillities which are limited by the equiptment he possesses. And I mean that BOTH accuracy and precision are lacking here.
Good waterblock reviews are few and far between. A good proper approach at this introductory review by webmedic will leave the door open for more technical and accurate reviews by him in the future if he follows through with his desires to obtain better testing equiptment.

[bigben2k]

Originally posted by gone_fishin


Reread my posts. I propose he could not tell the difference to come up with an order of block performance..A, B, C. What I said was he could only validate this by having a goal which matches his testing abillities eg. These blocks over here SEEM to perform good and these blocks over here do not SEEM to perform good.
The conclusions would then be a more honest representation of the testing abillities which are limited by the equiptment he possesses. And I mean that BOTH accuracy and precision are lacking here.
Good waterblock reviews are few and far between. A good proper approach at this introductory review by webmedic will leave the door open for more technical and accurate reviews by him in the future if he follows through with his desires to obtain better testing equiptment.

All right, let's put aside the question of wether or not you would submit your block for WebMedic's test.

What I'm hearing from you, is that you expect WebMedic to run a short series of test, and declare a winner. So on to WebMedic: Is that what you intend to do?

I've stated it before, the best that you can do, is give out a performance curve for a block at different flow rates. Since you're only testing 3 flow rates (3 pumps), then I would tend to agree with Gone_Fishin, simply because you didn't make the effort to graph the WB's performance curve. 3 points on a graph won't do anything. 5 points might give us a rough idea, but it is rough.

If you were willing to graph the performance for at least ten (10) points, (i.e. 10 different flow rates), then you would have a serious testbed.

In any case, I happen to know that you are leaning towards a quick user type setup test. As long as you remember that what you're going to test will NOT declare a block a winner over any other, and that you specifically explain why, and that you make no point in stating that one block is best, then you've got yourself a decent test report.

Here's why: your tests are too specific. It's like trying out 4 different tires on your car, and a few months later report which one performed best, and wore out the least. For any car buff, ya'll know what I'm talking about:
1-you never use different tires on the same car.
2-The different tires will cause a car to pull towards one side or the other, wearing out another tire.
3-Each tire is designed differently, for a specific wear time
4-Each tire has a slightly different diameter.

It's the same thing with waterblocks. Each waterblock is designed differently. You can't use the same waterblock on the same computer, because you're only testing it for one set of variables. (i.e. a Maze 3 might perform better with an Eheim 1048, but a Swifty would beat it with an Eheim 1250).

In short, WebMedic, I'll take in your results, but I'll keep in mind that each block was tested for only a few circumstances. I can interpret the data (where most can't) to figure out what I need to know.

[decodeddiesel]

I did read your posts in the first place, thank you. However this statement is what prompted me to ask you the question I did.

Originally posted by gone_fishin
Sorry but there is just no way in hell he could show the difference in individual blocks when the margin of error in his equiptment is higher than the difference in block performance.

So once again if his testing methods are so grossly inaccurate/precise as he would not to be able to show the difference between the tested blocks then how would he be able to seperate them based into 2 different performance catagories?

[decodeddiesel]

So here's a bomb of a question, (and I am not trying to put Joe on the cutting boards here, rather I would simply like to use his test bed as a reference), what does everyone think of Joe's WB testbed?

[pHaestus]

I think Joe's testbed is a good example of what one can do at their home with a reasonable amount of money. The thermocouple under the CPU core is better than most other CPU temp measurements (compunurse on side, in socket, etc), and it at least has close to the needed resolution. Joe uses a digital flowmeter, though I am not sure of the tolerances of the swissflow. Joe uses a big pump and varies GPM to generate a range of performance typical of what users could see. Joe also (most importantly) is VERY familiar with his setup and so can spot anomolies that may affect the results. Joe is also very honest about what his testbed's capabilities and limitations are.

However, thermocouples aren't gonna be useful for the kind of temperature differences that we are interested in (I think they are +/-0.2C or so) and the under core temp is an indirect measurement that is related to the CPU temp, but not perfectly so. And proper resolution on the flow rate is the big thing that I can't really comment on (dont know much about the swissflow).

[webmedic]

Originally posted by unregistered
"check maxim's technical document"
are all diode 'readers' going to be based on that chip ?
no difference due to different bios ?
(that not what I'm hearing at all)

and a means to cross-correlate ?

I don't know about this but epox uses winbond for there reader.

[webmedic]

Originally posted by gone_fishin


Reread my posts. I propose he could not tell the difference to come up with an order of block performance..A, B, C. What I said was he could only validate this by having a goal which matches his testing abillities eg. These blocks over here SEEM to perform good and these blocks over here do not SEEM to perform good.
The conclusions would then be a more honest representation of the testing abillities which are limited by the equiptment he possesses. And I mean that BOTH accuracy and precision are lacking here.
Good waterblock reviews are few and far between. A good proper approach at this introductory review by webmedic will leave the door open for more technical and accurate reviews by him in the future if he follows through with his desires to obtain better testing equiptment.

Ok agreed here. So what do you propose I should focus on for ths round of testing?

[webmedic]

Originally posted by bigben2k

All right, let's put aside the question of wether or not you would submit your block for WebMedic's test.

What I'm hearing from you, is that you expect WebMedic to run a short series of test, and declare a winner. So on to WebMedic: Is that what you intend to do?

No It's not as simple as that. On a side not I will rotate the blocks and not just test a single block 5 times and then go on to the nest block. I alrade have a large enough sample to start on this. I want to get the testing procedure down fist which is the reason or at least my reason for being here to take suggestions and work out the testing procedure before I start anything. There are a great many block that will not be here for a few weeks anyway.

Also I will work on clarifying the testing methods page to be more precise. I think stated above I will also use the block as it comes mounting and all. in this way it will be more a test of the block as the manufacturer send it. This in theory should also, how to put this, take into account the margin of error induced by a bad mounting mechanism. That is to say results will be not as good due to the mounting but it will better represent what the actual manufacture is selling and what to expect from such a product. I should think that the 5 test runns will also show a larger margin of error if the mounting system is not good.

I've stated it before, the best that you can do, is give out a performance curve for a block at different flow rates. Since you're only testing 3 flow rates (3 pumps), then I would tend to agree with Gone_Fishin, simply because you didn't make the effort to graph the WB's performance curve. 3 points on a graph won't do anything. 5 points might give us a rough idea, but it is rough.

I do have a few more pumps I can through in. Also to note we will be testing the flow rate of each block with a specific pump but I will not be able to change the flow of the pump to force each block to flow at the exact same flow rates. That is further down the road when a better pump has been collected and a magnetic flow meter.

If you were willing to graph the performance for at least ten (10) points, (i.e. 10 different flow rates), then you would have a serious testbed.

Again I do not have that capacity now but I could get 5 or 6. See the answer above about better equipment

In any case, I happen to know that you are leaning towards a quick user type setup test. As long as you remember that what you're going to test will NOT declare a block a winner over any other, and that you specifically explain why, and that you make no point in stating that one block is best, then you've got yourself a decent test report.

Of course as stated before it will aslo talk about cost, ease of use and setup, mounting, I'm popen to suggestions here.

Here's why: your tests are too specific. It's like trying out 4 different tires on your car, and a few months later report which one performed best, and wore out the least. For any car buff, ya'll know what I'm talking about:
1-you never use different tires on the same car.
2-The different tires will cause a car to pull towards one side or the other, wearing out another tire.
3-Each tire is designed differently, for a specific wear time
4-Each tire has a slightly different diameter.

It's the same thing with waterblocks. Each waterblock is designed differently. You can't use the same waterblock on the same computer, because you're only testing it for one set of variables. (i.e. a Maze 3 might perform better with an Eheim 1048, but a Swifty would beat it with an Eheim 1250).

In short, WebMedic, I'll take in your results, but I'll keep in mind that each block was tested for only a few circumstances. I can interpret the data (where most can't) to figure out what I need to know.

I'll make sure to state what the results are and indeed I spent the whole first page of the testing methods expaining this verry thing.

I will not be using mbm to monitor temps. I'll be using cpucool. It will read to .1 and log to a file up to once every second. This is much better than mbm for temp monitoring. Of course taking into account the margin for error but at leat this is somewhat more precise.

Sorry for spelling if it is off I'm trying to write to fast.

[gone_fishin]

Originally posted by decodeddiesel
I did read your posts in the first place, thank you. However this statement is what prompted me to ask you the question I did.



So once again if his testing methods are so grossly inaccurate/precise as he would not to be able to show the difference between the tested blocks then how would he be able to seperate them based into 2 different performance catagories?

That is why the word "SEEM" was in big block letters in both of my posts

[webmedic]

One last word on this I'm not deluded I don't think this will the block shootout to end all block shootouts. I do understand the limitations of the hardware being used and will make sure the reader knows the same. Unlike other waterblock shootouts. Hopefully this will not just be another one of those. But it will not be billa level YET either.

[decodeddiesel]

Originally posted by pHaestus
I think Joe's testbed is a good example of what one can do at their home with a reasonable amount of money. The thermocouple under the CPU core is better than most other CPU temp measurements (compunurse on side, in socket, etc), and it at least has close to the needed resolution. Joe uses a digital flowmeter, though I am not sure of the tolerances of the swissflow. Joe uses a big pump and varies GPM to generate a range of performance typical of what users could see. Joe also (most importantly) is VERY familiar with his setup and so can spot anomolies that may affect the results. Joe is also very honest about what his testbed's capabilities and limitations are.

Thats about my take as well. Yes the limitations of actually measuring the cpu for temp exist, although the undercore probe is a pretty good method of core temperature measurment from what I've seen. But what has been said is correct, given the flaws, inconsistancies, and lack of precision (due to inacceptable resolution) with onboards ondie readers, and your (pH's) diode reader it would be impossible to find "actual" core temp. and therefor to achieve a "control" for CPU temp in the experiment.

However, thermocouples aren't gonna be useful for the kind of temperature differences that we are interested in (I think they are +/-0.2C or so) and the under core temp is an indirect measurement that is related to the CPU temp, but not perfectly so. And proper resolution on the flow rate is the big thing that I can't really comment on (dont know much about the swissflow).

Nor can I comment on the swissflow, I for one would love to purchase one but it's really outside my budget at this time.

[gone_fishin]

Originally posted by webmedic
One last word on this I'm not deluded I don't think this will the block shootout to end all block shootouts. I do understand the limitations of the hardware being used.

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.