Some Interesting/Confusing Data from testing

[freeloadingbum]

The more I think about it, the more I'm convinced that temp gradient in the hose is responsible for the large difference in wattage between the cascade and maze4 at the same cpu temp. The flow through the outer 2 channels on each side of the maze are problably picking up little or no heat while the center channel may be picking up 50% or more. The cascade with all the flow starting at the core combined with the high turbulence should definitely see a better mixing than the maze.

If the above is true then changing the depth of the sensor ( if possible) would yield a large change in the maze wattage (50 watts perhaps), while yielding a small change in the cascade (0 to 10watts)

If this also proves true, then blocking off the flow to the maze's outer channels (filling in with silicone for instance) should give a nice performance boost by increasing the velocity of the inner channels for the same flow rate. I'm guessing as high as 1.5C at 1 gpm.

Edit: spelling

[pHaestus]

GIGO is an old programmer acronym. Short for Garbage In, Garbage Out. Those CPU Watt calculators aren't grounded in reality. Sure I can calculate a C/W, but it'll be crap.

[Les]

pHaestus
Were the tests performed with the TBredB 1700+ at 2200MHz and 1.85V ?
Don't think MHz and V were specified in article.
Info would be useful for cross-ref to Incoherent's thread.

BTW Benchtest com Calculator indicates 113Watts (100% usage) for a TBred 1700+ at 2200MHz and 1.85V.

[Aardil]

Beginning to wonder if splitting atoms wouldnt be easier :shrug:
A great thread just wish I understood what all them pretty lines ment.

[Myth]

#42 Ahh!

I know the eventhough we have spent a lot of time geting the calculations i the program as correct as can be.. it will always be more of a theretical result than a real life one, and it we do get similar results compared to the BTW benchtest calculator. So we prob. use the same formulas. The question realy is how far the result calculated this way is from the real life result which he is trying to calculate using the formula..

But i get your point, and think i will wait and see if we get some workable formulaes, and perhaps they could be added to the program..

[Incoherent]

An observation pHaestus

If you are using anti-freeze in your system, the value for heat capacity should be less than 4186J/kg*°C. Unfortunately I don't know how much but I would guess that it would be in the region of 10% less with the 1:4 mix you are using. Not particularly significant when comparing waterblocks but something to remember if we want to compare data from different power measurement techniques. It does bring your W results down to a perhaps more realistic level but doesn't explain the slope.
Again, I suspect an offset. A bit of qualification for this statement in these two charts.

This a bit colourful maybe but it is showing the calculated delta T curves vs flowrate for various power levels. Also shown are the Ts from your data for all the blocks. As you can see the delta T's adjusted by an offset of -0.07 deg fit very closely to a 70W heat source.

The second chart shows the new power calculation vs flow rate. This is a bit ideal but might point in the direction of a (small) probe error.

Another thing, I think that pressure loss needs to be measured for these blocks at different flowrates. I think that presented as it is the Cascade seems to be massively ahead of the competion (a conclusion I believe is not incorrect) but for a given pump/system the flow will be less through the Cascade, and hence the performance delta will not be as much as you'd conclude from simply looking at flow rate. I think this has been discussed elsewhere.
Not to take ANYTHING away from your review/testing methods mind you, or from the Cascade. Both are fantastic.

Cheers

Incoherent

[Les]

Incoherent
You have been busy.
Good thinking. Never thought of introducing a known error.
I use Scott Gamble's data for anti-freeze correction.
Yep 10% seems about right.
Graphs look eminently sensible. Will check your Excel manipulations for stupidity - will only post if find.

[Incoherent]

Originally Posted by Les

Will check your Excel manipulations for stupidity - will only post if find.

Be my guest Les. Excel sheet is here

Cheers

Incoherent

[Les]

Using pH's but will cross-check.
Thanks.

[pHaestus]

Incoherent you are 100% right about needing pressure drop vs. flow rate data. My digital pressure gauge was acting strangely when I was testing that MCWChill; I can see if maybe it was the low temperatures that were messing it up and if so generate. Otherwise I bid on ebay periodically but haven't yet gotten lucky.

I took the day off today and just went downstairs to look at the test system. The pump is running but the PC isn't on. There is a deltaT of 0.06C across the Maze4 wb at 1.5GPM; raising the flow rate to ~3gpm raises the delta T to 0.07C. Guess I'll swap the inlet and outlet and see if that offset persists (in the opposite direction); if so I'll just go back and adjust the outlet temps by the appropriate amounts. Winder where that came from; I specifically remember putting both these probes into water and they read exactly the same as water temp changed...

[pHaestus]

As my above post seems to be a bit TOO fortuitous I thought I'd snap a pic:

[BillA]

you should expect a diff when flowing
the question is at zero flow rate
one would suspect that the origin will not be attainable
- but this is not the principle cause of the Wattage discrepancy

everything has been on a couple of hours beforehand, eh ?
(between the Flukes and astd dmms 2 hrs is the minimum warmup time for me)

[Les]

Originally Posted by unregistered

you should expect a diff when flowing.....
...

Perhaps 0.00045 C for the Maze4 at 1.5gpm ( 5.68lpm,0.19mH2O extrap JoeC, using W=lpm*m(H2O)/6=0.18watt)?

Edit. The graph:


Diff between dT @1.5gpm and 3gpm is ~ 0.0013 C

[Incoherent]

Originally Posted by pHaestus

As my above post seems to be a bit TOO fortuitous I thought I'd snap a pic:

Hmm.

I figured about 0.0649C. I guess math matches reality once in a while. Cool.
Better verify that the lower thermometer is the outlet T? The offset I see is negative.

A table of the offset vs flow rate with no heat load would give us a calibration.

Happy.

Cheers

Incoherent

[BillA]

on the topic of: Some Interesting/Confusing Data from testing



recognize it ?
gonna add whole new complexities to testing our little toys

[freeloadingbum]

Is 24.71 the water temp you test with? The offset might be a little different at 28C for instance.

[pHaestus]

I generally test around 19-20C; the pump imparts a substantial amount of heat into the water, and when the fans on my radiator aren't running this can warm the water up a fair amount. I just swapped the probes (put the inlet probe into the outlet probe's fitting and vice versa) and now the top reading is a little higher. I'll let it return to equilibrium and then check it in an hour or two. I'll replace the probes as they were for all of the earlier wb testing and see if I can't generate an offset vs flow rate curve for you guys tonight too.

This is actually a VERY good thing; I can deal with a little offset in my thermistors a hell of a lot easier than I can more fundamental problems with the experiment (like large secondary heat inputs from mobo traces).

[pHaestus]

Bill ummm trying to guess what could give you such a curve... Heat pipe of some sort??

[pHaestus]

OK I am officially dumb. I spent about 30 minutes looking at my test system this evening and FINALLY realized that the probes aren't all configured the same way that they used to be. When I replumbed everything to get the wb inlet and outlet probes closer to the waterblock, I must have moved one of the matched set of thermistors around because that probe is now between the pump outlet and the radiator inlet. I have a fourth probe that wasn't in use that is brand new. It reads 0.02C higher than the probe I have in the inlet. What I'll do tonight is run a flow test with and without PC running. This will give me the offset over the whole range and should be able to verify that this was the problem with the earlier data.

I am JUST smart enough to be dangerous

[pHaestus]

Ok makes sense now and looks sensible. Here's W estimates from flow rate and delta t across waterblock (adjusted 0.02C to account for difference in probe without PC running):




Block used in today's testing is the Swiftech MCW5000-A; deltaT CPU-wb in looks like this:



If you take the average watts from top graph (73.14W) then you can estimate a C/W:



All becomes clear now (and more test data as f(MHz and VCore shall be collected over the weekend)