"Personalized" Radiator Testing on the Torture Rack

[shazza]

No worries, kgtiger - I know it's late in Perth. It's unlikely that your water temp is going to get high enough to worry about it. For example, if ambient is 25 degrees, and you have a 5-10 degree delta between air and water (which, from the graphs I've seen in skinnee and Martin's tests, is probably where most of us are), you still have a long way to go before you need to worry about your pump. In fact, your CPU might well have shut down by then.

Skinnee can correct me if I've got it wrong.

[skinnee]

Thanks. That's what my husband tried to tell me, but I'm paranoid from having killed my share of hardware

Yeah, its never fun killing hardware... but it comes with the territory.

*snip*

Perhaps I am just too tired to think straight.

Hope you are both well.

You lost me... but I wouldn't even venture a guess on water temps because I would probably be wrong.

edit: shaz, you've got it. You can only estimate.

[kgtiger]

Thanks Shazza and Skinnee, I know you can only give an estimate to my long winded question, sorry I lost you there Skinnee.
As I have never had any experience with building water cooled PC's before, or an PC for that matter, I don't know to much about it. However not knowing has too, has never stoped me from learning how too, and I am stoked that I have found a place with knowledgeable people that are open to giving there time to answer questions to people that are interested to learn, thanks again for that.

I will be taking the leap into building my first water cooled PC hopefully in the next 3 months.

[shazza]

Looking forward to your build, kgtiger. Keep reading, learning, questioning, and planning.


OK, Got the Crystalfontz 635 with SCAB hooked up. Got my makeshift "water-proofed" probe attached and now just checking things to see if they make sense. This is a 30 minute run, where I measured Air In @ the radiator, and Water in the XSPC reservoir (CPU at 4 GHz, 1.46 Vcore, 98%+ load crunching):








Now, off to investigate the major force behind my moving Delta ...

[BrokenArrow]

Nice Shazza. Nothing beats the accuracy of those Dallas temperature probes.

I have always wondered what affect of the surrounding ambient temperature plays on a reservoir mounted temperature probe versus one that is actually in-line with the water flowing by such as one inside a Tee.

I am sure it is very little difference, but your probes would probably show the difference...hint hint.

Also, this ambient affect may not be so apparent on the torture rack as opposed to a reservoir inside a closed "heated" case.

[Vapor]

Looking forward to your build, kgtiger. Keep reading, learning, questioning, and planning.


OK, Got the Crystalfontz 635 with SCAB hooked up. Got my makeshift "water-proofed" probe attached and now just checking things to see if they make sense. This is a 30 minute run, where I measured Air In @ the radiator, and Water in the XSPC reservoir (CPU at 4 GHz, 1.46 Vcore, 98%+ load crunching):








Now, off to investigate the major force behind my moving Delta ...

There's a 'delay in response' because of the volume of water. If you look, your air curve is some 30px to the left of the water curve (no units on the bottom, )

Remedy by including a warm up for however many minutes it takes your system (room ambients included in the 'system') to reach equilibrium

Looks like a great start for sure though

[Raptor-X]

Looking forward to your build, kgtiger. Keep reading, learning, questioning, and planning.


OK, Got the Crystalfontz 635 with SCAB hooked up. Got my makeshift "water-proofed" probe attached and now just checking things to see if they make sense. This is a 30 minute run, where I measured Air In @ the radiator, and Water in the XSPC reservoir (CPU at 4 GHz, 1.46 Vcore, 98%+ load crunching):








Now, off to investigate the major force behind my moving Delta ...

Could you have had a central Air Conditioning System kick in at about the 75% point on your timeline? The air temps are falling off at that point..

Edit:
Stable air temps should help stablize your deltas bigtime.

[shazza]

Bingo ... I figured you smart guys would figure it out. Also, a shout out to skinnee for telling me to use WinTest to log my temp sensors instead of CC2. Also, from reading Vapor's test methodology for water blocks, I saw that he found OCCT more stable, so I finally got that to work on this setup (had to enable Everest to allow external program access, cause I was getting crazy voltages using the built-in OCCT software).

It looks like the A/C cycles regularly, and this gets picked up by the Air-In temp reading. Then the water temp changes, but with a lag time (I expected that, just didn't expect to see the magnitude of the effect). This was easier to see on a chart from a longer run period.

So, I guess this means I might have to use a longer testing period - maybe 2 hours? And, as per BrokenArrow's suggestion, I'll also add a temp probe in-line, once I get the new expoxy. Here's a rough chart showing a longer run this afternoon (this was taken during the start of an OCCT run, so the first minute is at idle):

[Vapor]

Do you mind making one more chart line for the chart?

A moving average of delta...start like 10min into the data and do "=average($Col$Row:$ColRow)" and drag all the way to the bottom

You should see the curve get progressively smoother, the end point is your final number

(the end point being the average of all deltas from minute 10 to minute XX )

[Raptor-X]

So, I guess this means I might have to use a longer testing period - maybe 2 hours?

You could try the following protocol;
1) Late evening - Outside ambients in Florida should approach inside target ambients. 70-75F
2) Turn off your central A/C. However, leave the recirc fan running to dissipate the heat your system is generating.
3) With no solar loading, your house temps should stay relatively stable for a 30-45 min run.
4) I believe that you will get a higher quality data set & resultant analysis if you focus of stabilizing your air temp. My 2 cents worth...
Regards,
Raptor

[shazza]

Thanks, Raptor. I agree stabilizing air temps will help a lot. Unfortunately, no real easy way to do it (currently in Arizona and the computer is in the garage), but it's getting cool enough that we soon won't need to run the A/C at all, so that might help.

Like this, Vapor? (You realize I haven't had to do my own spreadsheets and charts for a long time ). And, I think I said my Delta was 5 degrees back in Post #9

[Vapor]

Yup, that's the gist of the testing you're doing....not so much looking for the momentary deltas, but rather use the entire XX minutes of data, and that's just a way to show how the delta closes in to a single number

After looking at that graph, I'd probably start later than 10min (see that upcurve in the beginning?)...though I'm not sure how long your system will take to reach equilibrium, you'll have to play with that

[Serpentarius]

shazza ... i wonder if you guys can test mcr480 as well?
i've been searching the forum for some info ... no one seems made the first thread ...

[shazza]

Thanks much for the help, Vapor (I'm sure I'll be reaching out again, soon). Yeah - I cut off some more of the early data - Delta moving average went from 5.0 to 5.1. I'm letting things run overnight to see what my temp curves look like.

Hi Serpentarius - I do plan on testing the Swiftech MCR 420, but remember, my testing is not going to be anywhere as precise as that by the pros. I'm doing this just to learn what I can about my setup, so not playing on the same field as the big boys

[kgtiger]

Hay come on, you are a big girl now. The fact that you are playing is the main point.

Keep it up Shazza it's good to see some results, well done.

[shazza]

@ kgtiger


arghhh ... forgot to turn the data logging on last night. But, started it early this morning. Did not turn on the A/C in the garage for several hours. Runs are much steadier, as you can see in the following chart. I think I've got enough info to get going now --- let's get testing.

[Raptor-X]

Great Job "taming the data".
Your testbed will make for some very interesting studies. I look forward to them. THANK YOU for sharing with us.

[skinnee]

you're taking to this like a duck to water, good work shaz!

[shazza]

Great Job "taming the data".
Your testbed will make for some very interesting studies. I look forward to them. THANK YOU for sharing with us.

Thanks, Raptor-X - you were spot on when you said to try with the A/C off.

you're taking to this like a duck to water, good work shaz!

A very old duck Thanks again for the help.


Trying to get some actual testing done, but keep runnning into little snags. For some reason, OCCT is acting up (maybe cause I'm using it to call Everest and using Win 7 64-bit ) So, back to Real Temp for logging Core Temps and WinTest for logging fans and temps. Running BOINC for 99%+ load.

I'm going to have to be judicious in my choice of data points unless i only do runs early in the morning or late at nite. Looking at this chart, you can see where I can pull in 3 15+ minute time periods during a 2 hour run. Not ideal, but will give me the trends i'm looking for. I took the average of the 4 Core Temp readings. I have another data set for the GT fans at 1000 RPM, but still working on the data, and it's almost tail time.








And, for kicks, some pics of the Crystalfontz 635 and SCAB:


It comes in an environmentally friendly box:






All the stuff, less two temp probes in use:













Forgot to mention I had to switch to the 4870x2 because I wore out the old 7600GS. Not ideal, cause the fan blows on the res, and it's noisy on air. You can also see my almost DIY case for the 635:





Whew, between XS, my Black Pearl build, and this stuff, I might as well get a full-time job.

[NaeKuh]

Whew, between XS, my Black Pearl build, and this stuff, I might as well get a full-time job.

Or you might catch naekuhitis.

[skinnee]

and now you know why my Torture Rack is half built, and I am still working on reviews that I finished testing for months ago.

your Air In/Ambient looks to deviate 2-3º over the 75 minute logging period, which is actually pretty good. You're going to have some variance unless you take measures to have finite control... you'll eventually end up at Environmental Chamber, just accept the variance and note your "normal" variance.

Whats odd is the trend of your dT compared to the variance of your Air In and Water temp...hmmm...only thing that comes to mind is BOINC, but I could be wrong.

[Vapor]

Science below!

The 'spikes' in dT should be counted!

Those are nothing more than the cresting/valleying of the air/water graphs being out of sync. When you change air temperature (if you have a space heater, below is a fun experiment to try), the water temperature will respond--but due to the volume of water in the system, that response is not instant, in fact, it's delayed very strictly based on the volume of water you have (and the efficiency of the energy transfer and the substances in play [mostly just water in this case]).

If you slid the water temperature graph X minutes to the left (I would look at when each curve crests to find what X is) and synced them perfectly, you'd eliminate the peaks but the general line would be a little higher. Why?

There are three factors to consider (for now):
1) temperature is the average of energy in a system
2) two systems (room air = one system; water in the PC = other system) take time to reach equilibrium when energy is changed
3) at the same time, one system is simultaneously increasing its energy (room is heating up slowly)

But because of the first two factors, any change in energy of one side of the equilibrium will take awhile to completely effect the other side. "How long" is dictated by the mass and thermal properties of the material. Add in the third factor (a slow increase in energy/temperature), and you get a "lag" in the display of the two. Because the room's energy is increasing (i.e., going toward the loop's energy) and the water's response 'lags,' you're actually undershooting the real difference in temperatures.

Now let's reverse the third factor--let's cool the room back down (quickly)! What you get when you do that is one side of the equilibrium is brought down to nominal energy within a few minutes is that the lag still exists, but now it's effect is negative. And because it's a faster change in energy, the effect in delta is magnified and the dT appears to spike.

So how do you fix it? Well, you can 'cheat' and just resync the energy (temperature) curves (not so easy sometimes) or you can roll with the punches of physics and use the fact that temperature is an average of energy and just average the temperature data you have!

First, let's just examine why resyncing is difficult: you need to either have incumbent knowledge of the mass (and in turn, volume) of the water in the system as well as knowing how efficient the rad is (though this plays a small part in the numbers, you do need it in order to be precise)...but if you knew how efficient the rad were, you wouldn't be doing these tests in the first place. Or you could just closely examine the data and figure out how long the 'lag' is and resync that way....but if you already have all the data, why not just simply average it?

Here's an example of why I wanted you to do a moving average of the data you acquired...below I've called it an "extending average" since a moving average typically has a finite length of time over which you're averaging....I'm just adding in every bit of data I get into the average and extending it ad infinitum.


For simplicity's sake, I used sine waves for Air and Water temperatures and a predetermined delta of 2C between the two (as well as a 'lag' of X=.25). You can see the delta curve oscillating (much like yours, just more regular in shape) between ~1.5 and ~2.5...but the number we ultimately care about is the average of all the dT's...and that's expressed in the moving average. As time goes on, that moving average hones in closer and closer to its actual value of 2. An important feature of the extending average is that it doesn't matter where I start or where I end (for instance, I started it at the worst possible time and ended it at a random time and the result still came to 1.992), just that the longer I go, the more reliable it becomes.

If I were to resync the sine curves, the green line would be flat at 2, but of course that would have required I know how to resync them (which isn't automatic when you're acquiring 'random' data).

In light of the graphs you've shown, you're only counting the valleys of the delta, which while they look nice, they only tell a part of the story about the performance. You need to include all of the data! If I were to redo my data with a worse radiator (higher delta) with more mass to the system (larger volume), the valleys end up looking the same (~1.7C), even if actual performance of the radiator has changed significantly.

Anyway, long story short...use all the data you get after a warmup!

EDIT: forgot to add the fun experiment Using a space heater, place it next to the rad and crank it up and do your testing (well, start the testing before you start the heater)...your air temps will come really close to your water temps as the two increase (appearing to close the delta)! It'll almost be like hotter temps = more efficient rad, but that's not the case, you're just messing around with physics

[shazza]

Thanks for the input, Vapor. I'd hate to let science get in the way of my little project, but I see your point.

It goes against the grain to remove any data - I hate having to make subjective guesses. Was just trying to eliminate the variable of the A/C cycling off and on. I think sticking to late evening / early AM and using all the data as you suggest is the way to go.

I did a couple of runs last night and a quick look at ALL the data shows me the following (this a ROUGH look at the data, just used simple average for each column - will redo in a bit):


Delta of 4.3 with Gentle Typhoons at 1400 RPM on MCR320
Air Temp: 27.7
Water Temp: 32.0
Avg Core Temp: 61.9
Avg Core T - Air T: 34.2


Delta of 6.3 with Gentle Typhoons at 1000 RPM on MCR320
Air Temp: 27.0
Water Temp: 33.3
Avg Core Temp: 64.1
Avg Core T - Air T: 37.1



This is @ 4.001 GHz and Vcore of 1.388 using BOINC for load.


Using the eXtreme PSU calculator, I come up with 241 Watts for the CPU. When I plug this info and my air temp into skinnee's water temp estimator spreadsheet, I get Delta of 4.9 for 1400RPM, and 6.8 for 1000RPM.

The CF 635/SCAB is controlling the fans. At 100% I'm showing about 1400RPM, at 58% I'm getting about 1000RPM. I did notice the numbers jumping around on one of the fans when set to lower power. So - I'd just use my 1000RPM numbers as estimates at the moment.


Seems reasonably close for all the variables involved. My main concern is being able to get consistent results so I can comfortably compare data for MY SETUP - thus eliminating the worries that come with using the assumptions when looking at professional test data. Good thing is, my temp sensors seem to be doing a decent job, and I'm refreshing my Excel skills.

Back at it ...

[skinnee]

.5º and .6º on the estimator... thats not bad considering the fan difference.

You do have a really good baseline though. Re-tune your excel skills one test at a time, too much and you'll go bat crazy.

[kgtiger]

Hay Shazza, how did the silicone work out for sealing your temp probes, and did you end up getting anything else to try sealing them?