"Personalized" Radiator Testing on the Torture Rack

[shazza]

.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.

Yeah ... I thought the results were reasonable. I know doing this is certainly helping me understand how things work with respect to temperatures.

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?

The silicone is working fine - may have a problem down the road, but so far no issues. I did get the 3M epoxy referenced in the link I gave, as well as some syringes, but now I need to order a few spare sensors before I start messing with them.

[shazza]

Hmmnnn ... got some preliminary results on the Swiftech MCR 420. Delta of 3.7 with the Gentle Typhoons @ ~1400 RPM (compared to Delta of 4.3 for the MCR 320 under same setup).



Here's a graph of the data for a two hour period from midnight to 2AM - actually have a lot more data, but it takes forever to set it all up - just about kills my MacBook Pro - Core 2 Duo @ 2.8GHz. If nothing else, this exercise makes me appreciate the i7 920 for data crunching.







Used the same scale as before for comparison purposes. Temps were very stable. BOINC used for load - consistently running above 98%. Now trying a run with OCCT, and I'll also be going back to recheck the MCR 320 ... my Delta seems a tad low from that last run, so will double check my data and rerun the testing.

[kgtiger]

Looks really stable, seams you have found the ideal time to test.

I see that you use Gentle Typhoons fans, has there been any testing here at XS of which fans are the best to use.
I have read else where that the Gentle Typhoons fans come highly recommended.

[alacheesu]

I'm following this thread with interest as I plan to do some testing of my own when time and money allows.

Grats on mod (I think?) status, Shazza. You're certainly someone that will fill that role perfectly.

[Angmaar]

Wow, there's a lot of information in all the graphs.
If you are stress testing with WCG/BOINC make sure you pick just one project (eg. HCC) because the amount of stress the different projects put on the CPU can vary.

( 800th post!)

[eth0s]

Awesome testing Shazza!

I have a feeling that as your testing methodology gets better and better, you may find yourself rivaling the testing "greats" that we all know and love.

My main point: I think you have communicated something important here, even with this preliminary test: This thread shows just how important ambient air temps are to so-called "water cooling". Your "sine-wave" graphs are really great: they really show the effects of the AC cycling on and off in your garage! I don't think anybody has demonstrated that before (I have never seen a graph showing changes in ambient air temps before). I realize that it was not your intention to show how sensitive the water temp is to the ambient air temp, but I think it is great data for everybody to see! (It may sound quite obvious to you and me, but there are plenty of people out there who still don't seem to understand the point that the performance of your "water cooling" system is entirely dependent on the ambient air temp.) Again great work! I think we have a new tester to rely on, in addition to Martin, Skinnee, Vapor, Bundymania and HESmelaugh! (or at least I do anyway.)

One last point: I think you also showed that the venerable old MCR320 is still a great performing radiator! I am impressed at how well it did in your tests. (I have an old MCR320, but I haven't used it for years, and I have instead preferred to use the more expensive competitors, but I am re-thinking that idea, because of this post! [Note: Just so you know I am not a shill for Swiftech: I am still mad at Gabe for selling me that friggin' Apogee GTX with an aluminum top! What the heck was he thinking when he made that? And what was I thinking when I bought it? Shame on you Gabe for selling that turd!])

Thanks Shazza, and keep up the good work!

[shazza]

...
I see that you use Gentle Typhoons fans, has there been any testing here at XS of which fans are the best to use. I have read else where that the Gentle Typhoons fans come highly recommended.

The Gentle Typhoons seem to be all the rage right now - they are very quiet and seem to be good performers. You can check out Martin's review here.

I'm following this thread with interest as I plan to do some testing of my own when time and money allows.

Grats on mod (I think?) status, Shazza. You're certainly someone that will fill that role perfectly.

Thanks, alacheesu (I think ) GL with your testing.

Wow, there's a lot of information in all the graphs.
If you are stress testing with WCG/BOINC make sure you pick just one project (eg. HCC) because the amount of stress the different projects put on the CPU can vary.

( 800th post!)

Congrats on 800+ And ... thanks for that tip, not something I would have thought of!

Awesome testing Shazza!
...

You are too kind, eth0s. I do not believe I see a white lab coat in my future


Thanks for the comments, everyone. I figured this thread is duller than watching my paint dry in my current work log, but I very much appreciate the guidance and suggestions.

And, as for figuring out the impact of ambient temps / Air Conditioning cycling, and more ... check out the next post for more lovely "background" data.

[shazza]

Another test run last night, using OCCT for loading. I was a bit bothered by the fact that I seemed to hover around 98%+ while using BOINC. Well, guess what - running OCT I was right around 95-96% (which is lower than some previous runs). Only other thing I had open was WinTest for temp monitoring. Who knows what is lurking in the background of the new Windows 7 OS???


GOOD NEWS is, I definitely seem to be able to get stable results. Bad news is, I haven't yet determined my protocol for comparison testing. I'd much prefer to use BOINC, because I'd like the rig to crunch all night. I'll give just running one project a try. I'm okay running BOINC as my load, even if it gives me a slightly different result than I get with OCCT, as long as I can get consistent results and also have an idea of the magnitude of differences between the two methodologies. After all, one of the main reasons to do this is to set up a crunching rig.


Before I get to the graphs, a comment on my preliminary test results for the MCR320 compared to the MCR420 - yesterday I saw about a Delta of 3.7 on the 420, compared to 4.3 on the 320. Today's results are showing a bit higher on the 420. Still working through whether it might be due to OCCT vs BOINC (or possibly the fact that a jug of distilled water was sitting almost in front of the radiator for last night's test - I've got the fans pulling air into the radiator, but still could have had some impact).


OK - this chart just shows a prolonged testing period of about 8 hours. First 4 running OCCT, second half at idle. You can also see where the trusty A/C kicks on. And, interesting to note that you can also see the "Air-In" temp dropping at the same time OCCT quits. Probably not a coincidence, but more likely a function of the ambient air surrounding the Torture rack being impacted by running at load - told ya these i7s are space heaters! I can also play around with some additional air temp sensors. [I have enough trouble with thermodynamics, I don't need the Heisenberg Uncertainty Principle toying with me]





The second chart is just the standard CPU 1 graph from OCCT:

[BrokenArrow]

Is your water temperature probe still in the reservoir? Or did you move to the Tee section?

[shazza]

Is your water temperature probe still in the reservoir? Or did you move to the Tee section?

Still using the reservoir. Did get my supplies to make up some more probes, just have to do it!

[Vapor]

In WinTest, is everything disabled in the "Packet Debugger" section?

I get high-99s for CPU usage % with small data set setting in OCCT v3.1.0. If I don't disable everything, I get like 90% (but I have 24 sensors in use and all 4 RPM counters going, so I bet that increases the overhead).

I'm using 64-bit Vista SP2, if it matters.

[shazza]

That could well be it, Vapor. I have been clicking the "Packet Debugger" button and unchecking everything, but can't confirm I did that for the 2am run (and, I bet I didn't, because I restarted the computer, trying to get rid of any nonsense I had running). Will redo tonight. Thanks again for checking in on the noob, especially since you must be tired from your recent review!

[Vapor]

If you can, put the CF on a separate PSU (P-PCs and a few other places sell laptop-like 12V PSUs with a molex connector and molex to floppy are available many places) and hibernate instead of shutting down. That's what I do (well, the CF and my pumps are on a Zippy that I only turn off to switch blocks)....that way you won't have to uncheck all the Debugger stuff and check all the logging stuff each time you power back on

Now I just mount, power on, enter my OS account, hit GO in OCCT and check the "Log to.." button in WinTest when I start a new test

[kgtiger]

Thanks for the link Shazza, much appreciated, I will take a look.

Keep up the excellent work, you know the more questions you ask, the more answers you get.

[xeon_1]

Shazza i have a question/request for you.
if you add a 4870x2 to that loop with the MCR420 would it hurt the cpu temps significantly?
Adding 15 to 20° extra?
Or would it be more in the range of 5° to 10°C
And what would it be with the MCR 320?

The reason i ask is simple i want to cool my setup with water and maybe overclock a little but the main reason is reducing the noise from my pc
Thanks in advance.

[shazza]

Xeon - I plan to do some testing with CPUs, but am away for a week, so it will be a while. I'd suggest getting an approximate heat load using the eXtreme PSU calculator, then using that with the radiator test data provided by folks like HESmelaugh and skinner.

My best guess is the 420 could handle it ok, but wouldn't hazard a guess at actual temps. Some are successfully running CPU am gpu on a triple rad. Temps would obviously be affected by fans, ambient, and OC.

[bentleya]

I have just been trying different ways of applying, "Load" to a CPU this morning, As i now have a crystalfontz 365 set up on my test bench. However i was trying OCCT 3 as of the logging capabilities, which is great. I have noticed my Core temps are 10 degrees less then when using core damage.

Has anyone else noticed variations between different programs?

[kgtiger]

Not that I know much about this, however If both programs are running at 100% load (or near that) I would have thought the results would have been the same?
Has air temp been the same?

[bentleya]

Yes, on both. i've stepped the cpu up to 3.6 GHZ, 20x 180 MHZ at 1.35V and i get the same temp reading as core damage. However i dropped it back down to the stock 2.67 GHZ and the temps are higher still with core damage.

OCCT, 60-58-58-54
Core Damage, 67-65-65-62


Air temp is 22 (within .4) in both.

[shazza]

I've not used core damage, so can't offer much. I do know that I get consistent readings with OCCT / Everest / Real Temp / EVGA E-Leet / HW monitor / BIOS - understanding that some of these use the same readings.


By the way - I have not abandoned my testing. Just been tied up with the build I have going on. Also, we are getting ready to pack stuff up to head back to Florida, so don't want to get too involved til I've got everything set up there.

[eth0s]

Also, we are getting ready to pack stuff up to head back to Florida, so don't want to get too involved til I've got everything set up there.

So it's winters in Arizona and summers in Florida? Shouldn't it be the other way around? (jk)