MCP 655 better on 1?!

H

headala

Limp Gawd
Joined
Jul 26, 2005
Messages
432
Hey, guys (and gal), I have had some puzzling results with my Swiftech 655 pump. Basically, I've experienced a 1~2 C drop in temps when I turned the pump down from 5 to 3 to 1.

Here's my setup:
Res -> pump -> swifty 120.2 rad -> storm orig rev -> AC twinplex GPU -> AC twinplex NB -> Res. Tubing is 7/16" from res through to storm, then 3/8" tygon from there to res.

CPU is Pentium D 820 @ stock 2.8 Ghz. MB is Gigabyte GA-8I955x Royal (Intel 955 chipset). All voltages are normal. Used Arctic Ceramique

Pics of system:
IMG_7199Medium.jpg


IMG_7198Medium.jpg


Screenies of temps. Note the times in Aquasuite. Ambient was constantly 79F. I had the system on idle for 20 minutes before, then logged data for 13 minutes with pump on 5.

pumpon5.jpg


Then shut down system, turned pump down to 3, restarted. Logged data for ~6 minutes, then turned down to ~1.5, logged for another ~7.

pump3to1.jpg


So, my questions are this:
1) Why do my temps fluctuate so much? Bad mount? I've tried to wiggle the storm to see if things change, but they don't.

2) What does the lower temp both from the cpu and the water with the d5 turned down mean? Is this peculiar just to my system?

3) Would overclocking change this at all?

Thanks for the help and info.
 
I would suspect that your testing procedure accounts for this oddity.

Try this:
From a cold start (system has been off for 8 hours) run pump on #1 setting and Prime (or pick your favorite stress test) for one hour.

Power off and let let system sit for eight more hours then repeat procedure with pump on #5.

Assuming that room ambient has remained constant you would expect the second test to produce lower temps.
If it doesn't, then you do have an anomaly to explain.
 
Shhhh, quick, someone delete this thread before TopNurse see's it :).


I have lower temps on 1 too. Maybe there IS something to that Aqua computer nonsense.
 
^^ and never you mind that one of the sensors used for this "test" is peaking and dropping all over the map.

why do they spike and dip so much? what is the actual temperature? why do i always say that evaluating performance based on internal diode readings is worthless?

a question to ask might be how fine the resolution on those A-C temperature sensors really is. the difference that you are noting is only a few tenths of a degree, which could well be below their threshold for accurate reporting.
 
^^ and never you mind that one of the sensors used for this "test" is peaking and dropping all over the map
Click to expand...

That's one of my questions in the OP. I'd love your thoughts on that if you have any.

a question to ask might be how fine the resolution on those A-C temperature sensors really is. the difference that you are noting is only a few tenths of a degree, which could well be below their threshold for accurate reporting.
Click to expand...

Could be true, but then the Aquareo would be very smart as to when it 'chooses' to be inaccurate i.e., right after I drop the pump from 3 to 1.

I'm not trying to say anything by all this or jump to conclusions, I would really like your help in understanding what's going on!
 
headala said:
That's one of my questions in the OP. I'd love your thoughts on that if you have any.!
Click to expand...
honestly, if i had a good answer for that one, i would have put it up.

could be the chip, could be the the motherboard, could be the software. there is so much fudge put into a thermal reading from the chip before you get to see it that you never really know just how accurate and consistant that number is.

i don't know where the ball is being dropped in your case, nor do i know how.
headala said:
Could be true, but then the Aquareo would be very smart as to when it 'chooses' to be inaccurate i.e., right after I drop the pump from 3 to 1.

I'm not trying to say anything by all this or jump to conclusions, I would really like your help in understanding what's going on!
Click to expand...
it's a conspiracy! the aquaero is in on it! ;)

seriously, i'm not trying to suggest anything quite that silly.

so, it never dips and varies down to those values when running normally at a constant setting of 3? there should always be some jitter to the values, if you have the reporting interval really high. i don't know if that's an option with the aquasuite or not, and i don't know how frequently you have it updating at.

it could be any number of things, and who know what's really going on in your particular loop.

another thing that could be happening, if you're using a typical thermal probe install via a T-line, is that when flow is sluggish, the sensor gets a reading from stagnent water that is not as warm as water flowing in the rest of the loop, where when the flow is moderate at setting 3 the sensor is getting warmer coolant swirled to it from the loop.

i hope some of the guys who have a background in fluids pop up and offer some well backed up theories to explain this one.
 
the 655 dumps a crapload of heat in the loop, when i was bleeding/leak testing for a few hours my rads were warm to the touch, although it is 1-2C cooler when i run it at 4-5 instead of 1-2
 
The inline heat dump for the D5 is not that signifcant. You can calculate rise in water temperature based on the simple wattage output, and you'll find that it's not significant. The heat dump of many pumps will raise the water temperature enough to make the radiator and tubing feel slightly warm. The delta T between the water temperature and the ambient air is not high, so very little heat will be dissipated from the radiator(s). Once you add a more substancial heat load, and the delta T rises, this is where you'll find that the water temperature will begin to "level off", even when adding more components that don't add significantly to the water temperature.

Touch your radiator and your tubing while your entire machine is on. Are they noticeably warmer than when the pump is the only running component? Not likely.

headala said:
That's one of my questions in the OP. I'd love your thoughts on that if you have any.
Click to expand...
I fully agree with Daishi on this one: onboard sensors are almost entirely worthless. If the sensor is relatively accurate, which is not likely (past certain thresholds, some sensors may multiply changes by two, for instance), then the only real explanation is that your radiator isn't performing highly effectively.
 
Not sure about you guys' pumps, but my MCP655 whines at lower speeds. It's actually quieter on 5 than 1.
 
Okay, thanks for the thoughts. Phide and DFI Daishi, what do ya'll think would be the best way to measure the core temp accurately? It's a Pentium D 820, so it has a full heat spreader. If it didn't I would just lay a thermal sensor next to the core...but how do I get a sensor close enough for it to be accurate?
 
my first thought would really be pump heat dump but only cuz of the fact that he is not Oc'ing. the storm would always like the more flow you get from settings 4-5.

for example, i switched pumps myself from the dd5 to the ddc and the much smaller heat dump did result in slightly diff temp characteristics at idle and load. just a thought.

oh and im required to echo wholly inaccurate testing procedures, on board sensors yada yada.

but then i will say, if it makes you happy to run your system one way then do that. its functioning and it works so enjoy.

op you have PM about something OT
 
headala said:
Okay, thanks for the thoughts. Phide and DFI Daishi, what do ya'll think would be the best way to measure the core temp accurately? It's a Pentium D 820, so it has a full heat spreader. If it didn't I would just lay a thermal sensor next to the core...but how do I get a sensor close enough for it to be accurate?
Click to expand...

Find a machinist with a very steady hand and cut a groove in the IHS which follows the specs for the Intel TTV. :) Then you just need to get a nice T Type thermal probe and some special thermal paste which hardens in the groove to keep everything in place. That would be the most accurate temps you can get.

About the phenomenon...bad temperature sensors, and slightly variable ambient temperatures. If you look at the data again, you notice that for the last half of the first test the temps seem to be floating around 33-34, with a good bit of 32's in there. The second test is floating around 33-32 with a good bit of 34's in there. So the data is actually very close. A very slight change in ambient temps could easily be to blame, or simply an inaccurate temperature probe. You are talking about very small differences in temperature here. Also you only ran the data gathering for 15minutes and then 6 minutes? That is far too short. Finally, you ran the test on idle, lots of times there may some program that is using up 5% CPU during one test and it isn't running on the next one, which would result in slightly higher heat output, and thus higher temperature. Try running the test as suggested, from a cold boot and testing 100% load temperatures. This ensures that you have as uniform a heat output as possible. Also gather the data for longer than 5 or 10 minutes :p .
 
You may not even be particularly interested in temperatures that come close to representing the temperature of the die. All one really needs to know is how temperatures change when other variables change. The problem with cutting a groove in the heatspreader and installing a probe in that groove is that you're sacrificing some cooling performance for little reason, unless the groove is very short and quite shallow. For most, being able to accurately monitor any point of the IHS is an improvement over any onboard sensor, so affixing a flat copper tip type probe to any side of the IHS is a reasonably easy and accurate way to go. The only issues here are that temperatures of the IHS will not fluctuate nearly as rapidly as the core will, meaning you'll need to load for a longer period of time before the temperature will level off, and changes in airflow over the probe tip can skew the intended data if the installation isn't pretty damn nice.

You're not going to find many who you can compare thermocouple temperatures with, as mounting locations will vary and type T thermocouples are only accurate to about ~1C, but you can get a reasonably solid baseline that you can compare when changing your loop around, re-mounting or adding components that won't tend to be influenced so heavily by other factors.

But if you really want to know what's going on, grooving your IHS and probing it and also adding one or more inline coolant probes is the way to go. Water temperature can sometimes be more valuable than the CPU die temperature.
 
You must log in or register to reply here.
Back
Top