Little Arse Radiator performance results

[Cathar]

I've tested the Little Arse Radiator

I used an Eheim 1250 pump, a Silverprop Cyclone 5 water-block, and the heatercore with the shroud as pictured in the linking thread using a 104CFM Panaflo at 12V.

The CPU heat load is set to 125W (~1903MHz @ 2.10v), and full load was measured using SiSoft Sandra Burn-in Wizard looping just the MultiMedia test for 1 hour.

There are no CPU idling programs used, so the AthlonXP is churning through electricity at idle like they normally do.

I'll just report the water temperature rise above ambient that I saw.

Idle: 2.2C +/- 0.1C
Full Load: 3.1C +/- 0.1C

When I say +/- I mean it. The thermal probe varies around those temperatures by that amount. This seems to happen very slowly and it does indeed oscillate up and down, probably due to various feedback air flows in my office.

I'll measure the performance now with the fan at 7V and post in this thread to update. The Panaflo at 12V by itself is noisy but bearable. Nowhere near as bad as the two Panaflo's at 12V side-by-side on the Big Arse where there is a distinct chopping roar as the two fan's airflows clash at the outlet side in the middle.

The Little Arse with the Panaflo at full speed performs about as well as the Big Arse does with one of the lower (and much quieter and less airflow moving) Enermax's at full tilt.

I must say, I'm surprised at its performance.

[Bravo]

Cathar, the figures are impressive, especially considering the space contraints

How much again ?

[SWEDKG]

Cathar, are those the specs of your Enermax ?

12V, adjustable between 6-13.8 volts
0.3A, 3.6W, 22-39.6dBA (67-95 cfm)
Speed settings 1500-2300rpm
120×120×25mm

The Panaflos are 25mm or 38mm in height ? you are talking about 105 CFM Panaflo, is this the H1A?

i'm about to get myself a rad this week, and i wonder if i should get an Enermax, or stick with my H1A and M1A at 5v...

[MWP]

Im not too supprised...

Compare the size of a good heatsink to a radiator.
Radiators are MUCH larger than heatsinks and xfer heat to air much better.

For normal use with CPUs (without peltiers or anything high power) there is no reason anyone should need a big radiator.... really using even a small radiator is overkill anyway.

[Cathar]

Okay, some water temperature results at 7V.

Idle: 4.0C +/- 0.1 above ambient
Load: 5.5C +/- 0.1 above ambient

The Enermax's are weird. I can't find any information that seems to be consistent about them. They are 25mm in height and given the rated amount of air that the Sunon and Panaflo's push at 12V (108CFM and 104 CFM respectively), I'd say the Enermax's push around 75% of that from just a subjective had in front of the fan kinda measurement.

The Panaflo's are the 38mm height -H1A's.

Bravo: It costs $90 for the bare unit. $40 for the shroud that will fit either side. $25 extra if you want me to paint it.

Who wants this one?

[BillA]

understanding that the temp gradient is:

air -> coolant -> wb bp -> die

a rise in air temp shifts all up, but not by the same increment
a decrease in air flow does the same
and again swapping to a smaller rad has a similar effect

note that this test was not apples to apples
swapping out the rads also changed the flow resistance hence the flow rate as well

the effect of a 2^C coolant temp change is probably 10 to 15^C at the die

be cool

[Cathar]

Hmmm, the on board thermal sensor reports CPU temperatures as (roughly) consistent with the water temperature increase.

The CPU itself also overclocks to the same level and stable. A stable CPU overclock (for me) is the biggest indicator that the CPU is running cool.

With the BA rad, the CPU would be become unstable at 1925MHz if the water temps rose above 27C (due to a hot day), and the same seems to be in effect here.

I'll measure the flow-rates later on. I have great trouble believing a 2C coolant temp change would cause a 10-15C at the die. I've run my radiators fanless at times causing 50C water temps with the Black Ice Xtreme and the CPU still read the typical rise above water temp.

AMD CPU's cease to function above about 95C regardless of how lowly they are clocked, this is known as the critical gate temperature above which signal integrity starts to fail within the CPU.

There is no way that there is a 5:1 die:coolant ratio happening.

[BillA]

I should, by now, know better than to post about temps
but, one more time

The Overclocker's Mantra

all temps are crap, but watts are worse
oooohhhhhmmmmmmmmmmmm

how many lessons do we need ?

you say your overclock becomes unstable at what temp ?
then, thinking backwards, what die temp might that suggest ?
IGNORE what you are reading, what is the CPU telling you ?
something about 95^C ?
and what is the temp that your "on board thermal sensor" reads ?
NOW you have a known relationship
-> you are observing 'compressed' temp indication
a great deal has been written on many forums about this, I'll not repeat any of it

so you put 125W where ? just EXACTLY WHERE ? do you know ?
"The CPU heat load is set to 125W (~1903MHz @ 2.10v), and full load was measured using SiSoft Sandra Burn-in Wizard looping just the MultiMedia test for 1 hour."
pardon me Cathar, but I'm getting wound up - again

these Watts are genuine BULLSHIT !; and you, by now, should know so without a ration of s**t from me
is the MultiMedia test for 1 hour maximally loading ALL CPU functions ?
has someone put an ammeter on the CPU leads ?

let me quote saba (without his permission, sorry) from a better place:
"What you really need is a stress-test program that will really hammer all aspects of the system. This means CPU, RAM, AGP, PCI and I/O factors.

Although it sounds rather drastic, try perhaps installing a simple web-server program and using Intel's IOMeter program to literally continuously hammer your system. Although it wont tell you specifically whats dead or not working, I find it really does put a huge amount of stress onto the system, which is always what stability tests are all about!

If I personally wish to stress-test my system, I normally install IIS (from your windows 2000 / XP cd's) and share some largish files (DivX movies work well). I set up IOMeter for 16 continuous WWW testing threads.

Starting a full recursive defragmentation within OO Defrag 2000 (www.oo-software.de, damn good defragger!) atop this is also very good. I also run two concurrent sessions of Quake II in a window, and let them demo cycle. Finally running 3dmark over the top of all this chaos."

THEN you MUST address the secondary path heat losses !!!
and until you address these factors you are playing with yourself
we have been here before, you're BSing

here is a fact, measured, repeatable
130 Watts with very low secondary losses = ~ 60^C at the die
and this is with coolant at 25^C, 1.5gpm, and a good wb

Cathar
your temps are only relative
your temp range is VERY compressed
your notion of applied Watts is fantasy

a suggestion:
your rad tests provided some useful comparative data (real data, and this is good)
your thoughts on die temps are speculation, and inaccurate
-> consider sticking to that for which you have data

be cool

[Cathar]

LOL!!

[Cathar]

Allow me to clarify. I'm not out to prove the C/W of the system, or to prove that the Wattage is correct, or even the temperatures of the CPU die.

Radiate gives a calculated wattage. Whether it's accurate or not, it's something that people can compare to.

The reason I use Sandra Multi-Media test looping is because, for me, I've never seen my CPU get hotter using any other stress test. Whether it accurately gives the Radiate program's output of 125W is irrelevant. The whole point is that it is a easily reproducible test that I can run to compare systems to. Whatever the wattage is that is actually being output by the CPU is consistent between test setups.

All I'm measuring here is the water temperature rise above ambient. Now you yourself has said that heat in = heat out, regardless of how fast the water flows or any other factor. Yes there are secondary heat paths, the size of which dictates the temperature reading on the on-board thermistor in the absence of any airflow about the socket which I've tried to do. The fact is that the thermistor gives reproducible results between different tests, however accurate or inaccurate it is.

Your rant above seems to be focussed on something that I never stated in my opening post, which makes me wonder where it's coming from? I measured the water rise above ambient temperatures with a CPU heat load that's reproducible (regardless of what it may actually be) and you're here on some rant about how this translates into extreme CPU temperatures, which isn't the point of what this thread was about. The CPU heat load as given by Radiate at least offers people who also have Radiate some basic form of reference to make comparisons with, again, regardless of actually how accurate they are.

I never stated that I believed the CPU wattage is accurate, but you seem to be intent on drumming into me something which I never stated as being fact.

As for CPU stability, temperature plays a role, as does voltage and actual clock frequency. Just because a CPU fails or is instable, doesn't necessarily mean it's hit 95C internally, it can just mean that the signals are moving too fast to retain a coherent state.

I can make the CPU fail at 1.75v at 1800MHz, where otherwise it's stable at 1925MHz @ 2.10v. It's certainly going to be a lot hotter at 1925/2.1v. This is just highlighting that CPU temperature is not the only cause of instability.

As for a 2C rise in water temps correlating to a 10-15C rise in CPU temperature, let's look at it this way. What then happens when the water temps drop by 2C? Is the CPU suddenly 10-15C cooler? What about if we stick the machine outside on a cool night and the water is now 10C cooler. Is the CPU now 75C cooler? If not, why not? If it's a hyperbolic curve, then what's the gradient of this curve? When does it suddenly start to become a 2C water rise = a 15C die rise?

Why does my CPU still work with 50C water temps at 1466MHz/1.75v, when it fails at 1940MHz/2.10v when the water is at 25C? The thermistor is reading lower in the second scenario too.

In short Bill, AMD CPU's have a peak operational temperature range of ~95C. Just because it fails, doesn't mean that the CPU is at 95C. It is wrong to to make that inference.

But getting back to the point of the thread, I'm just reporting some water rise above ambient temps with some reproducible heat loads that people can easily use to make their own comparisons. Not everyone has Quake 2, etc.

[BillA]

no negation of the comparative value of your tests
good results, more people should do 'em

you are probably correct, in reviewing the thread I think I ran off the rails and caused my own wreck
(saw 125W and could not contain myself)

due to my total lack of direct experience using a CPU as a heat source (for those too oft stated reasons), and because of the extreme variability that the same program seems to yield between different users; I do lose sight of the user's desire to compare something, even different somethings
(after all, a hobby is just that, a hobby)

re the coolant temp / CPU temp 'ratio':
I do not (yet) have enough of the right kind of data to define it, but I do understand the general relationships
yes, as the heat load increases the differential increases; and I suspect that the slope at a particular point will be the measure of the performance differential between wbs
and the increase is exponential

the compression of the mobo sensor readings is relevant particularly because (IMO) it's output is skewed in the opposite direction (under reporting increases with temp)

the crux of my problem (but not yours) is that some readers (nay, damn near all) will interpret a 2^C temp diff in the coolant as being the same for the CPU temp

and this is not so
at very low heat loads the rise will be similar
at a 125 Watt (actual system input) the difference could indeed be 10^C, I would guesstimate more

BTW: you can determine the heat input if you know the accurate flow rate and the temp differential across the wb or the rad
- the system equilibrium temp rise is a different beast completely; involving the total mass of the system, all kinds of heat paths, etc

be cool

a question (off topic):
why use a stability test influenced by frequency to determine cooling effectiveness ?

[Cathar]

Here's a little side thing I just did. It's not accurate in any shape or form, but I guess we can speculate somewhat for fun. The values given below are very rough. This is not scientific in any way.

Have ~300ml of water in the closed system, but plus about 1kg of copper all up inlcuding the waterblock and radiator.

The CPU was at the "125W" heat load running Sandra. Turned the fan off and watched the water temps rise.

The water temps rose by 5C in 5 minutes.

Specific heat capacity of water => 4200 J/kgK
Specific heat capacity of copper => 385 J/kgK

All up, it would take ~1650J to raise the system by 1C.

5C in 5 minutes is 8250 J in 300 seconds, which is 27.5W of heat.

Now the radiator will also have been passively cooling the water a tad, plus there are other factors involved, but given a worst case scenario let's double it and say that the CPU is pushing 55W of heat into the water. Again horribly unscientific but the true value is probably somewhere between 28 and 55W.

I suppose this would somewhat correlate with what you're saying Bill. That the actual CPU thermal output (well - the bit that goes into the water) is much less than what Radiate would otherwise indicate.

Might also go a long way to explaining why overclockers report fantastical C/W figures with their systems when hard scientific tests (as performed by yourself) show that such C/W's are pure fantasy.

[BillA]

nice calcs Cathar

now you can understand my frustration

oooohhhhmmmmmm

be cool

[Cathar]

a question (off topic):
why use a stability test influenced by frequency to determine cooling effectiveness ?

I'm just trying to simulate the highest "heat load" that I can that should be equal or above what most other people will see. I guess my theory is this. Since I've got a thermal probe that's accurate to 0.1C and even then fluctuates over a range of about 0.3C, then in order to reduce the ratio of size of the measuring error (assuming it's a constant 0.3C which it may not be....) when dealing with temperature rises of 1-5C, I try to make the figures as large as possible.

If I ran the CPU at say "50W" and the water temps were 2C above ambient with a 0.3C variance, that's a pretty big proportional error. If the CPU is at "125W" and the water temps are at 4C above ambient, then the proportional size of the 0.3C error is much less.

That's basically where I'm coming from. Of course it could all be totally the wrong thing to do, but I'm targetting the overclocking crowd who could care less what the water temps are at stock CPU speeds/voltages.

[BillA]

"totally the wrong thing to do"
not being cruel here, but yes
gigo
or as I prefer: you can't make a silk purse from a sow's ear

following the same reasoning, have you considered using Btus and ^F ? (bigger numbers, no ?)

no way around it (that I have found), resolution costs money
and accuracy, being the product of resolution and calibration, costs a lot of money

back on topic,
the only 'problem' with small rads is that they have no bragging rights
as for 98% of users simply the change from air to water is sufficient

be cool