One for Martin.

[sirheck]

I did a little load testing of a single Swiftech 220 radiator.

SYS specs are.
Q66@3.6ghz. Actual voltage in BIOS is 1.4


8800gtx@Core 605mhz,shaders@1413mhz,memory@924mhz.
A mild overclock for this card.

A single ST 220 rad with a D-tek fuzion V-2 and X-38 NB cooler i put on
my X-48 rampage. Along with a ST mcw-60 for the 8800gtx.
A laing 3.1 pump i soldered to a 3.2 with a XSPC res. top and 3/8
tubing. In a coolermaster 690 case.

Ive had this set-up for about a year. Though its had anywhere from a single 220 to a single TC 320 to a 220 ST and 320 TC along with a TC 320 and
ST 220 together, but now i am using just a ST 220 rad only.

I know Martin (along with most here) likes detailed readings and measurements. So i did the best i could with what i have.

This is with Prime95 running for around 30 minutes, along with
F@H,s GPU client.



Ambients are,


You can see the Flukes temp probe just to the left of my hand.


The wire poking in the rad just above my memory sticks is a thermal sensor for my Scythe Kaze fan controller.


The fans up top are running at 1200 rpms according the Scythe fan controller.

Here is my Fluke meter measuring the output air from the top of the rad.


The Scythe is reading 35.7c, (within 2 tenths of a degree of what my
Fluke meter reports) the left reading below 1140 rpms








My question is, I see alot of people saying a 220 rad cant handle
a Core i7 CPU alone.
I just cant see how a Q66 kentsfield along with a NB and 8800gtx pumps out
more heat

[MpG]

It's not that a 220 can't handle it, but your fluid temps are warmer than what most would call "performance", and I'd go so far as to say that your CPU and GPU temps back that up. But yes, no reason that setup shouldn't be fully stable.

[sirheck]

I have rather high ambients compared to what most post for testing.
Most are in the low 20,s.=21 to 23 from most test ive seen.

Oh Snap, I just realized i didnt post my ambients.

[Vapor]

Isn't it amazing what a single 220 rad can do?

I was testing out my C0/C1 920 earlier today on an MCR120....4200/1.42v with HT in OCCT like it was nothing (held to like 72-73C with 21-22C ambient with a Fuzion V2)

"i7 needs at least _____" has been vastly overplayed

[sirheck]

Isn't it amazing what a single 220 rad can do?

I was testing out my C0/C1 920 earlier today on an MCR120....4200/1.42v with HT in OCCT like it was nothing (held to like 72-73C with 21-22C ambient with a Fuzion V2)

"i7 needs at least _____" has been vastly overplayed

You were the one i mentioned in another forum about supporting this theory.

Edit: Damn i also forgot that the thing to the left of the last pics is my 32 inch monitor flooding the side of
my case with heat, (It has 2 120mm intake fans).
Yes my compy is behind my monitor.

[MpG]

"i7 needs at least _____" has been vastly overplayed

Agreed. Hell, for the longest time, I was cooling both a Q66 @ 3.6/1.425 and a 8800GTS 640 with an old Koolance Exos-2. Fully stable with fans at ~1000rpm. And that's like a 220-lite.

Doesn't mean my three PA120.3 are going anywhere, though.

[Vapor]

More rad always = better Usually the easiest way to decrease temps at a given ambient...or you can decrease noise at the same time without sacrificing cooling (if your ears want you to).

But it doesn't mean a lonely little radiator can't do the job either

The heatdump of an i7 has been overstated, IMO (though for those coming from a Yorkfield, it is an increase in heatdump...but it's nothing we haven't already dealt with in the past few years). Going from an MCR120 to an MCR320+220 stack resulted in a temp drop comparable to (or even less than) the temp drop I had with my QX6700 at 3600/1.49V (going from MCR120 to MCR320+220).....which means the heatdump is "comparable to (or even less than)" that of my QX6700's

[jagass]

I agreed...

[Martinm210]

Awesome work!!! It's like feeding a kid candy, can I have more PLEASE...

FYI,
The only way to really know what heat you have is to measure it. I found that the PSU calculators are a bit overly conservative. My Q6600 for example should be cranking out over 200 watts at the overclock and Vcore I'm running, but I measure quite a bit less.

I'm not a big thermo guy, it was my only class in college I dropped out of..

But after digging around the net and playing around with some numbers I found what I need to calculate heat dump. You can fairly accurately measure this if you can log temperatures of a known volume of water over a period of time.

This is how you can do it, and how I did it just for fun to see if I could.

-Find a larger thermos or other container you can plumb some barbs into.
-Now create a loop for one component (or more if you want) without any sort of radiator.
-Now fill this loop with cold or cool water because you need some time to measure the water rise and it's only getting warmer and warmer without a heat exchanger.

This is what mine looked like:


Results and calculation:


This is the only way I know of that gets to the bottom of heat and takes into account all the heat losses through the MB, etc.

It would be interesting if several people would do this so we could get some sort of database put together. It's obviously outside the scope of one person testing considering the cost of processors and setups.

Anyhow, per the calculators my Q6600 should be consuming 200 watts, but I measure about 120. Maybe I screwed up the math somewhere, but it seems to make sense and at least in the ballpark. I guess I always assumed heat dump may be something like 80-85% of the power consumption, but it's alot more complex when processor efficiency is also influenced by how cool it's running..

Carry on with the goodness

[Hondacity]

the q6600+nb+8800gtx total heat..ummm less than 400w total, but close to around 370w, core i7 when overclocked hard with say 4.2ghz with ht on @ 1.55v will produce around 270w, this is very close to my results awhile back. also close to extreme.outervision.com's numbers

but here's the thing with the i7 it produces almost twice the heat of a q6600. its true we can use 2x120 rads...heck with a gpu in the loop our temperatures would still be better than air coolers aka heatsink fan combos.

anyways i did try the core i7 with 3 gpus and 4x120 with 4 fans, that was over 1000w worth of heat, but with the good rad my temps were very much away from the tjmax....and my video cards were much cooler than stock.

anyways thanks for the numbers, try the core i7, you'll like the heat in the winter

More rad always = better Usually the easiest way to decrease temps at a given ambient...or you can decrease noise at the same time without sacrificing cooling (if your ears want you to).

But it doesn't mean a lonely little radiator can't do the job either

The heatdump of an i7 has been overstated, IMO (though for those coming from a Yorkfield, it is an increase in heatdump...but it's nothing we haven't already dealt with in the past few years). Going from an MCR120 to an MCR320+220 stack resulted in a temp drop comparable to (or even less than) the temp drop I had with my QX6700 at 3600/1.49V (going from MCR120 to MCR320+220).....which means the heatdump is "comparable to (or even less than)" that of my QX6700's

i have my system idle,..then occt'd the cpu...just look at the power consumption...put in efficiency of the psu and you have a power hog or should i say a heater in winter



when running occt at 4.3ghz ht on at vcore 1.55

[sirheck]

Yeah i guess my advise (and probably most of you here as well) for some
one wanting to run/watercool an i7 system with gpu,s would be
something along the lines of......

i7 and single gpu = A triple or quad rad, single loop.
i7 and CF or SLI = dual loops with 2x220 rads at least.

I do have another (brand new ST 220) and plan on running
those new 4770,s in CF but have to figure out how i want
to add it to the inside of my CM 690 case

Gonna have to put it on the bottom via cutting the hdd cage
out or maybe stand it up in the front bottom of the case.
Either way im gonna have to do some modding.

[[XC] riptide]

<snip>

Carry on with the goodness

Just checking.... your graph is titled 1350ml of water? But you use 3500ml / grams in your calculations and displayed in the pic?

[Martinm210]

Just checking.... your graph is titled 1350ml of water? But you use 3500ml / grams in your calculations and displayed in the pic?

Good catch!

Ignore the title, I'm the thermose is a 3.8L unit, and by the looks of it and what little memory I have these days, I'd have to guess it was 3500. I think I just copied that graph from an earlier trial where I used less volume. If I remember right that smaller volume just heated up too fast, and that's what prompted me to buy this larger thermos for the test. The longer the test and larger the volume of water, the more accurate it should be.

[Bond Number]

Martin,
I have question concerning your heat dump test versus an in situ test, where one measures temperatures before and after the water block and the flow rate with an inline meter.

Are you worried about the accuracy of the flow meter? I see this and the performance decrease of an inline flow meter as the biggest drawbacks. However one would be able to log heat dump as a function of cpu load during their daily use. This might be useful data for people who do not run their cpu at full load. Also one could look at how heat dump varies with temperature, i.e. for the same operation what is the efficiency of the processor (how much current leakage) at different temperatures.

Also in your current setup what is the flow pattern? If you let the hot water enter the bottom of the thermos it would help to eliminate any temperature gradients. But from you picture it looks like the flow is pretty turbulent so the gradient should be small.

[silverphoenix]

Awesome work!!! It's like feeding a kid candy, can I have more PLEASE...

FYI,
The only way to really know what heat you have is to measure it. I found that the PSU calculators are a bit overly conservative. My Q6600 for example should be cranking out over 200 watts at the overclock and Vcore I'm running, but I measure quite a bit less.

I'm not a big thermo guy, it was my only class in college I dropped out of..

But after digging around the net and playing around with some numbers I found what I need to calculate heat dump. You can fairly accurately measure this if you can log temperatures of a known volume of water over a period of time.

This is how you can do it, and how I did it just for fun to see if I could.

-Find a larger thermos or other container you can plumb some barbs into.
-Now create a loop for one component (or more if you want) without any sort of radiator.
-Now fill this loop with cold or cool water because you need some time to measure the water rise and it's only getting warmer and warmer without a heat exchanger.

This is what mine looked like:


This is the only way I know of that gets to the bottom of heat and takes into account all the heat losses through the MB, etc.

It would be interesting if several people would do this so we could get some sort of database put together. It's obviously outside the scope of one person testing considering the cost of processors and setups.

Anyhow, per the calculators my Q6600 should be consuming 200 watts, but I measure about 120. Maybe I screwed up the math somewhere, but it seems to make sense and at least in the ballpark. I guess I always assumed heat dump may be something like 80-85% of the power consumption, but it's alot more complex when processor efficiency is also influenced by how cool it's running..

Carry on with the goodness

looks like a homemade Calorimeter, nice!!

[Martinm210]

Martin,
I have question concerning your heat dump test versus an in situ test, where one measures temperatures before and after the water block and the flow rate with an inline meter.

Are you worried about the accuracy of the flow meter? I see this and the performance decrease of an inline flow meter as the biggest drawbacks. However one would be able to log heat dump as a function of cpu load during their daily use. This might be useful data for people who do not run their cpu at full load. Also one could look at how heat dump varies with temperature, i.e. for the same operation what is the efficiency of the processor (how much current leakage) at different temperatures.

Also in your current setup what is the flow pattern? If you let the hot water enter the bottom of the thermos it would help to eliminate any temperature gradients. But from you picture it looks like the flow is pretty turbulent so the gradient should be small.

You theoretically could measure the heat in watts by measuring flow rate and temperature delta, but the problem quickly becomes the resolution of the tools. To do that you would really need inline temperature probes with a resolution and relative accuracy down to .001C which takes some high dollar lab grade tools. The crystalfontz is only good to about .2C depending on the temp range, so it's not good enough. You may be able to get better than that using multiple probes and averaging logged data, but I'm still not sure it would be good enough.

I think it requires something like 350 watts if I remember right to make a 1.5GPM flow rate change 1 degree, so most of the time the temperature differential is nearly as much as the error in the probes. You might be able to improve the temperature resolution by adding restriction and slowing flow rates way down, but then you'd need a higher resolution flow meter.

The $3 thermos is pretty cheap in comparison and more accurate than what I have on hand. It's too bad though because measuring heat dump or loss by having some really high resolution probes would be the ideal way to measure both heat dump as well as measure radiator performance because it's really cut's down the time to test and eliminates any error of other heat losses that are outside those two points of temperature measurement.

The coolingmasters guys over in France did just that when they did their radiator testing, so I know the tools are out there. You just need temp probes with a relative accuracy of .001C to do it, which they did.

It doesn't matter what the flow pattern is. At 1.5+ GPM the flow is so turbulent and so highly recycled the temperature is virtually constant throughout the loop with the exception of the very slight .1C jump across the block and pump. You can see that in the logged chart. The blue line is actually data points that were collected by the crystalfontz on 1 second intervals across the entire test, so temperature is very much a natural slope. The curvature is error from the system losses, that's why I averaged it with a straight line trendline to account for it.

[Bond Number]

Thanks

Me and friend were thinking of doing so experiemnting over the next couple of months, so thanks for the insight.

[Russ_64]

Btw, my results with a Phenom II 940 OC to 3.6 and 2 ATI cards in single loop with a dual rad are about the same - it is too much heat for one loop (ambient is around 21c and 24/7 loop is approx 55 - 60c).

[relttem]

Martin,
I have question concerning your heat dump test versus an in situ test, where one measures temperatures before and after the water block and the flow rate with an inline meter.

Are you worried about the accuracy of the flow meter? I see this and the performance decrease of an inline flow meter as the biggest drawbacks. However one would be able to log heat dump as a function of cpu load during their daily use. This might be useful data for people who do not run their cpu at full load. Also one could look at how heat dump varies with temperature, i.e. for the same operation what is the efficiency of the processor (how much current leakage) at different temperatures.

Also in your current setup what is the flow pattern? If you let the hot water enter the bottom of the thermos it would help to eliminate any temperature gradients. But from you picture it looks like the flow is pretty turbulent so the gradient should be small.

You really need to calibrate everything if you want to get accurate results. I have seen thermocouple (high end) that are off by a degree. Thermocouple accuracy is usually a function of the temperature you are reading and the temperatures that the thermocouple is made for. There are all types of thermocouples - some for liquids, some for air..etc etc. We always calibrated our flow meters using a gram scale. We would simply pump the fluid out of a large container into a smaller container that was on a gram scale for a specified amt of time. We would do that a few times and take an average. We use the equation:

Q = mdot Cp dT to calculate how much heat our water is pulling from the chip.

knowing the mdot you can calculate the Reynolds number:

Re = 4*mdot/(mu*D)

mu is the viscosity of water, which you can look up online
D is the diameter. This gets tricky when flowing through a water block - you have to compute a hydraulic diameter, which is going to be different for every block. But, the Reynolds number will let you know if the flow is turbulent - more or less if Reynolds > 4000. But, if the Re is below 4000 it can be made turbulent which would be the case in water blocks.
Turbulence increase the convective coefficient, which is important in heat transfer involving a moving liquid (air, water, etc), but the trade off is higher pressure drop, which means you need a stronger pump.

[sirheck]

I was tired of the 120.3 or my other 220 rad hanging off the back of my case
so i added the 220 to the front of my CM690 case.
I even cut and ground the front out using my air powered die grinder and
cut off wheel with everything still inside the case

It still works just fine.

[MomijiTMO]

This thread has made the logic part of my brain wake up.

Thanks to everyone who posted and I'mma scrap that upgrade plan [adding 2 360 rads]

[Irocing]

Hi,

Just came across this thread.
Advanced 11 but 690.

360 Split

Pic

Later

[MadHacker]

I ave my i7-930, Ram , motherboard & video card (5770) all water cooled running through 2 140x3 rads, yate loom fans running at 1100 RPM
my CPU is overclocked to 4.3GHz, and when i run Linx 20 passes by the end i can feel the water is getting rather warm...
guessing 35-40 deg, no thermometer on it so i can't say.
I'm not worried about it because i normally don't run my machine that intensive...
just thought I'd give a heads up of my findings as unscientific as they are.

[Elpy]

So I guess neither of you bothered to notice that this thread is from 2009 and what on earth do your posts have to do with this thread? O_o. Afaik this thread was about using only a 240mm rad for i7 and neither of you even have just a 240mm.

[MadHacker]

sorry didn't notice the date...