Thinking of this TEC setup...

[BrianH2O]

Current system

Two loops with 1 res where the 2 loops mix.
1st loop
Res --> ViaAqua 1300 #1 --> BIX #1 --> Maze 3 --> res
2nd Loop
Res --> ViaAqua 1300 #2 --> BIX #2 --> DD GF4 block --> res
Each BIX has Panaflo 120mm Med blowing with no shroud

Current temps with 22C ambient are
Water 27
CPU 2.66@3.1@1.625V -->28 idle 32 load (F@H)
Radeon 9700pro -->28 Idle 31 load

I was thinking of a 226w for the CPU powered by a Meanwell S320
And an 80w TEC for the GPU powered by my 520 Vantec Steath PSU.
Add an 1/2" spacer and shroud to the BIX's

Will this work? What temps should/can I expect? Are the 2 BIX's enough to cool all the heat generated?

I expect the total heat will be around
CPU --> 100w(?) CPU + 20a*12v = 340w
GPU --> 40w(?) GPU + 8a * 12v = 136w
Total around 476 watts. Is this an accuarate figure?

My goals are to OC the cpu a bit more to around 3.2/3.3 and get better OC on the 9700Pro with a voltmod.

Thanks

[Warlord2]

looks good to me, that rads should work just fine, Ive been seeing a lot of 0-15c temps with the 226watt pelt, I donno about the gpu though, I thought I read they are starting to put out quite a bit of watts now so I donno what your temps will be on the gfx card with a 80watt pelt

[Kato]

I'm running my 80w on the Vantec 520W, and the GPU temp is around 20-23C with default voltages(1.66V), 25-27C when the card is Vmodded to 1.92V. I'm thinking of switching the 80w in my mcw50-t to a 126w, to bring the gpu temp below 20C...
But since you got 1 BIX for each tecs, I would say your gpu should be colder than mine.

[BrianH2O]

Thanks Warload2 and Kato

Kato
Some one in another forum suggested that I use a 172w @ 12v on the radeon 9700pro. I user the Kryotherm program and a TEC Sheetsheet to verify that it would provide lower temps and use less power also.

Are you using the 9700pro also? If so, have you removed the shim? If not, it would make a hugh difference in you temps I believe. My WC temps dropped 4C idle and 6C load. When I took off my heatsink, there was barely an impression in the TIM. What kind of OC are you getting with the Voltmod?

Warload2
I believe that the new gen of cards are putting up to 40w of heat.
I will be putting a 1/2 spacer and a shroud on the rad also, should help my cooling a little..

[Kato]

Actually, I'm using a Ti4200:D
Without Vmodd, I hit 330 on the gpu, with stock cooling...
I've tried 1.92V on the gpu, and it seems to be stable at 360, but I need to test more...
So you are saying that if I power a 172w from a 12V source, it would still be enough to cool it?

[BrianH2O]

Kato

Yes, if you use the Kyrotherm software (search around for it, easy to find), here are my results

80w @ 12v w/ 40w heat source /20C abient
Amps----------------> 6.94
Heat output(w) --> 131.34w
Cold side temp ---> 4C

172w @ 12v w/ 40w heat source /20C abient
Amp ------------------> 5.68
Heat Output (w) --> 119.10w
Cold Side temp ----> -1.9c

Therefore, less power, less heat, colder cold side, all sounds must better to me.

I think TEC's may be fun to play with.

[Paxmax]

Hi Brian H20!
If I were you I'd let that 226W TEC also benefit from some cooling from the other BIX aswell. Cause one BIX is no match for the mean 226Watter TEC. I think you'll be dissapointed with the cpu temp if you run it on just one rad.
I for one cannot understand why ppl buy BIX, when heatercores are so damn cheap!!.. okay I got to admit... they don't look as good as BIX.. oh, maybe thats why?? :o

[BrianH2O]

I've been think some more on this subject....

Since both loops share the same res, both will be helping cool the hold system. Since rads work best with a higher temp different between it and the air, both rads would be working at "max" efficiency.

If I did a series flow (ie from one BIX to the other), the temp diff would drop and make the second rad NOT as efficient as they could be.

Another suggestion from another forum, both both rads on the GPU loop, but run them in parallel. The loops would be as

Loop1
Res --> Pump --> Maze3 --> Res
Loop2
Res --> Pump --> Splitter --> Bix1 & BIX2 in parallel--> Jointer --> DD GF4 --> res

Here was the thinking... For the first loop, the water flowing thru the very hot Maze3 would be very fast. This would allow the water to get in/out without it getting very hot

For the second loop, the split would would slow down the water while it passed thru the rads giving a chance for rads to really work. Then join up again. Yes, this would reduce the flow abit, but the GPU is just not that hot.

I had another idea... use 2 x 172w @ 15v attached to the Maze3. This would give me better cooling with less power being used.

Any thoughts/ insights on this line of thoughts?

Oh and Paxmax, Yes i purchased the BIX's because I needed a self contained system and, at the time, seemed to be the best fit. I think both of them together will be able to cool the system to my liking, ie 10 to 15C for the CPU and the GPU... maybe a bit colder.

[Warlord2]

I had another idea... use 2 x 172w @ 15v attached to the Maze3. This would give me better cooling with less power being used.

well I think colin did this and the results basically sucked, he couldnt take away enough watts even with his big bong cooler. What I think would be pretty cool would be mulitible small pelts, maybe 2 or 3 120watters, this I think will bring the deltaT down and you wouldnt have as much heat to get rid of.

[I.M.O.G.]

hey guys, mind if i join you? ;)

Here was the thinking... For the first loop, the water flowing thru the very hot Maze3 would be very fast. This would allow the water to get in/out without it getting very hot

For the second loop, the split would would slow down the water while it passed thru the rads giving a chance for rads to really work. Then join up again. Yes, this would reduce the flow abit, but the GPU is just not that hot.


i have found this line of thought in other threads before also. it makes sense in one way and i understand the thinking, but i think you will understand where i explain this theory goes wrong.

Slowing down the water as it goes through the rads will not make the rads work harder/better. there was an excellent point about setting the res and rads up earlier which i will quote:

Since both loops share the same res, both will be helping cool the hold [whole] system. Since rads work best with a higher temp different between it and the air, both rads would be working at "max" efficiency.

both rads will help cool the entire system and zero efficiency will be lost in splitting the line - havent thought of splitting it in the resevoir before, i like that. and it is right to say the hotter the rads are, the more heat transfered because of higher temp difference between rad and air.

So this is what i want to bring attention to. it is most desirable to have the hot water flowing through the rads at the fastest pace possible. the faster the warm water flows through, the faster the rads will get hotter and the better heat transfer there will be (higher flow rates means more heat being added to radiator). slower water going through the rad does not equal better cooling, it equals heat being added to the radiator at a slower pace - a result that is clearly not wanted. im sure you see what i mean. good luck with your setup! :)

[I.M.O.G.]

i could have added this to the last post but it was long enough already and i just thought of this... if someone was having a lot of trouble removing enough heat due to large pelts or what not, and perhaps they were considering going with parallel/series radiators, it would be a much better solution (at the cost of an additional pump) to add a loop going from res->pump->rad->res. create a dedicated heat exchange loop and i am positive the results would improve. the flow would be maximized in this loop and the water would be getting cooled more efficiently.

[BrianH2O]

Thanks I.M.O.G

Then I should leave both loops with thier own BIX? This will allow the water to be max cooled just before entering the water block.

Currently, my temps are as follows @ 22C ambient
CPU...28c idle, 32c load (F@H) from MBM
GPU...28c idle, 33c load (3dmark) as measured on backside

This is without a shroud on the BIX's. For normal water cooling, I do not think a shroud on the BIX's make any big difference. The total heat being put out is max 140w.

With the peltier setup I am suggesting, I will be cooling at least 420w of heat. Then a 1/2" spacer with a shroud will be required.

So in summary, here are the mod's I am thinking of making
1) 1/2" spacer with a shroud on the RAD's
2) Modify my DD GF4 block with upper and lower 50x50mm 1/4" copper cold plates with a 172w peltier powered @ 12v from my PSU
3) Get the DD Maze3 cold plate with a 226v peltier powered by a dedicated 12v PSU.

When using the Kryotherm program, what values do people usual use for the Rc. Rh, Rins parameters. I have been using .05, .05, and 2.0. Are these OK?

How does I estimate the heat output of the CPU? My 2.66 at stock puts out around 66w. For the Kryotherm, running at 3.1 w/ 1.625v, I estimated 100w. Is that too high?

[Since87]

Originally posted by BrianH2O
When using the Kryotherm program, what values do people usual use for the Rc. Rh, Rins parameters. I have been using .05, .05, and 2.0. Are these OK?

How does I estimate the heat output of the CPU? My 2.66 at stock puts out around 66w. For the Kryotherm, running at 3.1 w/ 1.625v, I estimated 100w. Is that too high?

I typically use:

Rc = 0.1
Rh = 0.18
Rins = 1 for CPU.

Rc seems to be about right.

Rh is very dependent on WB, rad, and pump, and has to be guesswork for most people.

Rins is mostly a guess, and will be substantially lower for a GPU because the GPU is soldered directly to the PCB. For a GPU I'd probably guess Rins=0.7.

For CPU heat, there are various calculators out there that attempt to calculate this, but their accuracy is very questionable.

I don't often use the Kryotherm calculator for trying to get exact temperatures. I usually use it to compare two particular operating conditions with all else being equal.

[BrianH2O]

Sorry, these replies are getting longer, again sorry, just need to know before I spend all this money...

Those numbers are not very encouraging, Since87

Using a 172w TEC @ 12v
Ambient = 20c
Volts = 12v
Object Heat = 40w (oc'ed Radeon 9700pro)
Rc = 0.1
Rh = 0.18
Rins = .7 for GPU

I get
Amp = 5.54
W=66.48w
Qc = 50.28w
Qh= 116.75

and Tob= 12.8c

12.8C on the cold side is way too high to consider I think. How accurate do you think that is. With an 80w, same parms I get
Amp = 6.71
W=80.58w
Qc = 45.6w
Qh= 126.54
Tob = 15.8

I pretty sure that people are getting less that this with an 80w peltier on thier GPU. As a matter of fact, Swiftech
GPU Block (http://www.swiftnets.com/products/mcw50-T.asp#), shows on thier graph a temp of 6c for 40w GPU and 80w peltier. In order to get Tob of 5.9 I needed the following parms:

Ambient = 25c (as per swifttech graph)
Volts = 12v
Object Heat = 40w (oc'ed Radeon 9700pro)
Rc = 0.05
Rh = 0.05
Rins = 3

I get
Amp = 6.85
W=82.14w
Qc = 46.36w
Qh= 128.50w

Like I said, confused....

[Silver]

Originally posted by Warlord2


well I think colin did this and the results basically sucked, he couldnt take away enough watts even with his big bong cooler. What I think would be pretty cool would be mulitible small pelts, maybe 2 or 3 120watters, this I think will bring the deltaT down and you wouldnt have as much heat to get rid of.

Dual 156w tecs on dual bongs using two small water blocks one pump and one side of bong to just recirculate. Works for me. Take Care all.:)

[Since87]

I'm not sure I understand the MCW50-T.

According to Swifttech's ad, "Module hot side is directly submersed into the coolant for optimal heat dissipation."

I don't see how they are doing this from the pictures though.

If they are putting the water directly onto the hotside of the pelt though, you would get a very low Rh.

My 0.18 number may be too conservative. I haven't found much reliable data to base a better number on though.

The C/W of most cooling systems is likely to be better mounted on a pelt than mounted on a CPU, because of the increased surface area. I did an analysis of Swifttech's aircooled pelt product. Using JoeC's die simulator results for the MCX-462 heatsink on Die and on 50mm Pelt, I was able to determine that the heatsink's C/W dropped by about 15% when mounted on a pelt vs mounted on a CPU. The MCX-462 is a very thick based heatsink though, and you are likely to get much larger C/W drops with thinner based heatsinks/waterblocks.

To paraphrase what I said earlier - without accurate numbers to use with it, the Kryotherm calculator is most useful in comparing one simulated system to another. Until we have more accurate numbers from real world systems, we won't be able to calculate the performance of real world systems with much accuracy.

[BrianH2O]

OK I get it .... No one can accurately predict what my temps will be like. I will just have to build it and see what happens (ie pray!)

Actually, I am looking at those swifttech blocks with the built in TEC's. Might be the best solution. They are a pretty reliable company, from what I've read.

[Les56]

Originally posted by BrianH2O



I needed the following parms:


Rc = 0.05
Rh = 0.05
Rins = 3



Possibly ok numbers for cooling a 40x40mm heat source.
would suggest maybe a lower value of ~ 0.01for Rc. This being a fair guess at the resistance of a 40x40mm heat channel through (3mm Cu coldpate +good TIM). .
An Rh=0.05 for a Swiftech would be in roughish agreement with the predicted cooling from a SWiftech 462 with a 40x40mm heatsource (Extension of sums shown here http://forums.procooling.com/vbb/showthread.php?s=&threadid=4960 .
Have not put any thought into a guess for Ri when cooling a GPU.

Would suggest a much higher Rc for cooling a CPU. My best guess would be ~ 0.19C/W to 0.26C/W *

* From here : http://forum.oc-forums.com/vb/showthread.php?s=&threadid=150430 "..my currently favored Kryotherm parameters may interest :-
Rc=0.19 - 0.26 (for 40x40x6 mm Cold plate with a 10x10mm heatsource)
Based on:
1) C/W(TIM,Die/coldplate) ~0.1 - from Bill Adams work(e.g http://www.overclockers.com/articles654/index02.asp ) ,making an allowance of 0.05c/w for die thermocouple offset.
2) C/W(Coldplate) ~ 0.09 to ~ 0.16 -- from Waterloo calculations using,possibly mistakenly, an h=10,000 - 100,000W/m*m*c as corresponding to the conditions for a 172W Pelt cooled by a wb with a C/W=0.00625 - 0.0625(cooling a 40x40mm heat source).

Ri=1
Based on my current attemps to cross-correlate Bill Adams data with CPU(Diode)/"Radiate calculated Watts""

[BrianH2O]

Would the Ambient Temp for any calculation be the water temp? I have been using the air temp of 22c. But the air temp is meanless when water cooling the peltier. Therefore a more meanful value of 27~35, depending on your system, would provide more of a "real" world figures.

Les
Possibly ok numbers for cooling a 40x40mm heat source.
So, using
80w TEC@12V with
Wob=40w
Ta=30c
Rh=.05
Rc=.01
Ri=1 ........I get 15c as the cold plate temp?
Ri=2 ........I get 9.7c as the cold plate temp?
Ri=3 ........I get 7.5c as the cold plate temp?

Maybe the default of 2 is better? I do not pretend to understand the math behind it all, but "real" would experences have shown better temps than 15c, I believe?

[Les56]

BrianH2O

Yes the ambient for calculations is the water temperature.
Agree with your analysis.
Maybe Ri=2 is better or maybe a 30w estimate of the load is appropiate.
Possible the best one can hope for is a feel for the influence of the parameters

[Since87]

Glad to see you posting here Les.

I would think Rins for a GPU would be substantially lower than for a CPU

A GPU has a very short heatpath to the PCB compared to a CPU where the heat has to travel through the pins and the socket. Of course there is the the issue of other heat generating parts near the one to be cooled; voltage regulators and video ram. These external heat sources mean the Rins number has to be a crude fudge factor because there are temperatures substantially higher than ambient involved.

[Paxmax]

Um, yes, basicly, the Rins would be lower for a GPU, but cooling it under ambient temp with TECs the Rins(through pins->traces->board) would counteract the cooling.