Question about Water Cooling Loop

[fx9]

I'm trying to jump on the water cooling bandwagon. I've decided to get a Swiftech 220 APEX Ultra+ water cooling kit but I'm not exactly sure how to setup the cooling loop. I might have an idea though, is this about right?

2x120mm Radiator -> Cpu block -> Northbridge Block -> GPU block -> Reservoir -> Pump -> Back to Radiator

or is it pump first then reservoir then radiator?

 

[fx9]

Anyone?

 

[THRESHIN]

that setup will work just fine. some may have a different way, but your method will work just fine. in fact i think its best to send the water to the rad after the pump so that any heat absorbed in the pump is removed. also the res before the pump is good too - if theres any air bubbles in the system (like when you start up for the first time) there won't be any issues with too much air in the pump which could cause fun stuff like cavatation

 

[Arcygenical]

Loop order doesn't matter at all... So, don't worry about that.

I'm worried that your double radiator won't cope with a CPU, NB, and GPU... Infact, I know it won't. Get at least a triple radiator.

 

[Headpunter]

Res then pump, and 120.3 radiator....

 

[nsx241]

Second res then pump. The instructions even recommend this order so you'll always have plenty of water feeding the pump (never let the pump run dry).

 

[qwakrz]

Loop order doesn't matter at all... So, don't worry about that.

I'm worried that your double radiator won't cope with a CPU, NB, and GPU... Infact, I know it won't. Get at least a triple radiator.

The dual Rad should be fine.

I run that dual rad on my highly overclocked (for the stepping) quad core Q6600 (pushing over 200W at full load) and my overclocked 8800GTS 512 without problems but I do have to turn the fans up from the lowest setting or things get warm at full load , a dual core chip and GPU should be fine.

As for the loop arrangment, generally, Res > Pump > Rad is the order so the pump pulls from the res and the rad cools the heat from the pump. As for the NB and GPU fit them in whichever order is easier to pipe.

 

[Arcygenical]

The dual Rad should be fine.

I run that dual rad on my highly overclocked (for the stepping) quad core Q6600 (pushing over 200W at full load) and my overclocked 8800GTS 512 without problems but I do have to turn the fans up from the lowest setting or things get warm at full load , a dual core chip and GPU should be fine.

Can I get a picture of your temps? With that sort of heat dump, I'd expect a coolant temp of about 13-15c over ambient, which would mean a CPU temp of about 35c over ambient.

Loop order doesn't matter at all though... 18w or so from a pump is nowhere near hot enough to increase the coolant temperature.

 

[fx9]

The dual Rad should be fine.

I run that dual rad on my highly overclocked (for the stepping) quad core Q6600 (pushing over 200W at full load) and my overclocked 8800GTS 512 without problems but I do have to turn the fans up from the lowest setting or things get warm at full load , a dual core chip and GPU should be fine.

As for the loop arrangment, generally, Res > Pump > Rad is the order so the pump pulls from the res and the rad cools the heat from the pump. As for the NB and GPU fit them in whichever order is easier to pipe.

Do you have a picture of your rig? What are your temps while running prime95 or orthos? A dual rad may be cutting it a little close. 2x120 rad approximately cools 300W, an overclocked q6600 and a 9800GX2 comes to around 280W adding the northbridge to the loop will surely bring it over 300W.

 

[qwakrz]

I dont have any pictures or screen shots of temp as the PC is currently being ripped appart to change from 1/2" that came in the Swifttech H2O Ultra kit to 10/8mm and a full face G92 heatsink to replace the GPU only block and I am re-mounting the radiator (may also add a new one for the summer). The old setup was with the rad placed on its edge ontop of the case without any mounting screws and the side of the case left off as the pipes were to big to go through the back of the case. The new setup will be mounted ontop of the case with a proper mounting kit and the pipes run through some holes in the top of the case with the rad on standoffs from the top.

Temps were in the high but safe range, under max load (Prime on all 4 cores and max CPU speed, I usually throttle it down when just browsing web or playing simple games) Realtemp was in the range of 50C at its peak at max fan speed, 63C with them on low speed and all with a room temp of 25-30C (its been hot the last few days). The return water was cool but the water from the GPU (after the its been through CPU and GPU) was noticably hot even just holding the pipe.

foxtrotniner, bearing in mind that you are running stock you should see temps about as good as the best air cooling. If you are building the kit yourself (and if you have room) I would spec a 3x120 rad, if its a kit and you cant change anything you should be fine with the 2x120. Overclocking to the extream may push it upto the temps I see though.

I will post pics and temps of the new setup when its working (this weekend).

 

[devilchrist]

I have the same parts as that kit.. i bought it in pieces. I am now kicking myself for not going with 3x120rad. I'll have to add another 1x120 rad to my case.

my E6750 at 3.8ghz 1.6v and 8800gt at 750/1800/100. my temps max out at 63/63 45. with ambient room temp being 28.

too high for my liking and with the 3x120 rad I believe I can get much better temps.

If you can get the 3x120rad. for extreme overclocking, if you're just going for minor to safe overclocking.. then 2x120 is fine

 

[qwakrz]

Ok, put my loop back together just to take these screen shots (its cobbled together on a desk so not the best of setups)...

-edit to make this clearer-

This is the system just switched on and stone cold. It is supposed to show you the ambiant air temp (give or take a bit)

-end edit-

Idle temps after switch on (not even leak tested it), gives you a reading for ambiant or as close as I could get it

Short run of Prime at max fan speed. Look at the peak temps as that is where they settled. There is no loading on my GCard currently so its not making much difference to the temps

Continuing run but with low speed fans (Zalman fanmate on slowest speed). Still under control but this is in a cold room, colder than the last few days.

Just stopped the Prime run, shows the coolent temp in the loop currently which is only about 8C above ambiant.

This is not a perfect setup, not by a long shot and the CPU is only clocked at 3.2GHz not the 3.4GHz it can reach at 1.54v. The loop is running on pure water as I didn't want to fill it with additive just for this short run and the heatsink compound is fresh but it shows how well the Swiftech 2x120 can handle high-ish loads.

I will do proper runs when the new kit arrives for you.

 

[Arcygenical]

Those temperatures don't look right to me at all... They defy the laws of physics

Starting idle temps of about 20c, and final load temps of around 48 for 1.45v...

Realtemp is detecting your TJmax as 85c... Which I believe should be 100c... Which means your temperatures are 15c too low.

 

[fx9]

I would not use realtemp to measure a q6600, I would use it for my e8400. I suggest you run Coretemp, your Tjmax should be around 100C for a q6600.

 

[qwakrz]

I have gone through and checked CoreTemp and RealTemp to my actual running temps of the chip. You can setup extream idle on the CPU (lowest multi and just enough Vcore to boot) so that you can use a temp probe to measure coolent temp and match it to the core temp (+-1C) CoreTemp is always 15C to hot for this test on my B3 stepping chip, RealTemp is within 2C of measured value using in-water temp probe from Fluke meter in the resovour before the pump and radiator. The 15C can be accounted for by CoreTemp using a TjMax that is 15C higher than RealTemp. One thing that I find weird about the B3 stepping of the Q6600 is that it is just 2x B2 stepping duo's so it cant have a 100C TjMax the G0 stepping is the higher temp stepping as it was a re-designed core for the duo's and quads with the higher TjMax.

Anyway........

The lowest temp in the first screengrab is literally me turning on the PC and calling up the core temp, that is as near as possible ambiant temp as nothing is up to temperature. That is not system idle temps which usually sit at about 28-30C, it is just to show what the current air temp is in my room (give or take a degree) and how cold the water loop is at the start.

Highest peak temp was 51C for the loaded test at slow fan speed. That was in an ambiant of 20C

As I said, they were just quick tests (no heat phase lasted more than 10 mins, most no more than 5) thrown together for the dual 120mm rad with the supplied fans from a swiftech H2O Ultra+ kit.

 

[Arcygenical]

What does Intel TAT tell you.

This is the tool I trust above all others... Since it was Made by Intel. No inference needed on how to discern TJmax.

 

[nsx241]

First, update you real temp to the latest version. Second, real temp has always set tjmax for q0 q6600 to 95C since I was using it months ago (was definitely through ver. 2.41) since this was what unclewebb measured manually and most people tend to agree with, so I don't know why it's setting 85C here (unless of course you manually changed it).

 

[qwakrz]

First, update you real temp to the latest version. Second, real temp has always set tjmax for q0 q6600 to 95C since I was using it months ago (was definitely through ver. 2.41) since this was what unclewebb measured manually and most people tend to agree with, so I don't know why it's setting 85C here (unless of course you manually changed it).

Its showing 85C because its a B3 not a Q0

TAT shows 100C and is 15C higher than the true temp reading of the water loop, CoreTemp followed TAT for its TjMax and both are wrong according to my data from my chip (Intel have gone on record to say that TjMax is not correct in any document published and varies from chip to chip, the only thing we can say with any certenty is that the chip will throttle when temp to TjMax = 0 but no-one knows what the true TjMax for any chip is so they can work out an accurate running temp figure and Intel are not saying so this is all educated guess work).

 

[Arcygenical]

Its showing 85C because its a B3 not a Q0

I've had lots of B3 chips... 7 G0's and 5 B3's and all the B3's had a TJmax of 100c or 95c... The only reason we question this, is that your temperatures are much lower than the expected values for your heat load and radiative capacity... Yes, Intel has gone on record to say the TJmax is different for every chip... But it's not likely that the entire enthusiast community is wrong.

A heat dump of about 200w on the CPU (and probably 50w on the GPU at idle, 150w during load) will lead to higher temperatures with a waterblock...

 

[qwakrz]

I've had lots of B3 chips... 7 G0's and 5 B3's and all the B3's had a TJmax of 100c or 95c... The only reason we question this, is that your temperatures are much lower than the expected values for your heat load and radiative capacity... Yes, Intel has gone on record to say the TJmax is different for every chip... But it's not likely that the entire enthusiast community is wrong.

A heat dump of about 200w on the CPU (and probably 50w on the GPU at idle, 150w during load) will lead to higher temperatures with a waterblock...

Did you double check that each chip did actually have a TjMax of 100C by measuring the real temperature of the heatsink or did you just trust that the temp reporting program was using the corect TjMax value (remember all temps are worked out from the distance left from TjMax)? If everyone assumed that CoreTemp or TAT had chosen (as the TjMax is chosen and not worked out by these programs) the correct value then all the temps will be wrong, which is exactly what I have found to be the case with my chip.

As I have stated, I have double checked the CPU temp in relation to water temp using the ultra low load test (see my 2 posts above on it) so that core temp is within 1C of the chip case temp / heatsink temp and this temp is then passed into the water loop, Using 100C TjMax the temps reported by TAT and CoreTemp are 15C to high compaired to my calibrated thermomiter. So, its either a bit of software is reporting incorrectly or my certified thermomiter is reading incorrectly, I know which one I would trust. Remember that this test is done at idle with the lowest power from the CPU possible, if there is any load on the CPU the core temp will be further from the heatsink temp and readings will be wrong.

As for my pictures, They are only an example as I would not have expected the loop to have reach the upper max values but it was close enough for a 5 minute test. The room temp was cool (about 18C) and the fans do not suck in any case air or other pre-heated air so getting low temps is possible. All I was out to prove is you can get good temps using a 2x120 setup and higher speed fans (not the super 110cfm I now have but still not silent). I will agree that you would be better to go with 3x120 for quieter cooling using slow speed fans.

 

[Arcygenical]

When I was cherry picking my particular chip (it hits 3150mhz at 1.025v, which astonishes me, really) I did comprehensive thermal tests for the 3 best performing chips (as I said, I had ordered 15). I checked the temperature of the waterblock with an IR thermometer which measures boiling water at exactly 99.6c (hard to get a good reading due to the movement of the water, so I eventually aimed it at a penny) with an estimated heat load of 27w (0.025*(0.95V^2)*(6x*200mhz)).

My temperature reading, for the Tjunction, was about 7c hotter than the base of the waterblock, and 8c hotter than a copper heat plate that I was using to get the temperature from the exact middle of the chip (a 3.5mm thick copper slab mounted at exactly 22lbs with ceramique paste).

Point of the matter is, unless your chip's IHS is a perfect conductor of heat (it's not) and your paste has absolutely no thermal resistance (It won't) your Core temp will always be hotter than your IHS/WB temperature... Even when the load is very low (around 27w) we can estimate how much hotter the slab of silicon is going to be using an equation (rearrange Heat = SHC * Mass * Temperature Change)

Since the SHC of silicon is 700J/kg... and I'm estimating the die of a CPU is about 2g of silicon (could be wrong though, might be more like 3g) and 27w = 27J/s the temperature of the wafer of silicon will rise 19c from the heat generated alone... Now, of course, the thermal resistance of the IHS comes into play (I will never know this figure, so I can only speculate) which might reduce the thermal transfer by at least 50% (acceptable) which is a temperature difference of 8c between the IHS and the die... Exactly what coretemp (and I) measured.

I did this for my 3 best chips (I eventually settled on my chip, even though it wouldn't hit 4ghz reliably, it's VID was so low, that I couldn't pass it up) and came up with *about* the same answer...

So, yeah, I've done the testing... And even then, your data (as reliable as it seems, since you did test) is an outlier compared to the norm. Meaning, we're all estimating 15c too high, or you're estimating 15c too low...

 

[sir jman]

loopp order doesnt really matter