High CPU Load Temps/Low GPU Temps??????

[NXLTrauma25]

Let me begin with stating first off I had this problem on my old computer with the same watercooling but hoped a new CPU block and new system would solve the problem, alas it did not.

The problem is I am getting great GPU temps on my 8800 GTS but the CPU temps are horrid. The ambient in the house is roughly 70 degrees Fahrenheit. The cpu sits around 35 degrees Celsius idle but then ramps up to 50 to 52 degrees Celsius on load. The gpu almost never budgets from 47 degrees idle to 50 degrees load.

I know for a fact I have seated the Block correctly and applied the correct amount of Arctic Silver. I have done a similar setup on a friends computer and have build plenty of air cooled rigs over the years and always done this correctly and had amazing temps. I have also reapplied and reseated the block on this and the old rig I mentioned plenty of times to no avail.

Now the hardware.

Phenom X6 1055T @ 1.28vcore 3.5ghz
BFG 8800 GTS 640mb

These two are cooled by the following loop.

MCP655 Pump
Black Ice GTX 240 Rad/SilenX 80CFM Fans in Pull Configuration
ThermoChill 120.1 Rad/Coolermaster 90CFM Fan in Push configuration.
DD TDX CPU Block
DD 8800 GTS Full Cover Block (BFG Watercooled edition came with this)
XSPC Res
Kandalf Case

The loop is as follows Res---Pump---GPU---ThermoChill---CPU---Black Ice---Res

I tried every combination of the loop and nothing changed the temps.

The Black Ice is mounted in the front of my case with the air being expelled out. There are 90 CFM fans mounted on the floor of the case pushing tons of cold air up in front of the rad for it to grab so no worries as to it being fed hot case air.

The ThermoChill is mounted on the exhaust port of the case by the CPU pushing air out with again 90 CFM fans pushing air up from the bottom of the case to it as well as a 90 CFM fan pushing air fron the side fan hole on the side of the case right onto the CPU and in front of the rad fan.

Here are some pics I grabbed with the phone...sorry for the quality.
NOTE: THE PIX WHERE THERE IS NO SIDE FAN ARE OLD...THERE IS A FAN THERE NOW.

So the basic problem I am having here is the delta of the CPU temps.
I have seen people run air coolers and less extensive watercooling setups than mine and had much better CPU temps but if the answer is simply that my loop is only going to give me these temps then so be it...just seeing if this is normal or not.

Sorry for the horribly long first post I apologize.

 

[OCD-OC]

just taking a shot here

first off, have you checked your CPU mount .. 9/10 times bad TIM or bad mount

Secondly :

Single loop, your temps across components will equalize, never going lower than the hottest component minus the cooling of your loop, does an BFG 8800 GTS 640mb run HOT on air ? around 70-90 deg c ?

What I mean is .. if you ran your loop with only the GPU .. and you get 50 deg C in the water

you run your loop with just CPU and get 30 deg C in the water

put them together and VOILA .. you still get 50 deg C as that hot water from the GPU is passing through every cubic mm of your tubes and stuff.

a simple test .. pull the GPU out .. (you have on board VGA to test it with ??)

run the rig as you have it already without the heat from the GPU.

does your CPU still run hot on load ?

if so it's a bad mount or another problem.

GL with fixing it

 

[NXLTrauma25]

Dont have onboard video to test it but yes the 8800s ran pretty hot if I can remember correctly. Idle I dont remember but I remember the load being up in the 80s and now it sits around 50.

The odd thing though is I see people all the time with even SLI setups and equal Rads getting better delta temps than me.

I checked the mount and it is snug as it can be...I even slowly backed off the 4 mounting points while in the bios and watched the temps get higher at idle...then screwed them back down nice and tight and watched the temps drop .

 

[OCD-OC]

I'm thinking there's actually something "wrong" here and by that I mean setup wrong or broken
.. others more knowledgable will rip this apart I'm sure.

I see your ambient at approx 20 C / roughly from 70 degrees Fahrenheit
an 80 C GPU is cooled to 47 c a 33 c Delta
and the GPU is within a couple of degrees of the GPU

Now I admit freely I do not know what wattage or amount of heat your rads fans and blocks should shift But ..30 degree delta is not brilliant to be fair, so yup something is missing

perhaps there is a flow issue, looking at your pics thats a LOT of pipework looks nice but maybe the 655 can't handle it all ?

not sure if the blocks are restrictive and halting it ?

if the GPU seems working I still think testing it without is the best starting point.

have you got .. or can you borrow a low end air cooled GPU to put a display on the screen so you can see the CPU temps on a few hours of CPU LOAD

 

[QuietIce]

to OCF!


You've got (I hope!) at least two 90 CFM fans on the bottom of your case pulling air (I assume) from below your case. How much space do you have between the bottom of your case and the desk it sits on??? When I built a top-mount rad machine I allowed 2" of space between the rad and case to make sure the airflow wouldn't be impeded. If you've got less than that then your 90 CFM fans probably aren't flowing at 90 CFM because their intake is restricted. Try taking the case side off and pointing a good room fan in there to see if your temps get better.

The odd thing though is I see people all the time with even SLI setups and equal Rads getting better delta temps than me.

I'd like to see these people. Your rig is barely at what I'd call an acceptable level of rad area for your components - and that's only because you're running very high-speed fans with rads built to use them. I like to have the equivalent of 2x120 rads per component so, in your situation, 4x120 would be a minimum (with 70 CFM fans). Your high-speed fans make up for this but just barely. I don't see how anyone can run a CPU and SLI at load with your set-up and still have great temps. In my opinion, 250 CFM (IF you're getting that much!) for cooling a CPU and a full-cover, high-temp GPU is really pushing the limit and it doesn't surprise me your CPU is running a little hot. In the loop below I've got 280 CFM for two CPUs and a GPU (not full-cover) and my CPUs run 45-50°C under full CPU load. When I load up the GPU on top of that (which never happens IRL) I BSOD because things get too hot ...

 

[m0r7if3r]

have you considered reversing your airflow so that you're getting cold air, I understand you think its good enough as is, but it's still picking up some heat, it shouldn't be too much work to turn it around, and if it fixes it, well that's problem solved, if not, what have you really lost?

 

[Conumdrum]

The TDX block is a pretty old design. Going on 3 years? All DD does is make a new top to fit the new chips.

It's pin matrix and flow paths are way behind the newer blocks. Could be that.

 

[QuietIce]

The TDX block is a pretty old design. Going on 3 years? All DD does is make a new top to fit the new chips.

It's pin matrix and flow paths are way behind the newer blocks. Could be that.

Oooh, good call! I didn't even look at the block he was using.

My Storms are fantastic blocks for my dual-core Opterons - the best there is and even better with a de-lidded CPU! But I sure wouldn't want to use one on my Phenom X4's, there's just not enough high-powered cooling area on those older blocks for those monsters ...

 

[Conumdrum]

Umm any reason you left the case side on for the pics?

"My engine isn't running right, take a look through the hood and see whats wrong..........."

Anyhoo, your temps arent that bad. Granted a AMD is much more sensitive to heat. Your still WAYY under any heat problems.

Comparing apples to oranges, my chip is different, but maybe you'll see what I mean.

My CPU loop is just the CPU.
My ambient is 80F.
CPU idling at 36 34 35 34
Loaded it up with prime 95.
Temps immediately jumped
49 48 48 48
10 min into a prime run
52 50 50 49


My chip is stock. My temps did the same yours is and I know my stuff is right.

When you say you are loading the loop are you running prime 95 small ffts AND Furmark at the same time? Only way to test a CPU/GPU loop. Please try that if thats not what you were doing, and let it run at least 20 min.

Awaiting proper test results.

 

[QuietIce]

Conumdrum, 50-52°C is about as high as AMD's will go with a good overclock. AMDs do NOT like heat - at all. On my WC loops I consider 40-45°C at load to be acceptable. On the other hand, even at high clocks and vCore it doesn't take a triple rad to cool an AMD to (what I call) reasonable levels.

I have no direct experience with the X6 but an X4 with a "normal" double-rad (120.2) arrangement and that vCore should be running under 40°C at load. My 940BE 3.6 GHz @ 1.36 (stock) vCore with a TRUE and Panaflo medium sits at 41°C (+20°C ambient). From what I've seen reported the X6's do run hotter but not that much. If he had my TRUE set-up on there he'd probably be running 45°C or less (since his vCore is lower than mine) ...

 

[evolvedpc]

This problem is not about whether the GPU normally runs at higher temps than the CPU - it is about the power output of the 2 chips, in watts, and the thermal resistance of their respective chip/TIM/block/water interface. In particular, the difference in load temps of the respective chips on air have absolutely nothing to do with their respective temps on water - especially when they're on the same loop.

The rate at which heat energy can be transferred between your rad and the air, or between your chip and the water, comes down to the thermal resistance of the interface and the temperature delta between the two mediums. For a fix thermal resistance, you must increase the temperature delta in order to transfer more watts of heat.

With this is mind, we see that the total power (heat) output of the chips (in watts) will determine what air/water delta you will need in order to shed that same heat via your rad. This characteristic is shared by the whole loop. The blocks have nothing to do with it at this point. For a given rad/fan setup and ambient air temp, total heat output will determine your water temp.

The chip/water interface, along with the respective heat output of each chip, is what will determine the difference in temperature between your chips and the water and (for a given ambient air temp) the temperature of each chip.

If one chip has a less resistant chip/water interface, then more heat (watts) will transfer into the water for a given chip/water temperature delta. A GPU typically has more surface area than a CPU, so it gets a head start on chip/water interface "bandwidth". However, GPU's these days often consume (and must therefore shed) more power than CPU's, so that advantage is eroded somewhat.

A quick net check reveals that your CPU has a TDP of 95W, while your GPU is rated at 146W, although your overclock on the CPU will probably raise it a bit from the stock 95W. I'm not familiar with AMD CPU's, so I could say how much, but it may well be shedding almost the same amount of heat as the GPU, only pushing it down a narrower thermal path - which would explain why the higher chip/water delta is needed.

Either way, it looks to me like you are getting the temps you should be able to expect from your loop. The GPU might output more heat, but it has a wider thermal path to push it down. If you have already re-mounted your CPU block a number of times with no improvement, then there is probably no room for improvement there.

A better block might lower your CPU temps by providing a less resistant path to the water and therefore requiring a lower chip/water temp delta to push its heat down that path. Better thermal paste might, too. (There is about a 3 degree performance difference between Arctic Silver 5 and Indigo Xtreme, according to this comparison.) You could also try cleaning your block and/or lapping your chip. All of these will help reduce the resistance of the chip/water thermal path - therefore getting you chip/water delta down and your chip temp along with it.

More radiator might help lower your air/water delta and therefore lower your overall temps. This would get your CPU cooler, along with your GPU, too. It won't make your CPU cooler than your GPU, though.

 

[NXLTrauma25]

I forgot to add the radiators dont seem to be suffering from heat soak as they are cool to the touch pretty much.

There is bout an inch maybe a bit more space between the bottom of the case and the fans on the bottom blowing the cold air in.

As far as radiator space this is what I have to work with in terms of space. I was under the impression that a ThermoChill 120.1 could easily cool a 8800gts or a CPU while the 240 Rad easily cooled either or as well.

I will grab the video card out of my parents computer in a minute and see what kind of temps I get.

Thanks for the quick responses and please let me know if there is anything else you can think of.

I will post the results in an hour or so.

 

[Conumdrum]

When you say they are cool to the touch, is that after loading the CPU/GPU after 20 minutes? Mine are also cool to the touch unless I load them up with a proggy for a while. Then they are warmer.

 

[NXLTrauma25]

During load the Rads dont even feel warm at all although the air coming from the fans does feel warm.

Well I am prime95ing it now with just the CPU in the loop and still the delta is the same.

CPU
Idle 30C
Load 46C and rising

Cores
Idle 17C
Load 34C

 

[QuietIce]

During load the Rads dont even feel warm at all although the air coming from the fans does feel warm.

Well I am prime95ing it now with just the CPU in the loop and still the delta is the same.

CPU
Idle 30C
Load 46C and rising

Cores
Idle 17C
Load 34C

You mean your original CPU temp quotes were from the motherboard's CPU temp sensor and not the core temp?
Man, it's no wonder you think there's an issue - and now everyone knows why I never suggest using a program that shows "CPU" temp for monitoring the CPU, it just confuses the issue.

Unless you have an X2 or X3 with unlocked cores (those won't show core temp) you should ignore the "CPU" temp provided by the motherboard manufacturer. The core temp is, by far, the better reading and with water cooling it's even more critical to use it instead of the CPU temp (which is really the CPU socket temp). If you're running 34°C at Prime95 load (even adjusting for the variance, which is maybe 10°C but probably 8°C) then you're good - there is no issue here.


What motherboard did you say you have and what program are you using to monitor temps ...?

 

[NXLTrauma25]

Asus M4A89TD Pro/USB 3.0

Would the higher delta in the CPU temp be as a result of the lower airflow over that section of the board since it is watercooled?

The reason I went by the CPU temp also is there is a known bug with the X6 core temps being radically off.

 

[evolvedpc]

During load the Rads dont even feel warm at all although the air coming from the fans does feel warm.

Well I am prime95ing it now with just the CPU in the loop and still the delta is the same.

CPU
Idle 30C
Load 46C and rising

Cores
Idle 17C
Load 34C

In order to "feel warm" your water would have to be warmer than your body -or at least your skin - so it would have to be getting up toward 38C. You probably have a chip/water delta of around 15C, so if your chip is 46C (and rising) your water will only be around 31C. i.e. your water should feel cool, but not cold.

 

[QuietIce]

Asus M4A89TD Pro/USB 3.0

Would the higher delta in the CPU temp be as a result of the lower airflow over that section of the board since it is watercooled?

The reason I went by the CPU temp also is there is a known bug with the X6 core temps being radically off.

The core temp sensor may not be calibrated well, meaning it has an offset (which I acknowledged by adding 8°C) but the precision is still much better. A difference of 17°C on the core temp between load and idle is, really, a 17°C difference - regardless of the actual temp. With the socket temp you get influence from environmental factors, as you've just demonstrated. You should also know that board sensors are seldom accurate themselves. Generally, the higher-priced the board the better the sensor but that's not always the case. Are you using Probe II for those temps? It's usually not too bad at "guessing" the CPU temp but it works better when you're cooling with air insted of water. As a check your CPU (in a CPU only loop) should idle 2-3°C above ambient.


Yes, I'd say you have some type of airflow problem around the CPU socket - and that's not uncommon for water-cooled rigs since there is no fan on the CPU keeping the air in that area stirred up. I would have thought the side panel fan would be enough but I'd also guess your output fans are starving for air (only an inch on the bottom isn't enough) so your airflow is bound to be a little odd in there. It can't be too bad, though, or it wouldn't take 10+ minutes for the socket to heat up like that ...

 

[NXLTrauma25]

I have used probe and other monitoring tools all reporting the same.

I will order some scythe 130 cfm fans for the case and radiator.
Do you think a lower Delta could be achieved with better fans and such or are rads maxed

I would have though when I tested with only the CPU in the loop the Delta would have been much less than 15 degrees..

 

[QuietIce]

I have used probe and other monitoring tools all reporting the same.

I will order some scythe 130 cfm fans for the case and radiator.
Do you think a lower Delta could be achieved with better fans and such or are rads maxed

I would have though when I tested with only the CPU in the loop the Delta would have been much less than 15 degrees..

Are you sure it wasn't higher than 17°C before you took the GPU out of the loop? What were your idle and load core temps then?


Higher CFM fans will help but raising your case up another inch is free and it should help as well. You're trying to pull 180 CFM through a space of 183 cm² when they are designed to have 288 cm². Even an extra 1/2" would help a lot ...