Stupidly High CPU Temps

[Disco_P]

So I have a GPU CPU loop set up as follows:

NZXT H440
2550K/3570K @ stock
GTX 780 @ 1200Mhz 1.2v
EK LTX Nickel Plexi
EK FC-780 GTX Ti - Nickel
EK FC-780 GTX Ti Backplate - Black
Alphacool NexXxoS ST30 240mm
Alphacool NexXxoS ST30 360mm
About 1.2m Primochill PrimoFlex Advanced LRT
EK D5 Vario X-Res 140
3x EK Nickel Compression fittings
3x EK Nickel plated brass barbs
4x Alphacool rotary barbs

On the GPU side everything works great, temps rarely exceed 60C under full load and that hefty overvolt whilst the case is closed and fans are silent.

But the CPU cooling simply does not work.
At stock under Intel Burn test load, the 2550K gets to around 70C and the 3570K reaches 75C. All the while the temperature of the water doesn't exceed 25C, the rads feel cool to the touch and the block itself feels no warmer than at idle/off.

So I know that the problem lies in the heat exchange between the Intel IHS and the block itself. But is this normal? I have mounted and remounted the block, always cleaning it and the CPU thoroughly and using Noctua NT-H1. The GPU staying cool shows that the pump, rads and reservoirs are working. Monitoring the Vcore shows nothing abnormal and when using my old NH-D14 cooler I would expect temps of around 70C when pushing my high overclocks.

I assume the 50C delta between CPU and water means that the block isn't working. Is there a solution other that doing an RMA?

 

[Disco_P]

It wouldn't be if the CPU wasn't on a WC loop and I didn't get the same results with a 2550K (sandy bridge - i.e. soldered).

Also compare it to tests done on the stock cooler


3C drop in degrees is a little underwhelming

 

[Disco_P]

Did you try different tim?
Like clu?
How old is your cpu?
It might as well be degrading...

I tried both the NT-H1 from Noctua and the stuff that comes with the block (Supposedly Gelid GC-Extreme)

Not really sure what CLU is (possible typo?)

My 2550K is 2 years old and the 3570K I got 2nd hand but cannot be more than 2 years old and according to the previous owner was only used in a media PC for the last year.

They *might* both be degrading but I would not place money on 2 CPUs both be going southside this rapidly.

 

[Disco_P]

On ivy/haswell water cooling makes virtually zero difference over air cooling, particularly with high stress programs like ibt. 3 haswell i7's I've owned have all been very hot runners. Hence why people delid ivy/haswell and replace the stock paste with liquid metal. 20c temp drops are pretty common after doing this.

I can explain the crappy tim with the 5C+ difference between sandy and ivy, but sandy should not be 70C on water in the first place.

I have ordered a couple of fittings that will allow me to remove and bypass the CPU block without too much hassle and then I will install an aircooler for testing and if I send the block back for RMA.

 

[Disco_P]

Why didn't they use solder on the 4790k ? If they did I would have brought 1

They used "Improved TIM" and in doing so proved that the TIM wasn't the problem. It was the shoddy contact it made between the dye and the IHS itself.

 

[Disco_P]

I have contacted the place where I bought the block about an RMA and hopefully they'll accept. I can only hope that the replacement will allow me to actually play games without the thing overheating and cutting out.

Then I'll only have to sort out OS, PSU and CPU. Fairly sure everything is breaking on me at the moment.

 

[Disco_P]

As far as I can tell there is no incorrect way of fitting the block as either port can be used as an in or out. I read the instructions again found here:
http://www.ekwb.com/shop/EK-IM/EK-IM-3830046990730.pdf

Step 2 and 3 don't make any sense to me. As far as I can tell it wants me to apply thermal paste and then clean it off?

 

[Disco_P]

I hadn't thought of that. I will have a look and try rotating it when I get home.

TThank you!

 

[Disco_P]

There actually is a correct way and wrong way, you have to mount the block so that it flows directly over the CPU die.

You can see the die is long in 1 direction only and the water in your block is designed to flow directly over the entire length of the CPU core.

If you have it mounted wrong then the block is only cooling the centre of it.

So make sure you have it flowing in the direction of the CPU core.

I am fairly sure that it had always been mounted such that the water flows along the entire length of the core. But I tried rotating in 90° anyway and it made it only slightly worse.

Are you are using the older LTX? You might need a new backplate that doesn't interfere with the socket backplate.

I only bought the block around a month ago and so am fairly sure it is the newer ltx. I also have the ek 115x true backplate that allows me to apply full mounting pressure.

 

[Disco_P]

I'll go through all my pics and find the most relevant ones but the link in my sig has my build log with 100s

 

[Disco_P]

Here are the 2 most relevant pictures:

Currently all the temp testing is being done without the side panels on to ensure maximum airflow.

 

[Disco_P]

Res -> pump -> GPU -> CPU -> 240 rad -> 360 rad -> res.

 

[Disco_P]

I highly doubt it is the GPU as when I do all my testing the GPU is idling at desktop. It will sit around 25C when idling and only about 1-2C hotter than the water. When I stress the GPU as well the water will heat up by 15-20C and so will the CPU meaning that under load it reaches the dangerous temps of 95C+. I will move the water temp monitor to between the GPU and CPU so I can directly measure the temp of the water going in.

Rearranging the loop will be a difficult procedure and I am fairly sure that loop order doesn't really matter more than a couple of degrees.

 

[Disco_P]

So the weird just got weirder. Or simpler.

I did literally nothing except put the 2550K back in (I didn't even bother with fresh thermal paste) and the temps no longer reach north of 50C at stock and when I pushed 1.4v through it, it did not go above 60C.

But the 3570K is still stupid hot. And I think I know why.

The CPU has been lapped. Poorly.
Or attacked by a gremlin.
Either way it doesn't look good and the temps still suck the big one.

I am going to do a temp test with an air cooler. This might not be fun.

 

[Disco_P]

Redid the tests with the 3570K making extra sure that the paste was spread well, the mounting pressure was high and that the water flowed over the dye.
Still super poop temps. 75C+ within minutes. Did a test a 1.3v and it shot to 95C+.

I don't know whether it is the more lapping or the crappy thermal transfer on ivy bridge but it is not good at all. I will contact the guy I bought it off as I was NOT told or shown anywhere that the CPU had been lapped.

But I am now much more happy with my 2550K temps and can now OC it again (even though I have no idea why the temps dropped).

So thanks a lot for all suggestions and now you can shout at me for buying a 2nd hand CPU (it would have been a steal w/o these issues)

 

[Disco_P]

Yeah I would tend to agree with you that the loop order is fine but it wouldn't hurt to try it in a different order then you know its definitely the cpu

Problem is that reversing the loop order will not be an easy task. Also the now good sandy temps indicate that the loop is functioning well and the GPU is literally doing nothing and sits at the same temp as the water.

A badly lapped one "could" cause this. It will not allow optimum contact as you guys know. Ideally quality lapping should produce a mirror finish effect. From the pic added looks like the previous owner lapped it with a combine harvester. Lol

Patented Combine Harvester Lapping Technology. BRB suing Samsung.

But even with stock TIM my 3770k never hit 60ºC @ 4.5GHz 1.236v under water. Yeah it'd help to replace with CLU/CLP but I think there's something else having an effect here.
The lapping, whilst not the finest of jobs looks to be pretty even, dont these things just have a nickel coating? If the lapping was on the **** you'd be through in places...

OP how is the TIM spread looking after removing the block? Nice and even?

Also, have you been checking Vcore on CPUID or the likes or are you relying on what the BIOS says? It's not uncommon for the two to be way off, you maybe have a huge V bump under load giving you nuclear temps

Edit:Updated my temp data after a check in my de-lid thread

I don't mean to brag but I am very proud of my TIM application. Even and thin coating with very little leakage
Z77 MPower keeps voltages in line very well. I set 1.2v in the bios and CPU-Z reports that under load they are 1.184 and sometimes flick to 1.192. I can get the multimeter out and check the V-check points if you are interested.

I have also seen your desire for a high end Z77 board. Have you considered the MPower - I trucking love mine.

 

[Disco_P]

When you say delidded do you mean you have replaced the poopy Tim and put the IHS back or is there direct dye to block contact.

Tomorrow I have a day off so will do some testing with my trusty nh-d14 to confirm it is the chip. If I can get my 2550K stable at 4.5+ again then I will be very happy to keep using that lovely chip.

 

[Disco_P]

Very nice, I would love to know the results.

 

[Disco_P]

Tasty. I have never had a good overclocking CPU. 2550K used to do 4.7GHz at 1.4v.

 

[Disco_P]

I have never pushed this chip beyond 1.4v.

I used to have a 780 lightning and it was pretty crap. £400 and it wasn't stable at 1189 at 1.212v. Reference 780 reaches 1241 at 1.212v and only cost £270. Plus the stock cooler is ridiculously sexy and the lights dance to music that might be playing.