Overclock.net banner

Direct die cooling water block

20K views 16 replies 5 participants last post by  foxrena 
#1 ·
It is known that direct die cooling help Ivy bridge and Haswell to lower temperature. But since the die is lower than CPU retention, the retention bracke has to be removed to let waterblock contact the die. The risk (pin damage and bad pin contact) associated with CPU retention removal is quite significant.

Since I have the tools I want to make a direct die block that can goes into the CPU retention so I don't have to remove it. Here it is, I hope you enjoy the pics. IMHO nothing looks better than fresh copper!

Testing results compared with CPU380 coming soon!
















Test done, see next post for summary.
 
See less See more
15
#2 ·
Summary:

In summary direct die mount does reach about 5C lower core temp than with IHS on! During benching the temperature improvement is very noticeable.



This is with direct mounting of KOC WXE over the die, without removing CPU retention, two nylon washers are added in the retention to ensure CPU mounting pressure. The waterblock then can be mounted with "figure 8" tightening method. I have repeated it twice and got consistent results.

I have also tried to add a thin silicon rubber pad which is slightly (very slightly, may be 5-10 thousands of an inch) thicker than the die. I thought it could help protect the die and balance the mounting. But the performance with this method is poorer than with IHS on by about 1C core temp. The reason I think is the additional forces required to compress the rubber, it offsets the mounting pressure and affects contact between silicon die and waterblock. So unless you are sure about the mounting pressure, adding rubber pad is not recommended.


If you decide to do direct die mounting, you are aware of the risks as I have sacrificed a processor. Be sure to use "Figure 8" method to tighten the screws and constantly check the compressing of springs are even. It is also good practice to tape the surface of the PCB to prevent scratching. Don't tape the surface mount caps! Because they are only a little shorter than the die, with tape it will be higher than the die.

Rule of thumb for mounting direct die:

1. Try to mount the heatsink as flat as possible and always use "figure 8" method to tighten screws/springs and only tighten quarter/half turn each time.
2. Never fully compress the springs. The mounting force will increase enormously after springs are fully compressed.
3. Too little pressure is also bad, which may cause poor pin contact in the socket. Normally I tighten the springs to 2/3 or 3/4 of the amount of full compression.
4. Use electrical tape to protect the CPU PCB surface from short circuiting. But never build tape higher than the die (for example don't put tape over the surface mount caps on Haswell cpu).
5. Center your heatsink/waterblock with respect to the die/mounting holes. Off-center mounting will cause tilting of the heatsink.

Even with this procedure I think this is still chance of damage since we are human. Think about the risk and gain before you practice.
 
#5 ·
Here are the results for Koolance CPU380:

Test methods:


4770K delid, MX-4 between Die/IHS and IHS/waterblock.
4.5GHz overclock, 1.285Vcore
AIDA ver2.80 stability test, running for about 5min, take core package temperature average

Results for CPU380, IHS on:

Water inlet temperature: 23.65C
Core package average: 73.7C
Delta core/water: 50.05C
Flow rate: 1.77 Gallon/min




Results for my direct die block, IHS on:

Water inlet temperature: 24.15C
Core package average: 73.2C
Delta core/water: 49.05C
Flow rate: 1.17 Gallon/min




So my waterblock has one degree advantage over CPU380. However CPU380 has larger flow rate. This is expected since my block is specifically designed to concentrate flow in the die area. Here are some pictures:

Here are some pictures:




 
#6 ·
I am trying to get direct die mount working. But something unfortunate happened - my CPU died!!

I guess the reason is that I scratched the CPU PCB when delidding it, and mounting without IHS caused short circuit since the retention is in direct contact with the PCB scratched area. Now I have it laying on my desk, here goes $300
mad.gif


I know that direct die mount can be done. I will have to be more careful next time.... fortunately Micro center is close.
 
#12 ·
Here are the results for the direct die mounting, without IHS:

Water inlet temperature: 23.65C
Core package average: 69.6C
Delta core/water: 49.05C
Flow rate: 1.17 Gallon/min


This is how I applied TIM:


In summary direct die mount does reach about 5C lower core temp than with IHS on! During benching the temperature improvement is very noticeable.



This is with direct mounting of KOC WXE over the die, without removing CPU retention, two nylon washers are added in the retention to ensure CPU mounting pressure. The waterblock then can be mounted with "figure 8" tightening method. I have repeated it twice and got consistent results.

I have also tried to add a thin silicon rubber pad which is slightly (very slightly, may be 5-10 thousands of an inch) thicker than the die. I thought it could help protect the die and balance the mounting. But the performance with this method is poorer than with IHS on by about 1C core temp. The reason I think is the additional forces required to compress the rubber, it offsets the mounting pressure and affects contact between silicon die and waterblock. So unless you are sure about the mounting pressure, adding rubber pad is not recommended.


If you decide to do direct die mounting, you are aware of the risks as I have sacrificed a processor. Be sure to use "Figure 8" method to tighten the screws and constantly check the compressing of springs are even. It is also good practice to tape the surface of the PCB to prevent scratching. Don't tape the surface mount caps! Because they are only a little shorter than the die, with tape it will be higher than the die.
 
#15 ·
Hi guys I have cracked 5GHz using TEC. Please have a look:

I tested 4770K under my TEC block (KOC T38). As expected, core temperature improve dramatically over water cooling, and I was able to get some decent results. This particular 4770K I used is about average, with VID at 1.056V. So I couldn't crack 5.1Ghz stable this time. Maybe a re-mount will help
rolleyes.gif
Anyway here are the results:
(The stock paste on silicon die was replaced with MX-4 in all tests)

5.3Ghz Super Pi, 1.55V, TEC at about -20C:


5GHz Cinebench, 1.44V, Score 10.88, TEC at -5C max


I can run 5GHz AIDA tests for a while, but I didn't test it for like 12hrs. The load TEC cold plate temperature saturates around 1-3C at 5Ghz. This is the TEC block I use. See This Thread for details.


Comparison of core temps using different cooling methods:

*Benchmark software used: AIDA (ver 3.0) stability test. Core temps are taken from the core package average after all temps are stabilized. Water temperature is taken at the water block inlet. Ambient temperature is between 23-26C.

Detailed results:
Koolance CPU380
KOC WXE, IHS on
KOC WXE, IHS off
KOC T38

This is something that can be best measured using TEC: Core temp v.s. IHS temp. Since I have a controller that can lock the cold plate at any given temperature. Here are the results:

Core temp scaling of 4770K:

Yes, the slope is >1! It means if you lower your heatsink temperatue for 10C, you will lower your core temps for more than 10C. It has been like that since the first generation of Core i processor. Ivy bridge has a slope of 1.33 in this temperature range.
The graph is easy to read, for example when cold plate is at 25C you will have average core temps of 71C, and when cold plate is at 0C you will have 42C average core temp.
GT core I think is the graphics core temp. I use PCI-E video card so the GT core is not enabled. It shows the temperature at that part of silicon die.
 
#16 ·
SetTitleMatchMode 1
SetTitleMatchMode Fast
 
#17 ·
Quote:
Originally Posted by dr/owned View Post

Direct die mounting without the retention bracket isn't really difficult to do. I've done it twice now. Because the waterblock (380i at least) extends past the footprint of the socket, so you can wrench the thumb screws down all the way since the block will eventually be resting against the walls of the socket and not crushing the die.
Been there done that since i7-920 (soldered IHS!).
But being able to keep the CPU retention is always nice.
rolleyes.gif
 
This is an older thread, you may not receive a response, and could be reviving an old thread. Please consider creating a new thread.
Top