PDA

View Full Version : Removing MOSFET heatsinks from DFI NF4


Jimbo Mahoney
06-03-2005, 03:34 AM
Hey guys

Just wondering if anyone has done this yet?

I know they are held on with grips which are soldered to the back of the board.....

I will be trying it this weekend and then fitting my MOSFET waterblock.

Just wanted to know of people's experiences...

Thanks

teko
06-03-2005, 03:39 AM
looking forward to see your results, as I would like to do the same, but is a bit of chicken :rolleyes:

Jimbo Mahoney
06-03-2005, 04:18 AM
Damn, am I gonna be a guinea pig then :stick: :( :cool:

teko
06-03-2005, 04:26 AM
Sorry mate, but I'm a chicken :chicken: And I have to make the wc for the mosfets first.. But I thought that they where just glued to the chips..

Jimbo Mahoney
06-03-2005, 04:38 AM
I thought they were glued or similar, but if you look closely, you will see two 'grips' either side of the base of the heatsinks. These pass through to the back of the mobo where they are soldered in place.

Jimbo Mahoney
06-04-2005, 10:32 AM
OK, I'm back and everything survived its encounter with the soldering iron!

It's a bit fiddly, but not hard.

I melted one prong of the 'fet sink grip thing and pushed the sink to the other side, so it levered that prong out, then repeated on the other side.

Ideally, you should clear the solder out of the small holes, as they can then be used to put small cable ties through to secure your mosfet waterblock.

I ued silicon thermal compund - a reasonable size blob - on each 'fet because there seems to be a LITTLE play in the block - ie sitting on three sets of 'fets, it tends to see-saw on the middle set a fraction of a mm.

I decided not to use AS5 or AS epoxy because they are both slightly capactive / conductive and the 'fets don't need masses of cooling anyway.

My main reason for cooling them with water is they get VERY hot (60'C) unless you have a fan moving air around / near them and I don't want any fans (except in PSU and on radiator).

My opinion:

Don't bother watercooling the mosfets unless you don't want a fan pointed at them.

Some temps (during Prime):

no fans - 60'C
120mm fan pointed at this area - 36'C
watercooled - 37'C


Now, whilst it may look like watercooling is worse, remember that the sensor is actually only NEAR the 'fets and will be severely influenced by the large amount of airflow a fan provides. With the waterblock, we are back to zero airflow around that area, so 37'C sensor reading probably indicates a 10'C lower temperature than 36'C with tons of airflow.

Hope this helps others!

Funny_S
06-04-2005, 10:44 AM
you got pics? :stick:

Jimbo Mahoney
06-04-2005, 12:03 PM
Here's one:

http://homepage.ntlworld.com/james.waynewright/Mosfets.JPG

Won't really be able to get any better ones, sorry.

The mosfet block only has 3/8" barbs, hence the tygon for that block.

The rest of the system is 1/2" black silicon tubing.

Adding the mosfet block drops flowrates to about 4.5 lpm / 1.2 US gpm (from 5lpm / 1.4 US gpm).

WiCKeD
06-04-2005, 02:46 PM
Wow you did each one? At least it looks like it didn't affect the flowrates much.

EMC2
06-04-2005, 04:13 PM
/me wishes a close-up pic before the WBs were added was available

teko
06-05-2005, 05:46 AM
Thanks for being guinea pig ;) I was only thinking of water cooling the one on the top of the board, the one for the mem volts.. But might go all the way as you have ;)

Jimbo Mahoney
06-05-2005, 09:07 AM
Wow you did each one? At least it looks like it didn't affect the flowrates much.

Well, the waterblocks you can buy are typically designed to cover all of them.

The one under the CPU and the one in the top corner of the board are still aircooled, but poking a themometer on them only shows them to be 40'C.

The three groups of three next to the CPU socket get VERY hot, especially when you are overvolting and running Prime or something.

Right now, @ 2.826 Ghz and 1.525 volts (1.488 actual), the 'fets are being reported as 51 'C
:slobber:

Remember there is NO airflow so that is probably a realistic temperature, and that's watercooled.

Either way, I couldn't get beyond about 2.7 - 2.75 Ghz Prime stable before, and now I can so it's either:

The watercooled mosfets help stabilise the CPU

or

The mounting of the CPU is better this time (I had to take it all apart to mod the board)

or

The CPU has burnt it.


My money is on 1 and 2 - the CPU temp is a couple of degrees lower (relative to ambient) and the 'fets must be loving the water 'cos their output is ridiculously stable under load.

Jimbo Mahoney
06-05-2005, 09:09 AM
/me wishes a close-up pic before the WBs were added was available

Why?

Do you mean after the removal of the sinks, but before the waterblock?

Under each sink are three tiny mosfets. Totalling 9.

The underside of each sink has some themal tape and they are held on firmly by the soldered grips which go through to the back of the board.

I think people can get similar results from aircooling them, although it's hard to say because the probe is also massively cooled then...

Jimbo Mahoney
06-05-2005, 09:12 AM
Thanks for being guinea pig ;) I was only thinking of water cooling the one on the top of the board, the one for the mem volts.. But might go all the way as you have ;)

No probs.

I don't think it's worth watercooling any of the 'fets apart from the ones for the CPU.

As I said, the memory one only reaches 40'C (at least the sink does anyway).

Unless you are running your mem from the 5v line using the 4v jumper? I wouldn't recommend that anyway, but that would make that 'fet work harder and hotter. It's better to use the jumper trick so that you get all the options in the BIOS, but it will actually take the voltage from the 3.3v line. You will get a maximum of your 3.3v line minus about 0.1v. With my line @ 3.6v, this would allow upto 3.5v and keep the 'fet cool 'cos it isn't doing as much work.

Luckily, the sweet spot for my RAM (TCCD) is 3v.

Jimbo Mahoney
06-05-2005, 09:56 AM
Right now, @ 2.826 Ghz and 1.525 volts (1.488 actual), the 'fets are being reported as 51 'C
:slobber:

I just realised why it's being reported this high:

prodding my finger on the black square components and the silver cylinders (caps?) next to the fets reveals they are really quite hot!

That would explain why the probe is reporting such high temps - it's the black square block right next to it!

Wish the probe was in between the actual 'fets.
:rolleyes:

EMC2
06-05-2005, 02:24 PM
Why?

Do you mean after the removal of the sinks, but before the waterblock?
Yes... so I can see which FETs and driver they used... particularly interested in which driver ;)

Under each sink are three tiny mosfets. Totalling 9.
Actually... 2 FETs and one FET hi/lo driver...

BTW, those "black square components" that get hot are the output inductors of the switcher... and yes indeed the "silver cylinders" are low ESR caps.

Peace :toast:

stealth17
06-05-2005, 02:38 PM
lol the only fets i would even bother cooling is the vdimm ones. the one with the stock sink is REALLY hot with high current. only if you have ZERO case air flow would i bother doing those....nice job though! hey this is XS even if it isnt practical right!

id like to see the black up close. it looks pretty!