Mounting Tests.

[jaydee]

Now that I got my test bench together I am working on my mounting technique. Now that I am able to acquire more reliable data I can see how bad mounting technique can cause issues.

Here is the results from my first 3 mounts of my Project X block.

Mount 1: (CPU temp subtracted from water in temp)
13.57C

Mount 2:
10.51C

Mount 3:
12.06

That is a pretty ugly thing. This is without using spring BTW. I am now going to dig up some springs and remount 3 more times. Trying to get a good mounting system down. This also shows how one mount from many of these reviewers just doesn't cut it!

[Cathar]

Nor even 3 mounts.

I generally just keep on mounting for about 12 times. Take note of the lowest temperature reported over those 12 mounts, and then keep mounting until I reproduce that value, or at least within 0.2C of it, twice more. If I don't see anything better then, then that's pretty much where I'm confident that I've seen what the block can do.

Anything less than the best (or very close to it) mount, is only just measuring mounting variations and not what a block is capable of.

[jaydee]

Yeah I was going for 8 mounts but noticed such a large difference after 3 I decided to try some springs. If things are closer after 3 I will continue the remounts. I really like this data logging with the dow sensors. Guess I need to pickup MS Excel one of these days to use the CVS logs to make those nice graphs. I got an old version of lotus with spread sheet but it's graphing is rather useless.

[jaydee]

With Springs First 3 mounts.

Mount 1:
12.13C

Mount 2:
10.87C

Mount 3:
11.93C

[freeloadingbum]

How much spring load are you using?

[jaydee]

Originally Posted by freeloadingbum

How much spring load are you using?

Who knows... :shrug: I used one long spring and cut 4 sections of equel length from it. Screw each one down in a cross pattern untill the gaps are almost but not competely gone. No way to tell absolute mounting pressure even if spring specs where known.

EDIT: Also it is enough to keep the block from moving at all.

[snowwie]

why wouldn't you be able to tell mounting pressure if you knew the spring specs?
assuming the block is mounted horizontally, of course

knowing your mounting pressure seems crucial...esp. for comparing blocks

[jaydee]

Originally Posted by snowwie

why wouldn't you be able to tell mounting pressure if you knew the spring specs?
assuming the block is mounted horizontally, of course

knowing your mounting pressure seems crucial...esp. for comparing blocks

If you ask a question like that then you must know the answer. How am I supposed guage torque? Torque wrench won't work, no way to hook up a pressure sensor..

Repeatability is the goal anyway. That is why you make several mounts with the same method. Same way pH does it and Cathar as explained above. With decent temp monitoring you can pick up the difference. The main problem is not pressure difference but uneaveness. Once you get your technique down then every mount should be near identical which is more important than knowing the pressure. As long as it is the same each time the absolute pressure measurement is uneeded.

When I get my die sim completed I got a way to measure pressure but it isn't possible with a motherboard. Basically the way JoeC does it at OC'ers.

[BillA]

your bp is pointed, NOT flat
lol
(I'm not laughing at you jd, just the long learning process you have started)

your mounting is shit
buy some springs (ask the spring rate), then you know the required compression to achieve a desired load
-> you measure the spring height

skip torque

[jaydee]

Originally Posted by unregistered

your bp is pointed, NOT flat
lol
(I'm not laughing at you jd, just the long learning process you have started)

your mounting is shit
buy some springs (ask the spring rate), then you know the required compression to achieve a desired load
-> you measure the spring height

skip torque

Thanks for dropping in. I will look for some springs with known spring rate today. Might be hard to find around here. Measuring spring height will be a challenge. Will see if I can work my micrometer in there some how.

[BillA]

suggest you look at springs already 'sorted out' for the socket/pressure you want
2 or 4 ?
then buy a mounting kit with those springs from your preferred vendor

I cut down a digital mic so I could lock it, and work it around all the other stuff
mount the wb first, then connect the soft silicone hoses
-> after 2 hrs jiggle the posts and wb, then wait another couple of hours
suggest using BillA Goop from AS, settles faster than most other stuff

[pHaestus]

I think incoherent's testing methodology has more or less rendered variations due to mounting pressure and TIM application irrelevant hasn't it?

Jaydee it took me a month or so to get really reproducible results. When I return to testing now I can usually get consistent results after a day or so of fiddling. Also I seem to get much more consistent results if I overtighten. power on, and then back off of the springs to the desired point.

[BillA]

"Also I seem to get much more consistent results if I overtighten. power on, and then back off of the springs to the desired point."

similar effect as my jiggling, work out some of the TIM

[freeloadingbum]

Here's a question. When deciding on what load to use, would it not make sense to use a load where the paste performance starts to plateau? For an example, on the chart included, it would appear that between 12 and 14 turns the performance of the paste (BillA Goop) doesn't improve much. Would the user not achieve less variation using 13 turns than using 6 turns (unfortunately in this case the actual load is unknown)?

Also, would using tubes (metal or plastic) cut to desired length, fitted around the springs, help improve repeatability? Then all you would have to do is thighten the thumbnuts (evenly) until they just about touch the top of the tube.

[pHaestus]

Repeatability may not be affected SO much because the source of the variability is presumably the thermal paste application (and how it spreads out of the interface when you tighten it down). I would assume that slightly off-flat wb mounting will result in a different final interface (and performance) than a perfectly aligned mount regardless of the pressure you use. That's separate from the force you apply in other words. By taking great care in how I do things I can generally get to within 0.2C dT of my best mount pretty consistently even with ~30 lb force.

You would really stack the deck against wbs that use the socket lugs or anything other than 4 hole mounting by cranking the springs down on the 4 hole blocks to far past specs while the other blocks are fixed at the mfgr-supplied "safe?" settings. I don't think you could put the same springs on a block that uses the lugs without shearing them from the socket...

[BillA]

non-flatness is a significant source of variation,
unless the mounting procedure is such that the same 'side' of the cone is pressed against the CPU each time (doable but difficult)
- this can be seen by reading the grease (if not disturbed by dismounting)

w/o a level baseline test results are meaningless, for me currently it is a 50lbf total load (non 3 lug products)
- but any value can be used, presumably related to the DUT's operating conditions

[freeloadingbum]

Originally Posted by pHaestus

Repeatability may not be affected SO much because the source of the variability is presumably the thermal paste application (and how it spreads out of the interface when you tighten it down). I would assume that slightly off-flat wb mounting will result in a different final interface (and performance) than a perfectly aligned mount regardless of the pressure you use. That's separate from the force you apply in other words. By taking great care in how I do things I can generally get to within 0.2C dT of my best mount pretty consistently even with ~30 lb force.

You would really stack the deck against wbs that use the socket lugs or anything other than 4 hole mounting by cranking the springs down on the 4 hole blocks to far past specs while the other blocks are fixed at the mfgr-supplied "safe?" settings. I don't think you could put the same springs on a block that uses the lugs without shearing them from the socket...

How do you know you're using a ~30lb force if you don't know the spring rate? It's not that hard to measure the spring load within +/-1lb if you so choose (+/- 1lb being better than nothing). I'm still hoping Cathar will know the spring load at 5 turn.

Also, without measuring the spring rate, how do you know that, at present, the cards aren't being stacked against the 4 spring blocks?

At this point we don't know what the load is at 13 turns (it may be 30lbs for all I know) but yes, if it's too high then it's not an option.

[pHaestus]

You're mixing apples and oranges here somewhat.

The "13 turns" thing is from the Aquajoe blocks, which used a proprietary mounting system and made it impossible for me to use the same springs as in other tests.

Cathar's "5 complete turns for AMD recommended force" is what I went with because frankly no other mfgr even mentioned anything about how far to tighten the springs down. And once I had started testing that way it was easier to continue with the same methodology than to repeat all the testing again.

As far as how I know that the cards aren't being stacked against the 4 spring blocks, I don't live in a vacuum. BillA tested many of the same blocks I have (LRWW,Innovatec rev3, MCW5000-A, MCW6000-A, AquaJoe, etc etc) and the relative rankings of his C/W vs flow and my dT vs flow curves look pretty close. And I recall Bill making custom mounting plates so he could use 4 hole mounting and same springs on all the old socket lug blocks (like Innovatech Rev3). So I operate under the assumption I'm in the ballpark.

[freeloadingbum]

Originally Posted by pHaestus

So I operate under the assumption I'm in the ballpark.

That's the problem. Why leave anything up to assumptions when you don't really need to. Isn't that what testing is all about?

[BillA]

hey guys, we know the spring rates for each that we use - and we sell them too
and I will provide the spring rate to any that buy for testing
(normally I define the compressed height for a given load)

if purshasing springs defeats the DIY concept, then buy the springs from a spring mfgr
or make 'em up as jd did

BTW, Swiftech spring assys generally have limiters of some sort

[jaydee]

I was using the 4 holes on an Epox 8K7A. Now I am using the 4 holes on a ECS K7S5A because the Epox was dead when I got back a few minutes ago. I was expecting that though as the heat sinks on the mosfets were literally sizzling and the caps been replaced once already.

So now I am going to swap out mobos and restart.

I bought some compression springs said to be rated at 16lbs per inch. A little stiff maybe but I think I can work with that until I decide which stock kit to order. Probably be the Swiftech as I want a MCW6000 anyway. Just as well buy the block and hardware at the same time.

So now to the question of how much load to apply to each spring? I assume I want to take to total load I want and divide it by 4? Then use that number on each corner?

Another thing crossed my mind as I was looking for springs.... What about the difference in weight of the blocks? Is the weight of the block going to change the pressure?

[jaydee]

And for the TIM I am using the silicone crap from Radio Shack. I figured very little to no setup time for it. Is that a bad idea?

[freeloadingbum]

Originally Posted by jaydee116

I bought some compression springs said to be rated at 16lbs per inch. A little stiff maybe but I think I can work with that until I decide which stock kit to order. Probably be the Swiftech as I want a MCW6000 anyway. Just as well buy the block and hardware at the same time.

Using 832 (or632) bolts, 15 turns should equal a 30lb total load at the 4 springs. Ideally, you don't want to compress the spring more than 80% of its total available distance as the rate starts to ramp up in the last 20% ( from what I read, but I imagine it will vary somewhat depending on the spring)

[jaydee]

I am using 6-32. I got everything together with the ECS board to find my Dow sensor for the water temp to has died... Odd. Well back to calibrating probes... Maybe I will get 3 of them together and get water out temp aswell...

[bigben2k]

McMaster has a pretty good spring selection, and a guide too.

Their guidelines also refer to compressing a spring at no more than 50% of total compression, otherwise you de-rate the spring (i.e. you "bend" it, so it won't behave the same way again).

It took me a bit of calculating to find the right spring, and I still didn't get it right; I got a spring that reaches the target compression at maximum spring compression, leaving me no "play".

Follow Bill here; many good tips: use a limiter, use a washer on top and bottom of the springs, and if possible (or necessary), use a sleeve; it'll help you get steady, repeatable mounts each time. Compress a little bit more than your target, then loosen it a bit, to help the TIM settle nicely. Whatever you do, just do it the same way each time.


I have yet to hear some good advice on cleaning up a mess, between remounts.