Scratch Build – In Progress Mr3D - Project Tatami (mATX computercase from scratch)

I was worried you wouldn't be happy with performance. Not having a center inlet can affect temps quite a bit and I have a feeling your base was just way too thick. You want as little material as possible between the water and the cpu heat-spreader, but I can't tell from the pics just how thick your block is. I think I settled on ~2mm for my base but I'm sure it's possible go with a lot less. And more surface area always helps so more/thinner fins.

At least for your test block, 76C on a 4.4ghz 5960x isn't that bad considering how much heat you have to deal with. I think most blocks on the market wouldn't be able to do much better than ~70C or so. I wish I had a beast like that to test my block on.

 

About the center inlet, that's why I made the test-block in the first place. Checking if there was a difference. I got ~3C cooler on a flow of water from the top, compared to a flow through the block. BUT I think those tresults are flawed/could not be applied on this setup. Because I had more or less non-existant measuring equipement. Just room-temp and cpu temp, the rig that was tested on only outputted ~100-140w, not 240w like mine does now. And last, that block, didn't have a optimised jet-plate. I don't think the jetplate itself is the important part here, but the inlet hole on the top is renerally too big. Checkout the swedish album I talked about and you'll see what I mean.

About the thickness of the base(bottom of fins to CPU distance). I had several people tell me that it would make a huge difference, but that is actually bull. Didn't include that in the update, but the original test-block had a thickness of 3mm. I tried it, milled it off to 1mm, and tried again. I got MAX 1C temp difference but that could've been measuring errors and slight variance in mounting on the 2 different runs. I did those tests yesterday, with a heat output of ~240w.

The CPU/VRM block used a distance of 4.5mm. Yeah, it sounds much, I'll agree on that, but the difference between 1 and 3mm was more or less non-existant then i don't think that could make a big difference.

I think the bad cooling-performance lies in the fin-arrangement. I need more and smaller fins.

 

That's cool you were able to test base thickness. Even though copper is one of the better metals for conducting heat, it still takes time for the energy to transfer from one side to the other and theoretically too much can act like an insulator. It's nice to know that it's fairly negligible in the real world.

And 3C difference seems about right for the center inlet. That's what the best performing blocks use, but I think having a jet plate is what makes a center inlet worth having. Without it I agree that it's not that much better.

Another thing to consider is that Intel usually makes the heatspreader slightly concave. Most cooler bases are convex to match so you could losing some contact with a perfectly flat block.

 

Yeah, I have to do some more tests on where to put the flow, and how much difference it makes.

About the convex surface, that is indeed true. I did some quick measurements of the cpu, it was ~0.1-0.15mm convex. I didn't measure the block, but it should be flat enough.

Looking at how the contact is when removing the block, it looked fine. The thing is that, when mounted with the force on the side of the part, the block will actually deform slightly and bend over the cpu. It's not so much like 0.1mm but should be some. That's one of the reasons I wanted higher mounting force and made the new backplate.

 

All of this.

The temps are good and inline with an octocore. Did you make a run of mountings to get a baseline? Could be a less than ideal mount.

shinji2k is also on the money regarding the bow of blocks,particularly EK blocks,it is requirement to get serious contact pressure. You could mitigate some of the issues by lapping the CPU....

As for the fin work,know anyone with a diamond slitting saw?

 

Well. I was going to say, "More surface area/woogly channels," but it seems everyone beat me to it.
-It seems a crime to have all that copper and not fill it with as much surface contact as possible.

 

Looking great

 

Good temps? The temps I've seen on other ~4.5GHz OCs is around ~65C...

I only made one mounting/block. But first I tried the CPU/VRM block, then i tried the testblock with 3mm base, then the testblock with 1mm base, and all the temps get close. Especially between the 3 and 1mm testblock, where it only differs 1C. So I wouldn't say it's the mounting.

What it could be is the convex surface of the cpu... I get a temperature difference of ~8C(sometimes I've seen 10C difference between all the cores) So maybe I should give lapping a go?

Well, I wouldn't lap it, but just surface grind it. I'll get ~1-2µm flatness if I grind it. What do you guys think of that?

I'll just have to use a smaller end-mill... The smallest we got is 0.2mm, but then I can only mill about 0.5mm deep. I think milling with a 0.4mm mill would be a better choice, could get 4mm deep with that one.

The thing with milling is that I could get the fins however I want. Wavey or whatnot. With a slitsaw I'll only get straight ones(and there fore only get on the same levels on the current blockmakers) I haven't ever seen a commercially made block with THIN wavy fins... And if I remember correct, back in the days when big and few fins was a thing, EK rocked with their waves. That's why I want to make a smaller version of that. Well, and also because waves should in theory create more turbulence and therefore better cooling capacity of the fins itself. Correct me if I'm wrong.

I'll have to do some testing on that! Make 2 similar blocks, only difference being one with straight fins, and one with wavey.

I said it above, but the reason I wanted fewer fins was to concentrate the cooling capacity directly above the die. Right after I milled it, the gut feeling told me it was too few, but I couldn't do more about it before I tried it. Do I need more and smaller fins? INDEED!

 

I've had good luck with just a huge water contact area filled with 1/32" fins/slots. (I say 'luck' because I just do stuff and hope it works.)

-Last block I did was a 2.25" square of slots.

Smaller fins? Go your own way, but I fear you will need a shrink ray and a minisub.
.
.
.
.
...I'm secretly hoping you have both.

 

Smaller fins? Why not waveier fins?






I NEED MOAR WATER BLOCKS!

You've seen the first block before, tho I just sandblasted it to get a "uneven microsurface", or whatever you wanna call it. Not sure if it'll make some difference, but we'll see.

Second one got more or less identical dimensions with the first one, but the fins are waved. The machining of the fins-only on this block took 6h...

The third block, running the fins now, but it's like the first block, but instead of making the fins 4mm deep, I'll make them 8mm. Well, it depends. I only got one 0.5mm mill that got 8mm clearance(the rest is 4 or 2mm) If this one breaks during machining, I can't make it. Estimated machining time for the fins is 13h for this block...

And the forth block, like the first or the third, but 2mm fin depth. I'll mill those fins tomorrow.






So... The CPU, checked the flatness of it while mounted in the CPU socket, and it wasn't that bad actually. Just below 0.1mm, but for the center, which is the important part, only a few hundreds of a millimeter.


That was all for this update, stay tuned for the upcoming days!


P.S. Some part of me is thinking "F#ck it, I should just pop the IHS off and direct-die water it..."(You know, where you push water on tho the actual die directly.)

 

that block is Sick. Make me one

 

Getting a water tight seal around the die would be extremely tricky --especially compensating for expansion/contraction with temperature fluctuations. Better to make a block that rests directly on the die, with a shimmed edge so you can't accidentally crush it. Good luck...

 

Plus, you're adding dissimilar metals into the loop, increasing the chances of galvanic corrosion - who knows what effect fluid additivess will have on a bare CPU.

Those water blocks are gorgeous, by the way.

 

I'll think about it, but all my crazy ideas comes up at night...

Didn't even think about the galvanic corrosion there.



So... Some results from todays testing.






Here we have 4 different blocks. 3 of them, the fins is the same, only thing that differs is the height of the fins. 2/4/6mm fin height.

On the last block, the fins are just like the straight ones, but waved. 4mm fin height.

And the difference? Well, I didn't study the results closely yet, but I've seen what happened during testing. The tests was run 2 times on each block to remove/limit my error when mounting.
I'm really disappointed with the results. 2-3C temperature difference, at MAX. And that's between the blocks with straight fins, between the straight and waved one is MAX 1C...

On to some more testing tomorrow, bed now.

 

There's still a huge chunk of real estate on those that hasn't been 'finned up'.
-Yes I'm nuts, but if you are going to do a ton of milling...

 

Don't think these are the last test-blocks I've milled... Going back to work to mill some more today!

 

The IHS for Intel chips is plated in Nickel.

Also,direct chip cooling doesnt really work,the DT sniper block which came out tried it....and failed.

Better to delid and run lidless,however you will need a shim to clamp the CPU PCB flat to the socket,PCB's warp when unsupported.

 

Well, that isn't true. Me and a friend tried direct-die cooling on one of his old Q6600 CPU. Sure, that's only 100w, but it still performed out of this world. Idle water/core delta was ~2 degrees, and load delta was ~10C...



I did the milling, and he did the testing.

Here is the link to the original swedish gallery: http://www.sweclockers.com/galleri/11772

We made 2 different blocks, but check the last one.

Yeah, PCB's are really weak, even with a IHS there are some deformation(just check one of my previous posts when I measured the CPU flatness. I know the CPU isn't flat in a resting case, but you can clearly see that it was deformed when mounted.

 

Saw a link to this via mod zoo (cheapskate) amazing work

 

Update time! (Finally...)

Sorry for my inactivity guys, just been too lazy to update the log the past time, but now I'm here with more blocks and also some results!





Here we have a direct IHS block, made to cool the IHS directly. By clamping it tightly to the CPU the o-ring seals the water against the IHS.






Another one is this one. Took 24h to only mill the fins/cubes. Yes. 24 hours... The program(G-code) was over a million lines and 28mb in size. For only text, that's A LOT! :hypnotized:




But those wasn't all. Here we have 9 blocks, with different fin-layouts.






Performance? Well, I'm slightly disappointed that I didn't see bigger differences between some of the blocks.
These results are with water only flowing through the block, no jet-plate/center inlet(beside the XSPC and IHS blocks).

So, how about a description to the blocks.

Rev 1.15: The first block I made. 0.6mm for water, 0.4mm fins, 1mm pitch. Area of 32x32mm. The fins are 4mm in height.

Rev 1.25: Exactly the same as Rev 1.15, but the fins are 6mm in height.

Rev 1.35: Exactly the same as Rev 1.15 and Rev 1.25, but the fins are 2mm in height.

Rev 2.05: "Findata" is the same as Rev 1.15, but I made them wavey instead. 4mm in height.

Rev 3.05: So, I wanted to see the difference between fins and no fins. The area is still 32x32mm, but with a height gap of 0.5mm for the water to pass through.

Rev 4.05: More fins, what would that do for a difference? The fins are the same as Rev 1.35, but instead of 32x32mm area, it is 36x42mm. 2mm in height.

Rev 5.00: Here we have it, the IHS block. Why? Well, I wanted to see the difference between a layer of copper and thermal paste(Rev 3.05), versus none(This block). Height between CPU and the built in jet-plate is 0.5mm.

Rev 6.05: To create as much turbulence as possible, I made this. The block have a cube pattern that's overlapping, just like a brick wall. So the water is forced to smash against each cube on it's way through.

Rev 7.05: So, to test the difference of different fin thicknesses, we have this block. It's similar to Rev 1.15, but 1.1mm gap for water, 0.9mm fin, 2mm pitch. Area is still 32x32mm and they are 4mm in height.

I also lended a XSPC Raystorm from a friend.