Okay, I've put back the steel screws and indeed the heat exchanger started working again.
I should have used more due diligence before embarking in this attempt with the plastic screws.
As usual when it's late and I'm sleepy, I should just go sleep and avoid compulsory shopping...
Turns out there are 3 types of plastic screws; nylon, Polypropylene and PEEK (Polyether Ether Ketone).
PEEK is a "thermoplastic super-engineering plastic".
Word of NBK, japanese quality since 1560. I'd say they sound quite reliable...
But it wouldn't work even with their screws.
Since they are a serious company, finally, they publish datasheets with the torque:
Their M3 screws in PEEK have a torsional torque of just 0.3.
And there's a note stating the recommendation torque is 50% of this value.
I didn't bother to make the calc but to match the 4.5 torque from the 8 x steel M3 you probably need dozens and dozens...
The only way to use these magic screws is with a custom design heatsink where you can plan to use M10/M12 screws.
They have a torque of 10.8 and 18.6. Half of that will probably suffice for 4 screws per high powered TEC.
So I'm going to stick with steel screws also with the next heat exchanger design.
I'm back testing with the setup and I've done enough to wrap up some findings.
The i4770k is now clocked at 4.6 GHz with 1.3v; I'm matching the Guru3D configuration since they have tested over the years a lot of AIO kits.
It's a good reference for benchmarking. I'm doing the same double wPrime 1024M benchmark twice and checking the max core temp.
I was expecting much more help from the pipes and rad/res/pump insulation.
Indeed going from the small single AIO pump to all these separate components had a negative impact.
I thought the Monsta rad would be the biggest detractor but I was wrong.
The Monsta insulation is working pretty well, the temperature difference dropped from 2-3c down to 1c.
But all the other components, probably combined with the mounting inside the case where the hotloop rads and fans are pushing, are bringing down the efficiency massively.
Now the TECs really struggle to bring down the coldsink to 15c and settle to 6-7 amp, while on the previous open bench AIO config they could go down to 9c.
The pipes are still very long, much more than they should, and I have a res with quite a lot of water in.
It takes a lot to precool the loop, much more than before.
The insulation is already working quite well, even with the Monsta in.
While with the open bench setup I had a deltaT between the coldsink and the water between 2c and 3c now I have just 1c difference.
It's a very good result considering how many parts are not insulated at all.
If you don't want to do something complex like my setup and just have an heat exchanger, use a CPU Pump block with a small res inside.
A very small loop with just the Monsta and the heat exchanger is highly efficient and easier to manage.
With this setup I definitely have to insulate very well the CPU block; the HK Pro is very efficient to dissipate cold too.
Had to shut down the top rad fans cause they were sucking away the cold.
It's something I'm going to do when I'll move to the main rig with the new controller and the tec-sandwich-castle on top.
Hope it's not only eliminating it as a detractor but helping enough to make the whole much more efficient.
The pump and res are a big detractor and I can't eliminate them.
But I hope to improve big time with the EK Volume FLT D5.
Less piping, no more separate top and res, just one unit.
I'll have to think about a proper insulation for the pump, I wrapped around polyurethane foam and didn't see much improvement.
I could confirm the coldloop pump speed setup matters; you need to run it at low speed to precool the loop.
Unless you have a perfectly insulated loop which is not my case; I have the Monsta rad in at least.
While under thermal load it has to run at full speed to remove more heat.
While the performances are not striking good right now they are not that bad.
Without the TECs running, just the Monsta rad, insulated, the max core temp is 70c.
It's the same level as a Kraken X63 and similar.
With the TECs at full amp and 15c pre-cooling max core temp at 57c; 13c improvement, 8c less than a Corsair H110i GT running at max speed.
But running at the limit is not the best; at 18c can still get 59c with much less power consumption.
The hotloop is gloriously oversized and works really well.
Which is good news considering that I want to put a whole lot more heat in there in the future.
The Microdrives are keeping the DC-DC converter, the TEC controller and the Heatsink at 35c with a very moderate fan noise.
I've already limited the fan speeds and stopped the top fans and I can probably reduce the speed even more.
With the correct fittings and piping and the flow at max speed I don't see any problem there.
Next short term task is to have the new EK Volume FLT D5 in the coldloop.
I'll improve the insulation and since I'm going to drain the loop I'll probably work to shorten the pipes to the correct length.
The Styrofoam is fantastic; there could be the arctic or a volcano behind, at touch is just ambient temperature.
But I need to cover at least the aluminium tape with a normal plastic insulation tape.
It's good at shielding from heat but not good enough at keeping the cold.
All as expected but I did want to feel it with my fingers
In the meantime I'll start working on the new heat exchanger design and the parts for the CPU block.
Now that I don't have to mess too much with the case I'm getting back the fun...