Next episode: More cooling design!
Having modelled the Corsair HydroX in 3D, I can now verify my assumptions on packing and fit - and explain my updated hypothesis on dealing with the chipset:
On the bottom layer of the board, you see the four areas requiring heat transfer: The CPU, the x570 chipset and two banks of VRMs. Whilst in theory the copper fins I have been using should be sufficient to deal with the VRMs, I'll try to integrate them in a combined cooling solution en-par or better compared to the the original ASRock design.
For the chipset, I'll have to re-purpose a CPU cooler as there is no product available OTS, and the plan is to use a heat pipe based solution with a standard 6mm pipe diameter that permits
- modding the CPU bracket to fit to the mounting hole pattern for the x570 cooler
- replacing the heat pipes with my own and connect them to a radiator outside of the board's footprint, enabling a clear vetical air flow combining forced air using a small (A4) fan with natural upward convection.
This is how I envisage the basic layout:
- one 6mm tube collecting heat from the VRM banks, possibly with a copper fin stack right above for added dissipation, linking to a three-channel contact plate over the chipset,
- two 5mm tubes connecting the chipset to a 40mm radiator with a vertical fan arrangement.
You see the pipes modelled in the length I have actually ordered them - and yes, the 5mm ones are too long (200mm) as I have been impatient and ordering them before verifying the fit. Wonder whether that's a good idea - not entirely sure whether heatpipes work only end-to-end or whether heat transfer can take place at any point along the pipe. May also inform the layout of the VRM pipe with three "bus stops"...
Another challenge potentially requiring a design change will be bending the tubes, particularly the 6mm VRM tube, to the radii required for everything to fit: Unfortunately, all contact blocks I came across have a 6mm pipe design, and component spacing on the board will require me to bend those pipes to a 10mm centre line radius. From what I read, that’s likely going to lead to ruptures- in that case I’ll have to use standard fins for the shorter bank of VRMs.
As my radiator of choice (
product link on Alibaba) comes with 5mm holes, I'll have to use 2x5mm pipes to link it to the chipset cool plate - plan is to use 1mm thermal strips to fill the gaps in the chipset block having 6mm channels. Alternatively, I may resort to soldering, but this is tricky as requiring low temperature solder and a very tightly controlled temperature window to avoid exploding heat pipes.
If my VRM collector pipe turns out to be too thick to achieve the bends required, I may revert to 5mm pipes for the VRMs, too.
One layer up, all as before: My modded EK Annihilator block - here shown as a complete loop with a placeholder terminal to connect to the GPU.
It features two placeholder G1/8 extensions that I cannot find anywhere as there is hardly any offer for G1/8 fittings that can match the plethora of stuff available in G1/4 - I may have to come up with something else there. G1/4 is not an option for my build as consuming too much space.
Over the heat pipe block used for the chipset, you see a modded CPU clamp that should work with the mounting pattern on the board - I have modelled after the actual piece you see
here (it's in the post), and with a bit of cutting, drilling and grinding it should come out like this:
Next up in the stack is the plexiglass block of the Corsair cooler - you see how it is positioned right on top of the EK Annihilator.
This allows to close the water loop (note that the terminal adapter is still a placeholder using EK hole spacing - still subject to redesign)...
...and to check for clashes:
There are none! So essentially that means that with a 3080/3080TI FE, I can use a disassembled, but unmodded Corsair block for what appears to be a comfortable fit!
Here you can see where the terminal block's o-rings don't line up with the Corsair block - its still a design I did for my old EK/2080ti setup.
Next comes the Nickel-Copper block...
...and my 3080 FE placeholder.
Thanks to PCI PCB thicknesses being standardised (1.55mm), I can now precisely assess the overall build dimensions taking "real" measurements from the Corsair block, and with the backplate applied...
...I get exactly to the 22mm I have expected...
...even giving me 1mm clear between case lid and backplate for added tolerance. I may want to lift up the GPU assembly though to touch the case to allow for added heat transfer from the backplate.
One HDMI and one Displayport remain clear and accessible, a second Diplayport should become usable with a bit of modding to the fan - that's all I need for this build, assuming that I'll re-route one DP to the board's thunderbolt interface.
That should all work out - so, how about the 3090?
Apart from the main challenge identified already, which is getting power to it (predictably requiring soldering it straight onto the card), the card would fit, but the block would need to sit slightly further towards the case's rear, resulting in a clash with the network module on the motherboard:
This is non-critical and can be addressed though by milling a notch into the Corsair's plexiglass block - very similar to my mod to the EK Vector 2080ti used for a previous stage of this build.
The terminal module will require some tinkering - as the 3090 needs to sit further to the rear of the case compared to the 3080, but the position of the terminal module is constrained by the space between CPU block and RAM, I’m looking at a gap between block and terminal should I want to use the part sketched for the 3080...
...where the ultimate challenge would be to come up with something that can work for both a 3080 and 3090 configuration. Reason for trying to keep the terminal module card-agnostic is to use the one-month lead in for machining and shipping from China for GPU hunting- and machining two terminals would be rather pricey as one off parts come at a premium. Perhaps I can use the module coming with the Corsair block to do the trick here.
The 3080 backplate appears to fit onto the 3090 despite the PCB sticking out on two sides: Key function of the backplate in my build is acting as a spacer between the PCB and the case, and ideally also to transferring some heat to it - this should all work with a 3090, too.
Why I ultimately may opt for a 3080 TI though (should it happen) is this situation here: the gap between GPU assembly and board components on the PCI connector side.
Exactly where I have my clash, I'd need to fit through the PCIe cable - this has already been the bottle neck of my last iteration with a 2080ti.
Whilst I could remove plenty of material from the plexiglass block when sacrificing RGB (not a sacrifice IMO, really...), a 3080 can leave much more breathing space for a cable, and the ambition would be to go for a PCIe 4.0 cable this time: Looking at the cables I got my hands on so far, however, they are much thicker than the amazing, paper-thin
HDPlex 3.0 silicon flex cable I have used for my 2080ti setup.
From what I can see, that's not an option for a 30x0 build, though, as I would have to switch the PCIe interface back to 3.0 in the BIOS to support the HDPLEX ultra flat cable. At that point I'd sacrifice speed of my PCIe 4.0 M.2 SSD - which is really the only component taking advantage of it. That's unless I can selectively "downgrade" the x16 port only - but from what I gather, it's all-or-nothing with the x570.
Strangely though, when using a 3.0 GPU such as the 2080Ti, the x570 just deals with it and still keeps the other lanes in 4.0 mode - but I expect to see a blue screen when plugging in a 4.0 GPU and then trying to squeeze that signal through my non-rated HDPlex flat cable. Would be interesting to know whether anyone successfully experimented with 3.0 cables in 4.0 mode - getting the HDPlex cable to work would be amazing!
exciting!), if ever available in this version of the universe,
