[Build log] Depth Charge a Framework Desktop Monster

[Arkratos]

Welcome to my adventure in watercooling the strangest set of hardware I’ve ever run!

Today is the start of my build log for the Quake con 2026 case mod competition and the goal of the build is… A manta ray! And to suit the nerd that I am and my childhood love of beast wars/transformers, I will call this build Depth Charge after the manta ray character.

Photo credit: Wikipedia

Background: Hi, I’m Cameron. I go by Arkratos/TheArkratos on most of the internet. I’m a hobbyist case modder. Usually I just enter the Quake con case mod competition because I used to be local to the Dallas/Fort Worth area. I also almost always design my own cases for competitions because no one really makes a case to support the weird things I want to do.

A few months ago I managed to take 1st place in the scratch build category with my computer I call Lost Atlantis. But I have to give credit to the guys that also got 2nd and 3rd place; those guys’ computers were just insane. 2nd place was a professional cosplay-grade shield from the new Doom game. It is also a working chain saw, obviously without “real” blades. It had a pull start that would trigger the blades popping out and rotating! 3rd place was a 160% scale (I believe) portal gun. It was amazing with removable panels, easter eggs in the design, all I can say is it was just phenomenal! Quake con’s Instagram video shows the builds nicely (although a bit quick): https://www.instagram.com/reels/DNJtp1RMqJ_/

So going into next year I decided I want to up the ante: I want to incorporate specifically large motion into the build. I remembered seeing this marvel of engineering from the Cooler Master Case Mod World Series 2018: https://bit-tech.net/features/modding/case-mod/cooler-master-case-mod-world-series-2018-the-winners/7/ The Crescent by ‘CPSM’. This computer has a “pumpless” watercooling system. The entire system rotates to treat the computer like an hourglass: gravity pulls the liquid through the computer from one reservoir to the other and then it rotates to do it again.

Base idea: Long story short, my plan is to take that idea and expand upon it; I’ll make my design not just an engineering experiment, but also something pretty enough my partner won’t hate me for hanging on the wall! So yeah, a manta ray. The wings will be the reservoirs (which hopefully will flap), the computer in the “torso”, and even use the cables as the stinger!

To CAD I go, and I end up with this rough mockup:

During the CAD phase, I start looking for hardware. Obviously I need an ITX motherboard, because with the reservoirs I’m already at around 3 feet wide. I don’t want to include a GPU in the watercooling, because I’m afraid it would complicate the watercooling enough it might not work. If I have a big air-cooled graphics card I kind of feel like the concept is a failure. That’s when it hits me… Strix Halo (Ryzen AI Max+) exists, and that is a beast of a platform that doesn’t need a dedicated GPU to game. So I reached out to Framework after talking with them at PAX West, and they have graciously sponsored this build with their Framework Desktop mainboard.

At this point the mainboard has come in, and I’ve been trying to settle on additional hardware. I believe at this point I have, but I’ll leave that update for another day.

[Arkratos]

[Reserved]

[FW4TeePee]

That’s extraordinary! Wow!

[Arkratos]

Sorry for not having a real update right now, I got sick… again… November was not my month basically went over a year without being sick and then get sick twice in a month… and then my work got slammed as we enter the holidays.

My real update will come this weekend but I do want to give an update.

Basically I have the design of the hardware layout pretty close to nailed down. I’m using an HDPlex 500w GaN powersupply and obviously the framework mainboard. However, the first sprinkle of spice is that I’ve been testing the new Intel Arc B50 Pro gpu. I don’t plan to use that to game or really for AI workloads or anything… it is specifically to support gpu acceleration in virtual machines because that little 70w card supports SR-IOV, meaning you can functionally divide it up and pass it through to multiple virtual machines at the same time, which is critical for me as I run linux as my main OS, but need windows for things like Autodesk Fusion to work on this project.

The other spice, that’s probably a little less spicy, is that I want a separate usb controller for each VM that I intend to run. So I’ve found a 4 port PCIe USB card, that has multiple controller and can supply 30 watts on two of the ports and 45 watts on the other two. Totally unnecessary but that does mean I can power even my 24.5 inch portable monitor off the computer.

Pictures to come but the short story is that to run a dgpu off the 4x slot on the framework board you need to use a riser cable that has supplemental power. And you might ask how am I plugging in a second card? well turns out there are M.2 → PCIe 4x risers too (which also need supplemental power because an M.2 only supplies something like 7 watts)

Sorry for the short update.

[Arkratos]

So to elaborate on my last update:

With a lot of help from other framework forum members I found that I could order riser cables from ADT-Link with the option for supplemental power, even if it doesn’t show the option on their website. You can place a note on the order “with a power connector”. The reason this is necessary is that the 4x PCIe connector on the motherboard is closed off and from what can be found online this is basically because if it was open it would need to provide the full 75 watts according to the PCIe spec. With it being closed off it only need to supply 20 something watts, so you will need to inject extra power if you plan to use something like a dGPU. I got the K23SL-TL model.

With a quick test of the GPU riser we are working!

Passing the GPU through to a virtual machine works flawlessly, resize bar is even enabled, which is really lucky because that’s not an option exposed in my bios.

The other fun was an M.2 → Pcie 4x riser, also from ADT-link, also with a power connector… I didn’t take a picture of this but here is the link: https://www.adt.link/product/M43.html
I got the M43UH model. So with a bit of playing around with layouts I settle on using M.2 slot that is on the bottom “front” of the motherboard. You can see it here, note that the rear IO is at the “top” of the picture:

After a few scribbles I settled on more or less this layout, top of the motherboard is the first photo, the underside is the second, IO pointed down in both:

Basically, mount the usb card off the M.2 at the “front” of the motherboard and put the GPU underneath the motherboard like in a sandwich style.

[Arkratos]

Oh and I have a bonus video, the B50 passed through to a VM:

https://www.youtube.com/watch?v=xAoXY1xr01I

I know some of you have already seen this because I posted it elsewhere on the forums, but it’s pretty relevant here.

Running Cyberpunk off the iGPU, then closing that and running Armored Core 6 off the dGPU in windows in a VM. Ignore the slightly lower than normal framerate in cyberpunk (I think it had something to do with running the VM at the same time?) and ignore the stutters when I opened up the VM I don’t think the game functionally liked me alt tabbing into it like that, especially when I only gave the VM 8GB of ram.

[Arkratos]

Sorry for the lack of updates, the combination of the holidays + the start up I work for got me CRAZY busy and then was a casualty of rippling effects of the Meta layoffs. So I guess I have a lot more free time to work on this project now

Well on the bright side, before work fully exploded on me I ordered an Bambu Lab H2C and built a cabinet so I can print ABS/ASA without gassing myself out with toxic fumes. I promise the build wont be “just another 3d printed case” because it wont be. But it will need a lot of 3d printed stuff to get the overall shape of the manta ray, but a lot will need to be made by hand, CNC’d, etc. The H2C does up my ability to do better plastics (ABS, ASA, Polycarbonate) and larger prints, which will be key for the sidewalls of the reservoirs.

Pictures of cabinet:

I should be getting a more substantial update coming in the next week or so. Sorry for the delay everyone, life happens but I’m back now.

[Arkratos]

Hi, sorry for the delays, but here I am with an update!

I 3d printed a basic layout for the motherboard, PSU, and PCIe cards:

With the Intel card underneath:

The bends on the PCIe ribbon cable were a little tight and I didn’t want to kink them so I 3d printed some shims to make sure I have a tight but gradual bend:

I think I need to clean those up more, since the pressure of the cable doing the bend is causing the shim to not sit flush… but that’s not a hard problem.

But I have a bigger problem at the moment:

While the mock up looks good, turns out manta rays aren’t THAT thicc.

Luckily N3rdware has been working on a single slot B50 cooler. It’s been released by now but I managed to get an early prototype… right before everything in my personal life exploded.
(https://n3rdware.com/gpu-coolers/single-slot-intel-arc-pro-b50-cooler)

So where the project is at:

1. base layout is mostly figured out
   a) some things can be tightened up and moved slightly to fit a bit more cleanly in the silhouette, turns out I did my CAD file poorly (I didn’t use components) so moving things will be harder than just remaking everything properly

2. 3d printer set up is good, so I can now print the side walls for the reservoirs, while CNC’ing Acrylic/Polycarbonate and the aluminum for the top and bottom.
   a) Polycarbonate is what multiple people suggested but I didn’t realize it shrinks about a full percent, so that might be an issue when I’m trying to get a perfect seal and screw holes that line up. Tests coming soon… but I might be switching to ABS/ASA, maybe even fiber reinforced if I need to.

3. CNC machine is still under construction due to the spindle upgrade
   a) will need to CNC both sides for both reservoirs and the waterblock pieces.

4. Cables… have ideas for routing and all that, but that will be a lot.

5. still need to work on aesthetics a bit, while I have a layout in mind for the components and watercooling… I’m not sure how I want to fit it together aesthetically. So this is where I’d like to ask you all:

Do we try to go more on the realistic side with windows to showcase components? or do we lean into the mechanized direction? Do we go steampunk-y? Do we cyberpunk-y? Let me know your thoughts!

[FW4TeePee]

@Arkratos

steampunk

Given the relationship between a manta ray and Jules Verne my suggestion would be steampunk-y

This is a phenomenal story @Arkratos I look forward to watching the documentary

[Arkratos]

yeah, that could be cool, I’ll look at more concept art that exists out there for ideas… but I’ll also say, I’m a sucker for this aesthetic: https://www.artstation.com/artwork/XnwN8Y

[FW4TeePee]

@Arkratos

I’m a sucker for this aesthetic

There are no rules to say the two aesthetics have to remain seperated

[eyeon]

Hi,

Do you have an update on the stability of the GPU riser setup? I have ordered the same ADT-LINK adapter with the power cable and plan to use it with a dGPU, but have read online that such adapter cables can cause issues because of the SATA power adapter (melted SATA power cables etc.), do you have any such issues be that instability or high cable temps?

Also, are you able to use the GPU with PCIe Gen 4, or Gen 3?

[Arkratos]

Sorry for the super delay… I’m sure you already have your answer, but it’s been stable and running at PCIe Gen 4 4x.

It’s been a while since my last update, so I wanted to let everyone know where the project stands.

The biggest reason for the slowdown is that I was laid off back in January. During my job search, this build had to take a back seat while I focused on finding my next position. The good news is that I’ve recently started a new job, so I can finally start putting consistent time back into this project. Not to mention my H2C printer has broken itself twice…

That timing is important because I still have the same deadline: QuakeCon, which starts August 6th. My goal is to have the build completed in time to bring it with me.

There has actually been some progress behind the scenes, mostly tinkering between going to meet ups, looking for jobs, and personal coding work to beef up my resume. (The tech job market has been TERRIBLE)

The biggest milestone is that I finally have my Shapeoko 3 CNC cutting aluminum reliably. I added a SuperPID speed controller to the DeWalt router, along with an air blast and a mist lubrication system, and after some preliminary testing I can now machine aluminum successfully. That unlocks a lot of parts that were previously stuck waiting for me to have either the funds or ability to fabricate them.

I also picked up the n3rdware single-slot cooler for the Intel Arc B50 Pro several months ago and have it installed now. I’ll include some pictures below. With that in place, I’ve also more or less finalized the internal hardware layout, which was one of the last major unknowns.

One design change I wanted to mention: the original concept included articulated wings that would flap. I’ve spent a lot of time trying to come up with a mechanism that looked good while being functional, and I haven’t found a solution that I’m happy with. The hinges cause the build to need to be at least 8 inches wider, while causing visual gaps in the aesthetic… With QuakeCon approaching, I’ve decided to cut that feature from the project rather than compromise on the overall build. I’d rather finish a polished version of the core concept than rush a feature that I can’t figure out how to make look good.

Now that life has settled down and the deadline is getting close, expect updates to become much more frequent. Time to start making chips or as I’ve been calling them danger glitter (seriously, aluminum chips off the CNC make awful splinters)

[Arkratos]

So not the most exciting update… well I guess that depends because it was pretty exciting for me… We have ALUMINUM parts!

I’m still working on the CAD files for the new body design, where the wings aren’t hinged. But while I was working on that I ran the CNC machine and did my first real functional metal part!

What is this part? Well it’s to make contact with the VRMs, RAM modules, and CPU.

I will be able to attach my CPU block to the top of it, like so:

I’m hoping to have a CAD update soon with the reservoirs and internal plumbing all CAD’d up this week because I’m hoping to get cutting on the reservoirs this weekend.

Also I finally nailed down my parts list to make the tilting mechanism, nothing too fancy but limit switches, a worm geared motor, an arduino, etc. I’m uncertain about if I’m going to 3d print the gears or need to CNC them. But either way, that’s coming, parts are ordered.

So what’s left?

1. CAD
2. CNC aluminum for wings, CNC acrylic for wings, 3D print mid section for wings… you’ll see.
   a) I also need to Cerakote paint any aluminum that makes contact with the coolant
3. CNC top of waterblock, I’m using the baseplate from an existing block… worst case I can use this block top.
4. CNC distro plate style tubes to reservoirs. This is a bit complicated to explain but with the wings no longer moving, I’m going tube-less and fitting-less. Well I’ll need a fill port but that should be it.
5. Custom Wires (I wont be sleeving to keep them compact so that shouldn’t be that hard… it’s just a 6 pin PCIe, 8 pin CPU, 24 pin, and 2 sata plugs… which I might not even leave as sata because they just go to the risers.
6. Tilt Mechanism

I have 4 weekends left, so wish me luck

[convexproblem]

This looks fun, is there a hole in the plate for the water block to directly cool the die?

I remember from nearly 20 years ago when I messed with liquid cooling that even nickel plated aluminum was a corrosion risk, maybe you need some 2-ethylhexnaoic acid corrosion inhibitor package in the water? (never heard of a car radiator corroding, it must work)

[Arkratos]

The shim just has that indent where the block goes to minimize the thickness over the die. I still need to do some thermal testing… Similar to what NFC did here, but they have a cutout: www.youtube.com/watch?v=Enuv8Ai4yoc

And corrosion is definitely something I’m worried about… anywhere where liquid can make contact with the aluminum needs to be treated. NotFromConcentrate (https://www.youtube.com/watch?v=od6fwgUv0fA) has done some testing and turns out using Cerakote to do a ceramic coating on the aluminum has been working wonders, so that’s my plan.

[convexproblem]

It’s 2C/mm of aluminum alone man. A second thermal paste interface will be even more esp if it’s not sanded very flat. You have no excuse to not do a cutout, you have a mill lol.

I don’t know if Cerakote is bulletproof but you might as well just run an automotive inhibitor to make sure.

[Arkratos]

Are you saying that’s the temperate delta I should expect per mm of thickness? “2C/mm”

My thoughts are that if I do a full cutout then the VRMs and ram modules will get 0 cooling, because I don’t have airflow. Idk, I’ll do a test here soon, but you may be right…

[convexproblem]

Yes, 2C/mm of thickness. I got about 1C/mm when I calculated it for copper, I forgot the exact wattage I used and I didn’t differentiate CPU vs GPU (since the CPU dies have higher power density). Thermal paste will be a lot unless you can get it extremely flat and achieve something like a 10 micron bond line.

The stock cooler is pretty thin, so heat doesn’t conduct that well from the VRM to the distant heatsink. The 4 RAM modules to the top of the APU have no airflow and have only a thin section of copper to the heatsink as well. If you mill a fins into the block over the VRM and the RAM it’ll probably work. The majority of thermal resistance is in the thermal pad and the power density is miniscule so you just need surface area. I have a thermistor around 1cm away from the CPU dies on the bottom of the heatsink for my PWM controller, and it reads about ambient+23 under max load. As such the stock cooler is more of a big heat spreader rather than high surface area heatsink for those parts.

AFAIK the RAM uses a little under 2W/chip at full power (a large % of which will be conducted downward through the motherboard instead of the plastic case), and the VRMs switching losses will generate something like 10W across the entire row of mosfets. If you’re worried about milled fins you can also thermal glue taller heatsinks or a section of metal pipe and run water through them.

I assume the reason for water cooling is to get better performance, so it makes little sense to give all the gains back to the aluminum plate. I have a copper plate so I have the option of soldering everything to it with my stove if I want to water cool, but aluminum is much harder to solder.

[convexproblem]

I got a few little copper fin stacks and some diamond powder today, mixed the diamond powder into some silicone thermal glue, here’s the result:

While running mprime I put my finger on the RAM area, barely warm, over the VRM and ouch it’s hot! I still think best solution is separate heatsinks everywhere.

Update: I didn’t use my thermistors to test before/after but I went to feel the metal today and the part of the plate over the RAM modules is now cool whereas it used to be hot, so the thermal glue and fins are doing something. For the VRM section, the area close to the screw is really hot but I can’t tell what the temperature is where I added the fins.