No you're right. I originally thought that SFX was going to work, but once I got everything drawn in CAD it didn't fit because the GPU needs to be located smack in the middle of the cylinder to provide the clearance I need to plug in the aux power on one side and the PCIe riser on the other. Then I considered a TFX form factor (since I have one in my spare parts closet), but the width wasn't the problem, I needed something with less height (in the conventional mounting orientation). The PSU is now a Silverstone SSF-FX350G.
I'm not terribly excited about the tiny 40mm fan, but it's an interesting new product to try out. This thing is smaller than some gaming laptop's power bricks and produces 350W
How's that for power density? The PSU fan will probably end up exhausting downward toward the mobo (unless reversing the fan produces no ill effects), which I realize isn't ideal, but it's probably going to have to be that way so that I can plug everything in with minimal wiring clutter. I plan on shortening the PSU cables to only what's needed (that will also be a fun new experience).
That PSU will be fine. Assuming worst-case figures all around, the efficiency chart is probably for 230V and that it's closer to 85% on 110V, then assume that your typical peak load is going to be 250W and you're looking at only 250W * 0.15 of heat loss in the PSU total, which is still under 40W and that a good fraction of that heat will be transferred to the vast surface area of the PSU's housing anyway. Under more typical loads of maybe 175W and with more optimism on the efficiency graph, it's under 20W.
At this point I'm planning on venting the top as much as possible, and I'll also vent/slot the IO panel cover on the mobo to allow some heat from the CPU to escape there. My hot box only had the existing openings in the top, but I can at least drill 2 additional ~1/2" holes beside the pressure gage without anyone noticing because the shroud shadows that area pretty well. With all those extra holes, I should almost double the 20CFM that my hot box tested for. Try as I might to seal it up, I'm sure I'll incur some additional airflow losses around the mobo IO panel protrusion, which will allow some additional CPU airflow out the back. However, I don't want to let too much out near the IO panel for fear that would stagnate/starve the airflow past the GPU.
To circle back to my earlier air velocity estimation, the case volume is only 0.254 Cubic Feet so even if I only get 25CFM, that means (theoretically) every bit of air spends less than 3/4 of a second inside the case.
And yes, the GPU has to be mounted with the display ports facing down for cable routing, but also because the expansion bracket is the only thing I can use to bolt the GPU to the component frame.
Your proposed layout may actually be fine because of the chimney effect with a single tube and a powerful fan at the bottom. Theory and practice don't often agree in fluid mechanics, not without access to very pricey modelling software! I certainly wouldn't expect to get the same CFM through the case with obstructions in it but I still think that it'll be fine, though possible a little noisy.
If it were my project the only thing that would concern me a little is the sheer amount of hot exhaust directed towards the centre area; PSU fan, all 95W from the CPU, at least half of the 120W from the GPU, let's call it 30W from the PSU and you can probably add another 15W for everything else on the board to call it an 'even 200W' hotspot. Sadly I don't think reversing the PSU fan is going to be effective - it's a shrouded-intake, negative-pressure design so the coolest air is induced at relatively high-speed around all the primary heatsinks and then dissipates more slowly internally to less important heatsinks in the PSU. If you flip the fan and do nothing, the fan is going to have to work MUCH harder because the air reaching the primary heatsinks near what has become the exhaust outlet is now higher-pressure, slower-moving air that is already pre-warmed. Rather than flipping the fan around in the PSU, I'd prefer to find a way to flip the whole PSU around if possible.
One alternative for you to think about is how much deviation from the cylinder shape are you willing to tolerate at the back, out of sight?
You've already made a 1.75" slot at the bottom for the motherboard tray, and you obviously need more cutout for cable-routing. If you were to enlarge that slot by half an inch in width and then extend it upwards a bit you could easily squeeze a couple of decent-quality 60x25mm fans in there. Go up to a 3" slot and 80mm fans become an option. If you were willing to do that you have a lot more options available to you for experimentation:
- Intakes at top and bottom, exhausts next to the hot zone in the centre at the rear,
- dual-chamber design with 120mm dedicated to the motherboard and a couple of smaller fans handling the top chamber, or given that there's empty space between the cylinder wall and the GPU you could probably put a couple of 80mm fans in there.
- chimney-effect assist by recessing extra fan(s) in the gap between the cylinder wall and the GPU.
Personally I think the simplest alteration that would improve cooling to all components would be to flip the PSU around in its entirety, the additional benefit being that you can MASSIVELY shorten the cables to remove a significant portion of the airflow resistance they cause. You'll obviously need to do something funky with right-angle IEC C14/C15 cables but you'd need to do that whatever orientation it's in, anyway