Just out of curiosity, do you still need those small fans on the Icarus heatsinks? Wouldn't it be sufficient or even better to close the "case" on the top and force the air through the heatsinks? What's the airflow (cfm) of these fans?
In case this is suited better for another thread, let me know.
No, that wouldn't work, the Icarus stock heatsinks are way too small for that and need quite a bit of airflow. Even a huge fan sitting next to them (but blowing from the side is barely sufficient, so at least the second-row boards wouldn't be happy with such a setup.
Sadly you can't easily swap the heatsinks on these boards...
I would think that the turbulence caused by perpendicular airflows would have lesser cooling capability than higher-velocity ducted flow that was relatively linear. So, fans in a push-pull and a cover on the case (clear plexi?) ought to give even better cooling.
That's my design spec - currently building the infrastructure, waiting for 25 ZTEX 1.15x FPGAs.
Top (visible) board will be more or less square and have five columns with 5 boards each per column. Each column will have an 80mm fan pushing air in from the bottom and another at the top sucking air out. The main board will be mounted on a 45˚ incline to show off and also let physics help the fans out. Each set of 5 fans (blowers and suckers) will be powered from the main PSU on a separate 12V circuit with variable resistor, so if the cooling is more than adequate, I can reduce noise by turning down the fans.
Each column will be separated from its adjacent neighbour by an acrylic (perspex / whatever) panel. I had the same idea as nyana regarding drilling holes in the visible mounting board to run the USB and power cable through - I'll probably paint the base board copper or something. Underneath the base board, the USB hubs and power daisychains will be hidden with another board (false floor), so the only visible cabling will be one USB cable (for the entire farm), and five 2.1/5.5mm female sockets for FPGA power. The variable resistors for the fan circuits will probably be mounted on the side of the main board.
The board will then look like 5 'tunnels' with a fan at each end, separated by perspex. I haven't yet worked out the gasketing design yet (since the top panel will need to be easily removed for maintenance) but the top panel will be clear acrylic too. I'm hoping to equip each of the 25 FPGAs with blue Zalman flower northbridge coolers, as they look fantastic... but have only managed to buy 7 so far, and an order of 18 off Amazon got cancelled by the vendor after messing me around for 2 weeks.
This will be geek eye-candy, but also I've learned from GPU mining that having near-vertical airflow works well, so the board will be mounted at 45˚ and the space behind the board used to store the ATX PSU (PCIe cables will feed 12VDC to the FPGAs using custom cables, no butchering of the ATX PSU cables, and a plain Molex connector will suffice to feed the fans), and the Mac netbook powering the system (it's a Dell Mini 10v, running Snow Leopard and my build of the ZTEX SDK - the little netbook was chosen as it consumes low power, and running OS X allows me to easily view the screen from my other Macs).
Should be neat and tidy, hopefully should run with passive coolers and *hopefully* not too loud. Pics when I'm up and running - I've got all the bits I need apart from the FPGAs and 18 of the Zalman heatsinks, and am building the assembly now. Hopefully, when the FPGAs arrive, all I'll need to do is screw them into the mounting posts already built into the top board, attach USB and power cables, drop the perspex lid on to create the 'wind tunnel' cooling, program the FPGAs and kick off the miner.
If it works well without any snags then I'll duplicate it when I get funds for the next load. I reckon a 600W PSU ought to be more than enough and should deliver clean enough power so the FPGA power conversion stages don't have to do much work... I've got Cooler Master PSUs, my only concern is not using anything other than the 12V rail (which I'll start another thread about)...