and to top it off I made two of them.
I decided to go with GPU rather than with the non existent ASIC. I see it as deciding to buy a v-8 sports car for $100k rather than wait for the $1k atomic powered flying sports car to be finished (like with BFL).
You know that all the new bitcoind roads are designed for the atomic powered flying sports car and that your $100k v-8 sports car will have to remain parked in your garage?
I work with both clusters and co-location, so I got a good deal on a massive amount of hardware and I got a good deal on coloco (air kept at 66F all day plus a staff).
My rigs are a cluster
Why a cluster? Mining bitcoins is a parallel task: you don't need any failover or load balancing tech.
of 4 nodes each and one embedded *BSD licensed controller of my own design. Each one has 8 8 core cpus, 512GB of RAM
High-end CPUs and lots of RAM? For mining? Seriously if you want to throw money away you might as well give it to me: I could invest it in hardware that would produce some income and not generate electric bills like there's no tomorrow...
and 12 GPU cards with logging and storage done in an encrypted VIA Mini ITX controlling a storage array (so an error won't crash the system by filling up system disks).
I use USB keys for each of my "nodes". On some of the first USB keys filesystems were remounted read-only on medium errors: the mining process was undisturbed...
The other rig is exactly the same, located in a different location with the same colo company and has hot failover capability with the other machine using something Linux HA heartbeat. So if one rig fails it takes over the tasks of the other rig as its virtual instances switch tasks (why I needed so much RAM).
I really think I spent too much for this ($60k for both. if I had not got the deal I did I would have paid about $90k) and tonight or tomorrow I should be able to start mining once it is set up at my colo.
You think so? You have a grand total of 12*4*2 = 96 GPU. If these are 7970 you may have a total of 96 * 700MH/s (overclocked) = 67.2GH/s for $60k.
The fact that you have 12 GPUs/node will certainly be a problem (there's a limitation of 8 GPUs/server in the AMD drivers).
So you will probably be restricted to 64 GPUs -> 44.8GH/s (still for $60k). Being in a datacenter, power won't come cheap: even if you manage to put these 44.8GH/s online you may not even make a dime mining Bitcoins (and unless the BTC price rise to absurd levels you certainly will begin to lose money in one or two month).
For a point of reference I spent ~$7k for 12GH/s one year ago: your hardware is 3x less dollar efficient than mine one year later. Mining is about optimizing the cost of hardware and the power usage, your setup is simply over-engineered.