... We have priced out metal core boards and that I think added about $3 to $5 more per board. So if that translates into higher hash rates for longer then we think that is reasonable cost as we do not want to do be doing RMA work for chip failures...
From a consumer point of view I would bet that additional cost would be insignificant, even when marked up, when you consider the cost of the devices and even the cost of carriage is likely to be substantially more than that upgrade.
It would be interesting, once the project gets a volume of hardware to market and given that you're looking at specifying a more industrial grade of product. If the ASIC fabricators would be willing to formalise a package-form to allow socketing rather than needing complete new blades. But that's quite far down the line I suppose.
At the moment, from the early concepts, I have this mental image of each blade being like a Pentium II/III processors (if you remember those), where it stands off from the motherboard and is pretty much a heatsink cartridge around a small board with chip mounted. Might be worth seeing how they handled these concerns? I know you've probably already considered this because you've said previously that you want the wasps/blades will be able to run independent of the hive, and having the chips as plugins themselves would take more of the control work off the blade and back to the hive which would mean they don't run independently.
I know everyone has an opinion and not everyone is an EE, so take these suggestions with a pinch of salt, I'm just finding the discussion very informative and the project fascinating.
For us in terms of re-doing the wheel each new chip that comes down the pipe we are fine with that variation given our design concept we don't need a lot of component changes just a reworking of the space around the footprint of the chip. We know that ASIC fabricators won't get on board with a single socket design that be too easy right? And the design is collective effort and has had back and forth on a lot of these issues to finally arrive at the current design. We are always open to ideas though that could benefit the project overall. One of the EE's was involved in the very first Intel chip designs so yes I am sure he is familiar with all of those sort of solutions.
In terms of the Hive it is really only there to push the power to the Wasp and provide consolidated data transfer it is a very basic modular set up. In terms of loading more on the Hive that is possible but we will have to see how this all shakes out after the first runs. We benefit from having a modular Hive in that cooling could be modular as well. Be it a unit inclusive of the Hive or individualized cooling for each Wasp. Personally I am partial to a
http://www.iceotope.com/ configuration.
But considering the cost involved a immersion bath system like the guys at Asicminer in HK could be a better solution.
http://www.youtube.com/watch?v=oZavKweMrP4 from Allied Control.
Then again maybe the Minion and the BitFury Wasps could "overclock" in air if you have the right fan heatsink PCB combo. Who knows... but there are quite a few of us in the WPC that are itching to get our hands on some of the Wasps to see what "aftermarket" cooling we can do to them.
A newer version of the module Hive will be out when the Wasps are ready. We are also really keen to work with fabricators on casing / cooling so maybe someone really wants to work on that in tandem with us.
