We have had a lot of enquiries about building these systems up piecemeal so here is something of an initial price list.
Where Avalon chips are supplied by customer to us there will be a discount of £8 per chip applied.
We are also looking at a discount structure for individual boards.
Errhm... I'm having trouble understanding what I need. If I'd want to put together the smallest/cheapest
CM4-miner, what exactly would I need to order?
CM4 board obviously.
Master controller, I think.
Full rack , I think.
How many ATX power supplies and power distribution panels?
Then if I'd like to expand with one more CM4 board:
CM4 board (obviously)
Controller pass through and master-slave ribbon cable, I think.
How many ATX power supplies and power distribution panels?
Then how many times can I add one CM4 board until the rack is full?
How much power will one system with one CM4 board use?
There are a few optional items that we have not costed as yet the main ones being add-on fans for single boards, water cooling options, small cases etc.
The slave Controller will have a small FPGA mainly used as buffering. The initial master controller will have a bought in ARM A9 module paired with a FPGA for buffering and functional acceleration. It is possible poth CM3/4 could be used as USB style boards if Controller firmware is suitable. We won't promise that as an initial function but it is on our list of features we want.
The number of CM4s in a rack is primarily set by the height of the Controller that sits in a vertical DIMM socket and the cooling solution. At the moment we are targeting 8 boards in a rack but 9 or 10 might be possible. More details on that as we get the design pulled together. The initial cooling design will use the same heatsinks found on CM1 but water cooling is the target for high density racks later on.
The connection of boards is a bit like what we did on CM1 except we have a lot more wires to play with and we can run LVDS as well. The daisy chain will really only be limited by the Controller processing power and as yet we don't have good enough numbers other than we think we are ok for 8 boards. Because the controller can be changed we have got a planned "upgrade" path where higher performance controllers could be designed and fitted later if needed. Obviously we could also use more master controllers as well if needed in a rack.
Ok minimum startup could be using a 1 x CM4 outside a rack like CM1. We might be even able to stack the same way. That is TBC. You will need a Controller and power supply. Almost certainly an ATX PSU as inital power budget per card is 60A at 12V. That may vary with the performance delivered so may be a little less or a little more. These numbers are interpolations from Cairnsmore2 testing. CM4 will have its 12V segmented into 2 or 3 sections to help with how PSU get wired up to the boards and shared between boards.
We are in negotiation for PSU supply but the option will be there for customers to reuse their own or buy locally to keep freight cost down. We may end up with a higher current PSU and less of them in a rack but that element is still in design so not finally fixed.
We will talk a bit more about the detail of CM4 nearer the release. The main thing is that we will guarantee the price per GH/s Bitcoin hashing delivered. That gives us some room to vary how we implement that hashing power. We know everyone wants to know the detail but we are still sorting out a number of things commercially and design wise so please bear with us on the CM4. CM3 is much more straight forward but we still have a little development there to get it all going. It obviously will only do Bitcoin mining and that is both a plus and a negative and hence a reason for doing CM4. Going beyond all of these we hope to expand what we do in these systems with CM5 and CM6 and so on but that is for after August.