Test board for Avalon chips.

Guys we are going to upset a few of you that are keen to go forward with CM3/4 and I apologise on behalf of Enterpoint for that.

We have temporarily lost one of our key Bitcoin team members to illness and this is very likely to impact our timescales. As a result we are not going to progress orders until we have a clearer view of the impact. This won't affect our ability to manufacture directly but there may be delays due to designs being later than planned and that is of course important to you guys. Our main problem area is the Controller software and to a small extent FPGA development as it is tied to when aspects of the software are ready. We do have a backup plan but that doesn't give all the hosting features directly and we feel that is a major loss to the design spec. We are also going to look for external resource to solve the software issue but dropping someone into the hot seat doesn't always work.

It is going to take us a few days before we have a clear view on all of this so please bear with us whilst we sort it out.

Yes we are taking Bitcoins.

The short answer is we don't have to be successful in selling kit as such our own mining activities with these products will make us much more than making and selling hardware. Bitcoin is also not our main business so also a reason not to offer silly prices. No point doing it if we don't make money or worse make a loss. Kit like this does cost more to make than simple small boards and you have the choice to go with what you think works best. What we are doing is simply opening what we are doing it up to other people.

You might disagree but there trends from both Avalon and BFL towards dearer hardware I might suggest that they don't make money making the hardware at the what I will call the silly prices they have been offering. On the surface they are both starting to look more like semiconductor manufacturers than equipment producers. The newer entrants to the market I might suggest that they have not costed things properly and will end up making a loss and ultimate will probably go out of business. It is easy to get caught in the glitz of Bitcoin and not base what you are doing as a business on hard logic and analysis.

You might also ask why Avalon is supposedly mining with equipment before shipping. I would suggest that is so they actual make money and not lose money on equipment sales. Same comment has been made about BFL in the past.

That is a close enough number.

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.

There are rack prices and individual board prices. We annoucnced initially a whole rack price with 8 CM3s. Each CM3 has 144 Avalon chips which if Avalon datasheet is correct gives 40,608 MH/s per board. Until we have a full rack system up and running we won't have exact performance numbers.

Prices don't include VAT if that applies. VAT usually only applies to EU customers.

The CM3 will need a Controller fitted (DIMM socket - master or slave parts as appropriate) as well as 12V, 3.3V and 5V applied. Internet can plug in at Ethernet.

The complete price does include if we supply the Avalon chip. However we only have a very limited supply so it really needs customer supply in the short term unless we are able to buy any from people that have them. That is partly why CM4 is launching at the same time as we have a better chip supply for it.  

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.

What should be the final CAD image below for cairnsmore3. We have a test module being built up this week to allow us to test the Avalon chips before we press the button on manufacturing the Cairnsmore3 prototypes.

Email is usually the fastest as I don't sit all the time looking at the forum.

Exactly CM4 will be there August.

They are not desperate enough yet. We have different things in mind for CM5 and CM6 and more on that when we are ready. They might make the CM7 slot but there is serious competition for that position.

BFL do kind of have a desperate look these days. Look's like dutch auction time to me.

That is only if it arrives before August of course and in one piece. So far we have always pretty much kept our delivery promises and not slipped 3 months or as some suppliers nearly 1 year on delivering a promised product. We have set our prices and if people don't like them they don't buy. That actually doesn't matter as we can probably make more out adding anything spare from this build into our mining rig than straight selling kit so for us it is as simple as that.

I could put this in context that as far as I understand batch3 from Avalon are charging 75 BTC for 66GH/s in their batch3 which I believe is running late and no ETA. On todays conversion that is about £73/GH/s so is that so very different from our £80/GH/s? If it doesn't sell we will simply put it into our own mining operation.

We are also not going to try and compete with manufacturers that have little or no standing cost base and might not be around in 3 months when you have a problem.

I might also speculate that BFL has problems because they charged too little for their miners and their 2.5X recent-ish price rise is an indicator of that. Avalon too have also raised prices through batch 1 to 3 presumably for similar reasons. We have been in business now for 24 years and I am look forward to celebrating 25 still in reasonable financial health and that means charging realistic prices as far as we are concerned.

Costs can be argued about what is good or bad. I have not been following other Avalon based designs in any depth so anything I say is based on limited knowledge. I think the Burnin boards are using a USB interface and that is one we wanted to get away from. It's not great for a big design as we found out with CM1. When you get to 256 COM ports you are stuffed in Windows and maybe Linux too. The CM3 design will have a controller running Linux and miner software so doesn't need a host as such with those issues. We do have a USB interface but that is for special cases or maintainance.

The K384 I don't know much at all so can't make a comment on what they are doing.

With Avalon chips there are some particular problems with offering warranties. I'm not saying this as any insult or other bad mouthing of what Avalon have done but wearing my pro hat they are a supplier with next to no track record and barely have any normal business presence or support. Now if you said to most companies like us they wouldn't touch those chips with a 100 mile bargepole. It fails all the business due diligence tests that are normal. So what does that mean in practicality? It means that you might have poor quality in the design or manufacturing. It might also mean non-working chips arriving. It might mean chips don't appear at all. As a business that represents a huge risk in warranty and litigation. It might also mean a lot of support work and costs. Probably worst of all it could be pissed off customers. I certainly don't want Enterpoint being described in the same BFL is currently. So that is the context we set our prices for CM3. Remember also that CM3 is mainly being done as a service to customers because they asked for it and we can do it in tandem with CM4.

ROI is always going to a difficult one. ROI also changes depending on whether you consider it in flat or BTC. A year ago people talked of an exchange rate that might some day exceed $100/coin. We have been 2.5X that over the last few months. That change has made a big difference in flat profit levels. If we went back to $10/coin flat profits might not be a profit.

In BTC you might say 1/10 of last years BTC earnings is now crap. At best Bitcoin mining is a gamble and you back you horse i.e. equipment to make you money. So in context in buying our equipment we think most customers have, or will, make money and hopefully that will continue to be the case.

To put this into context this is a solution for more less now and we do see CM3 and CM4 being superceeded by CM5 and later "engine" boards. In Bitcoin product lifetimes are always going to be very short and the design we are doing is influenced by that. 90% the work that goes into the early CM3/4 systems will be reused in CM5 onwards and it is entirely possible in time that a single rack could even be comprised of 1 each of CM3, CM4, CM5, CM6, CM7, CM8, CM9 and CM10 to be extreme. That forward progression is our big design aim and the underlying technology can change very quickly to the latest and still work with older elements of a rig. With firmware updates the controllers we supply should be able to run with the newer technology as well. Eventually the standard may have to change say because of some restriction in I/O bandwidth but when we get to that point we will attempt to support a migration usage for "old" boards. That might be a faster Controller or one that has say 10G Ethernet. Whatever the challenge is we will try and bring the older boards along with the new designs.

If you are going to try and fine tune to use 100% the power supply of a DC you are likely to generate costs much more elsewhere in your rig in some other inefficiency. Some would argue that the DC itself is an unnecessary cost and that is why we want to make it capabile of running at home. We also think the potential for heating homes is much more useful than convenience maybe of a DC. It is certainly slightly greener. Also if you are in the camp argueing for hashing to be distributed a lot more then DCs are bad news. You could have 10000 miners basically in same building and they don't not know that. Lose that one building by whatever reason and you lose a big amount of the network. In our own DC plan we have done some thinking about all of these aspects.

It is commercially too sensitive to for this level of detail to be revealed in open disscussion at this point. Outside of the performance offered, unit power, and what the cost is to customers that won't ever be discussed by us in this forum before customers have units in their hands. However you are free to speculate as much as you like.

Ok so what could be done is that systems are adjusted to fit a DC capabilities. So the obvious way is to fit less cards per system. So 10A at 120V is only 1200W and that is assuming a good power factor. We are maybe looking at each CM3 card taking about 400W so 2 or 3 cards per power input might make sense. If a Goliath rack was fitted with 4 PSUs you could power 8 boards taking 4 x 10A feeds you have available. If you really want to push the space then 1.5 racks is just about possible or stay comfortable at 8 boards or 1 full rack.

The big difference with Goliath is that it is designed to go big and it has a good packing density compared to K16.  You also get the pain of manufacturing removed by paying us to do that in the cost. When you start working at these sizes it becomes a very different design and manufacture problem compared to a K16. That is not in any way being critical of K16 and it has it's place in the mining field.

Over the last 16 months that we have been involved in Bitcoin mining we have learned a lot and I think we still are learning more. We did CM1 last year and that has proved to be a very good product, popular with customers, and with excellent reliability except when placed in torrent of water. Outside of the early phase deliveries the number that have come back to see us on warranty I can count on one hand. That product whilst very simple has taught us a lot in what needed to be done in the Goliath Miner. CM2 which nobody outside of Enterpoint has seen in the flesh has also helped in furthering our ideas and developments. Another little known secret is that we have our own mining software and that might be cut into these miners if we think that is the best way to go. That may offer product stability over third party builds that we have no control over "updates". We will talk more about this aspect as we integrate the design in July.