|
I'm not sure that question has an answer. The original Compac didn't have a "target" hashrate, that's the whole point of making it with adjustable voltage and frequency. Set it where you want it.
I can say that for the ~10W it took the one-chip to generate 20GH, this one should get closer to 30. The 5W hashrate, instead of around 13GH, should be more like 18.
|
|
|
|
|
Hadn't really thought about it. I guess Compac SE is pretty boring compared to 2pac.
|
|
|
|
|
On that note, I'm starting to lay out a two-BM1384 Compac. It should have the same overall dimensions as the existing Compac. The heatsink is the same size, so super-high-end overclocking won't be as safe without modding, but that's okay. You should get a lot more hashrate out of it than the old Compac at any given power level.
To make a batch worthwhile will probably require at least 100, preferably 250 or more.
Not sure on price yet, wait until I get the layout done and see whatall parts will be needed.
|
|
|
|
|
So, question.
How many folks would still want BM1384 Compacs, if I decided to run out another limited batch to help raise money?
How many folks would want a two-chip BM1384 Compac (much-improved low-end efficiency and hashrate), or say a small 4-chip that took a barrel jack and USB. These guys would still have the same base features of the existing Compac - manual voltage control, etc.
The point is, something that can develop quickly, build relatively cheaply, be something people might want and can raise money to put toward better things. A lot of people have chipped in donations, but wouldn't it be nice to buy something instead?
|
|
|
|
|
But it would be nice if they could get back to me about picking up another hundred or so samples, cash up front. The last reply I got from Punin was September 27th and I've only emailed three times since then with no responses yet. Might take a bit more badgering. Right now I can prototype one pod and two sticks; with another 100 ASICs and 15 comms I could prototype a TypeZero board, five more pods and another 7 sticks. That'd give me a lot of stuff to play with and some to send out for y'all to play with.
|
|
|
|
|
It would be incredible if I could get chips on some kind of net-30 or thereabouts. Chips will easily be the largest single expense; for something like the Compac they'll likely be around half the total material cost but for pods and bigger it's more likely 80%.
I asked Punin a month ago what it would take to buy another 100 or so sample ASICs and more of the comm chips so I could build more than one working prototype of everything. I have enough on hand to make one pod and two sticks and that's it.
Yes, the big boards would fit right up to S1/S3/S5/C1 sinks. Likely it'll be a double-sided board with ASICs on the backside like the S5; since we'll be operating with three independent strings there'll be a buttload of distinct "rails" on the board so PCB cooling isn't really safe, which is why S5 and Avalon6 both did top coolers. Unlike at least one other S5 refit kit, I won't require you to have a working (or modified in whatever way) S5 controller; the boards will run on USB so you can tether 'em off a single controller of your own choosing, kinda like Avalons but not quite. And the PCIe jacks will be placed so S3 cases will still work.
|
|
|
|
|
Between the two of you and the cores I already have, that's about 12K chips. Heck, we should really figure out a way to get Bitfury's attention since Punin hasn't responded to my last several inquiries.
|
|
|
|
|
If you're looking for, your subject line should probably say "WTB" - Want To Buy, instead of "WTS" - Want To Sell.
|
|
|
|
|
I think it's a lot more likely that I'll go bankrupt before BitFury does. PSU sales are down, I don't have a new product to roll out unless these miners come to life, hosting is barely paying the rent (and it was never intended to pay for anything else) so really if there isn't good news (and probably some loans) in the near future I could be packing it in by year's end. BitFury is likely to last longer than that.
The pod is designed with a single PCIe jack and a standard barrel plug so it could run off a full PSU or a brick. I'm giving it power monitoring (input current and voltage) so if you're on a watt-limited brick you can watch the power draw and make sure you're not burning something up. Based off what numbers I've seen, it should probably draw in the range of 35 to 100 watts depending on user setting, with stock around 75W for 700GH. I hope to build them with a heatsink and fan already installed. Target footprint is 8x12cm, with 8x8 as heatsink and fan so a standard 2- or 3-wire case fan can be used.
I would really like to be able to tell Bitfury we had a guaranteed order of 10,000 chips. That'd probably get someone's attention. More than 3500 Compacs were made and sold between mine and bitshopper.de's licensed design, so another 3000 sticks would mean 6000 chips. 500 pods would mean another 5500 chips for 11.5K chips total. That's likely at least a $50-70K order.
Back when the first word of Bitfury's 16nm chips surfaced, Punin stated that I'd be getting some (promise made good in August) and he also stated he intended to reopen a webstore from which the MOQ would be one chip, because he wanted to support community endeavors. I'm really hoping at least part of that happens. We need to make sure Bitfury stays aware of just how big their sales could be for this project, and just how good the PR would be for them being the only outfit to openly cooperate with the mining community's hardware devs in almost two years.
|
|
|
|
|
There was an S3++ at one time, limited distribution but they were testing the S5 string concept with S3 chips and already had an S3+ model.
|
|
|
|
|
Price will depend a lot on what Bitfury charges for chips; release date will depend on supply of chips and also how quickly we can dev. I'd be willing to start working on manufacture as soon as we prove basic functionality, because there'd be a month or so of lead time on big stuff (PCBs, etc) to finesse the code.
Everything is software-controlled voltage.
|
|
|
|
|
I've pretty well finished an 11-chip pod layout, and have most of a 2-chip Compac layout done. The controls setup is pretty much finished, anyway. Today I sent off for prototype PCBs of a modified 2-chip using the Compac as a base, but expanded for additional testing. I'll finish the Compac PCB based off any changes I have to make after testing this board.
The Compac will run 2 chips, and I expect around 20-25GH at stock 2.5W setting but a practical peak above 100GH, possibly closer to 200GH for extreme extreme overclocking (extensive modified cooling, etc).
The 11-chip pod would stock about 700GH from around 75W DC, with room for both over- and under-clocking. I expect a bottom-end around 450GH for 35W but I guess we'll find out.
Plans for the TypeZero (S1 refit) right now are basically three of the pod circuit on a single board, so 33 chips and 2TH 200W stock per board, with room up and down. The nifty thing about doing it that way is, each string would have independent voltage adjustment, fuse and master disable, plus a temperature sensor local to each string. This allows for a lot of control and flexibility, plus some fault tolerance.
Just because I can, I'm gonna plug the project donation address 1BURGERAXHH6Yi6LRybRJK7ybEm5m5HwTr. Prototyping ain't free, and I really want this project to succeed. The only things needed for that are funding to keep it going, and Bitfury to get us chips, and I can't really do either of those things alone.
|
|
|
|
|
I can do a 1200W server PSU with breakout board and 10 18" PCIe cables (16AWG wire) for $90 shipped within the US.
|
|
|
|
|
Current bid is $17.50, which is honestly kinda impressive. Maybe it's a collector?
For reference on age, the current generation of ASIC miners will do the same amount of work for 1% the power draw. There now exist single chips which will do the work of that entire ten-blade set.
Also, there's an error in the listing concerning the hashrate. The BE100 ASIC used on the ASICMiner blade has 28 cores and operates at a stock 12MHz for an effective hashrate of 336MH/s. Each blade has 32 chips operating in parallel for a blade maximum hashrate of 10752MH/s, or 10.75GH/s. A set of ten would run 107GH, not 1070GH.
|
|
|
|
|
Actually what I was getting at is, no intrinsic voltage control and a nominal core of 0.45V unless you can adjust at the PSU. So, like the S5, Avalon6 and 162-chip S7, undervolting is difficult and not an inherent feature.
|
|
|
|
|
Yep, working on mine right now. I didn't really glean anything from the text since I can't read Russian, but there was some valuable data in the pictures. I now have a better idea of the performance points of the Bitfury chip, since I haven't been able to test hardware yet. One of the drawbacks to being a solo operation, with product manufacture and maintaining a 100KW data center on top of design work, is that design can't always get done quickly.
|
|
|
|
|
More 26-chip unregulated strings, same as the other thread.
|
|
|
|
|
If they daisychain like the Avalon4/6 you could buy one controller for like 50 units if you had USB hubs and such.
|
|
|
|
|
Show Posts
