GekkoScience BM1384 Project Development Discussion

[sidehack]

You might want to shift it over to the support thread though.

[QuintLeo]

To me DOS stole more from CP/M & MP/M than the PDP 11 OS.

 To be picky, DOS was a (initially ripped-off later paid for) CP/M varient, based on "Seattle OS" that was a licenced CP/M "clone" written for the 8086/8088.
 CP/M however took it's interface almost directly from RT-11, even maintaining the "PIP is BACKWARDS" command structure among other obvious tipoffs.

 So the RT-11 to DOS steal was indirect, not direct.

Much of the NT design team was "lured" away from DEC

 If I remember the timing correctly, it was more that DEC was dying/had just died and they needed a place to go to work.
 Similar reason and timing on how Intel ended up with the Alpha IP (bought out of DEC bankrupcy IIRC), which they used as part of the input into designing the Pentium.



 Gotta wonder when sidehack is gonna change the name of this thread (if possible) from "BM1384" to "BM1385"....

[sidehack]

Or something else. Bitmain still hasn't budged about sample chips. I just tried again a few days ago, even asked about old BM1384 stock but I got nothing again. There are still other options.

[ATCkit]

Or something else. Bitmain still hasn't budged about sample chips. I just tried again a few days ago, even asked about old BM1384 stock but I got nothing again. There are still other options.

Maybe Spondoolies would sell you some? Their new rig is a beast only for industrial operations- but seems to have more efficient chips (0.15W/GH/s)  :

http://www.spondoolies-tech.com/products/sp50

Sure would be nice to have smaller miners with that kind of efficiency.

[sidehack]

Now don't tell nobody, but I'm talking with them about getting a few samples to play with. I was surprised but by no means unhappy to see they shifted to a smaller chip that's also stringable. If I'm working the numbers right, it's around 20W which puts it more in the A1 range than the 10W Bitmain chips we're used to, but the package should also cool more efficiently. Posted specs indicate it's a bump die of some kind, which will probably be more annoying to work with than something QFN (for routing, which depends on the actual pinout), but not nearly as bad as a big fat BGA like the Rockerbox was, or like SFARDS and 21e6 are using. All in all it's a pretty encouraging part.

[Mudbankkeith]

133Gh per chip

A 4amp 20w stick miner?  in time for santa's flypast?

[VirosaGITS]

133Gh per chip

A 4amp 20w stick miner?  in time for santa's flypast?

That sound pretty crazy, at 0.15 10w would be about 65GH. Maybe you could do 20w with proper cooling.

It doesn't sound very doable in term of powering and cooling it efficiently for a USB stick, maybe for a pod design where you have a couple of chips and you cool it off with a CPU heatsink?

[sidehack]

That's gonna depend on a lot of things coming together in a short amount of time. It took a lot longer to get the Compac ready than it probably should have, but a lot of that lost time was power issues. The buck design right now is pretty solid and wouldn't take much tweaking to power that chip at least moderately well - probably not all the way to the top, since that's close to a 40A draw and I really don't want to see that handled with a single phase. I could see 10W delivered to the chip, which would probably be about a 12W overall for the stick - for reference, we've pushed Compacs up to 15W input power but that was at a higher Vcore so the core current was lower and that's the real limiting factor for a regulator design.

Any dev this time would be much more heavily focused on software. I don't know a thing about Spondoolies' protocol except that it's both nonstandard and complex. Their rack gear uses an FPGA to translate between the controller and the chips, but Technobit implemented it in a micro on some of their Rockerbox things last year so I know that's possible. I'm sure it'd take a lot longer for Novak to figure out the code than it'll take for me to redesign the hardware for a stick.

There's still a whole lot of questions up in the air. All I'll say is it's not impossible.

I've been rolling the idea of a 4-chip CPU cooler pod for a couple days and so far I'm liking it. There's a few different options for how to flesh it out, so until we crunch the numbers on the best base design I won't know anything for sure. Again, it's not impossible.

If Spondoolies cooperates with sample chips (and they've historically been better at that than probably everyone except ASICMiner, may they rest in peace) the only real limiting factor will be dev capital. We've got a plan to free up dev time in the next few weeks, but having time doesn't always mean having money. All our income is based on sales, right now that's between Compac sticks and PSU hardware. Hosting is basically a breakeven concern and mining revenue accounts for maybe a hundred bucks a month. We run a tight ship and can operate pretty lean compared to a lot of shops, but the level of leanness we're running right now isn't really sustainable.

[Mudbankkeith]

I tried one of these on a stick last year
[https://www.dropbox.com/s/vdgpz6m9d45z2cg/2013-12-21%2021.56.26.jpg?dl=0]

[VirosaGITS]

That's gonna depend on a lot of things coming together in a short amount of time. It took a lot longer to get the Compac ready than it probably should have, but a lot of that lost time was power issues. The buck design right now is pretty solid and wouldn't take much tweaking to power that chip at least moderately well - probably not all the way to the top, since that's close to a 40A draw and I really don't want to see that handled with a single phase. I could see 10W delivered to the chip, which would probably be about a 12W overall for the stick - for reference, we've pushed Compacs up to 15W input power but that was at a higher Vcore so the core current was lower and that's the real limiting factor for a regulator design.

Any dev this time would be much more heavily focused on software. I don't know a thing about Spondoolies' protocol except that it's both nonstandard and complex. Their rack gear uses an FPGA to translate between the controller and the chips, but Technobit implemented it in a micro on some of their Rockerbox things last year so I know that's possible. I'm sure it'd take a lot longer for Novak to figure out the code than it'll take for me to redesign the hardware for a stick.

There's still a whole lot of questions up in the air. All I'll say is it's not impossible.

I've been rolling the idea of a 4-chip CPU cooler pod for a couple days and so far I'm liking it. There's a few different options for how to flesh it out, so until we crunch the numbers on the best base design I won't know anything for sure. Again, it's not impossible.

If Spondoolies cooperates with sample chips (and they've historically been better at that than probably everyone except ASICMiner, may they rest in peace) the only real limiting factor will be dev capital. We've got a plan to free up dev time in the next few weeks, but having time doesn't always mean having money. All our income is based on sales, right now that's between Compac sticks and PSU hardware. Hosting is basically a breakeven concern and mining revenue accounts for maybe a hundred bucks a month. We run a tight ship and can operate pretty lean compared to a lot of shops, but the level of leanness we're running right now isn't really sustainable.

Thank you for the update sidehack, if its just funding, maybe we can come up with someone community funded, ala kickstarted but less vaporware. Personally i'd be willing to buy in early with what i can and i'm sure i'm not the only one who would crowdfund you.

If the $/GH on the sticks were good i would totally buy tons of them to help you out but thats not doable, so i think there could be a lot of interest in something around or under 0.2$/GH.

If you come up with a plan for such PCB's (upgrade for S1/S3 path) or a pod/cpuheatsink-cooled that is concurrent, i think you would do great, considering how many people throw money at vaporware pre-order.

[philipma1957]

133Gh per chip

A 4amp 20w stick miner?  in time for santa's flypast?

well maybe on the under volt down clock method 2 amps 10 watts  and you get 60gh per stick


I can do 2 amps easy on my hubs  up to 8 sticks per hub  so I would really like these for sticks

[sidehack]

$0.2/GH is going to be hard to hit, I think. The best thing going right now is the S7 and that sells at what, about 38 cents per? I know prices are going to drop as more stuff appears, and I'll do the best I can but twenty cents is a hard target. That's about what a year-old S3 is selling for right now.

The $/GH on the stick, I've said it before and I'll say it again it's the best ever $/GH for a stick miner and I can't make it any cheaper. I know you're not complaining, but I'll reiterate the fact regardless. Something like a pod with about eight times the hashrate and only three times the price, I'd like to see that get going but my guys tell me it wouldn't sell either. I just have to wait and hope SPTech can come through for us with chips and I can do something good with 'em, and in the meantime keep selling Compacs and PSUs.

Regarding something community-funded, I do believe one of Kickstarter's rules for a hardware project is you have to have a working prototype before running the campaign. They'll fund manufacture but not development. I like that rule, and we try to run the same way. The first step on the path to BFL is conning someone else into funding your development. Well the first step on the path to BFL is apparently hiring known scammers to run your business, but the second step is getting folks to fund your development.

I don't know about doing TypeZero boards and a pod concurrently. The pod would come first since I'd much rather have something small and less complex to work on as far as ironing out control systems and multi-chip comms. The only real difference between the pod and the full-scale board would be chip count and probably a beefier power system, but all the digitals would transfer straight across so there wouldn't be much time between a pod and the full-size boards. I can work on non-control parts of the big boards while waiting on prototype parts to assemble and test the small boards.

If I had the money from a single typical vaporware scam sale (not even the whole scam, just a single customer's purchase) I could probably afford dev for the entire project.

[VirosaGITS]

$0.2/GH is going to be hard to hit, I think. The best thing going right now is the S7 and that sells at what, about 38 cents per? I know prices are going to drop as more stuff appears, and I'll do the best I can but twenty cents is a hard target. That's about what a year-old S3 is selling for right now.

The $/GH on the stick, I've said it before and I'll say it again it's the best ever $/GH for a stick miner and I can't make it any cheaper. I know you're not complaining, but I'll reiterate the fact regardless. Something like a pod with about eight times the hashrate and only three times the price, I'd like to see that get going but my guys tell me it wouldn't sell either. I just have to wait and hope SPTech can come through for us with chips and I can do something good with 'em, and in the meantime keep selling Compacs and PSUs.

Regarding something community-funded, I do believe one of Kickstarter's rules for a hardware project is you have to have a working prototype before running the campaign. They'll fund manufacture but not development. I like that rule, and we try to run the same way. The first step on the path to BFL is conning someone else into funding your development. Well the first step on the path to BFL is apparently hiring known scammers to run your business, but the second step is getting folks to fund your development.

I don't know about doing TypeZero boards and a pod concurrently. The pod would come first since I'd much rather have something small and less complex to work on as far as ironing out control systems and multi-chip comms. The only real difference between the pod and the full-scale board would be chip count and probably a beefier power system, but all the digitals would transfer straight across so there wouldn't be much time between a pod and the full-size boards. I can work on non-control parts of the big boards while waiting on prototype parts to assemble and test the small boards.

If I had the money from a single typical vaporware scam sale (not even the whole scam, just a single customer's purchase) I could probably afford dev for the entire project.

I understand. The 0.2 target was calculated with S5/S5 chips. For instance 300$/1300gh=0.23$/gh and i figure by the time your project go from dev to production to hitting people's door, it would have to be a bit lower still.

Of course if you do it at 0.1-0.15 J/GH the price per GH would naturally have to be higher.

And i know the USB was a toy and i'm glad i got one, in the end it i paid 45$~ for one and it was to support the project and play with it.
But to put in more money it would have to be something closer to 0.2$/GH than 1-2$/GH, of course higher depending on the efficiency, but i was comparing it to the current stick's efficiency.

Basically whatever you build has to fill a niche that beat other's in that category, either it be noise or price at various efficiency. Anyways thats how i see it. And thats why i got one stick, since it beat all other sticks at awesomeness

[sidehack]

Right now I'm more likely to get SPTech chips than I am BM1384, and given they're not talking to me about BM1385 either I doubt that'll happen. If I come out with anything it'll be new chips from someone and be capable of operating in the 0.15-0.25W/GH neighborhood. I'm not competing with the S5 - it's old news. I won't, and can't, compete with used hardware prices on new hardware with new chips a year old. If Bitmain and SP stay in the 80dB consumer and 17000W industrial products, it seems likely the "niche market" of people who want to be able to run miners like they've had access to for the two and a half years before right now will be wide open.

[Mudbankkeith]

4 Spondoolies chips will show maybe 500Gh.   That would be quite a mean little pod.
While 8 giving 1000+ would be a kick in the pants to Bit(stingy)main. Especially if it (and 3 brothers) would mount on a redundant S1/S3 chassis with a Pi(crust) on top.

[VirosaGITS]

Right now I'm more likely to get SPTech chips than I am BM1384, and given they're not talking to me about BM1385 either I doubt that'll happen. If I come out with anything it'll be new chips from someone and be capable of operating in the 0.15-0.25W/GH neighborhood. I'm not competing with the S5 - it's old news. I won't, and can't, compete with used hardware prices on new hardware with new chips a year old. If Bitmain and SP stay in the 80dB consumer and 17000W industrial products, it seems likely the "niche market" of people who want to be able to run miners like they've had access to for the two and a half years before right now will be wide open.

If Bitmain doesn't release a S7 mini, Bitfury continue being picky about who they sell their stuff to and SP just keep building bigger and bigger ASIC, then i very much hope you can get those SP chips to make something custom. (Or get a very big crane to move those silly SP units >.>)

I don't intend for you to specifically compete at the S5 level, if you compete at the SP50/S7 level, thats a okay too, we might be looking at the need for 0.37$/GH + or less 5 cents at current values, and thats a-ok. Bitfury also allegedly have 0.06W/GH iirc?

If you had some 4-5 chip pod that did 600-750GH/s for 0.2W/GH i would gladly fork something around one BTC for one and i'd probably take a couple per month. Assuming we can find a cost efficient quiet-ish Heatsink to slap on that, i'm sure many others wouldnt mind loosing their pockets a bit.

A nice feature the community could come up for such miner would be it fitting in a stackable plastic frame maybe? Then you have their fan pull the air into the heatsink, outward, in one direction ish for easier dumping of hot air. Or whatever.

[sidehack]

The pod board would not be designed for S1 compatibility. They'd be two separate PCBs.

If we get far enough into pod to make it viable, I've been needing to talk to a couple people that shipped me a few cases of Freezer 7 coolers last year for about fifty Technobit Minion boards that never got shipped. The coolers are still sitting in boxes on my hosting shelves, been there over 13 months by now. I could look into buying them off the guys and making that an optional sale alongside a pod PCB.

I think four chip is going to be easiest to wrangle, which leaves a 500GH top-end for the pod. I say top-end but until I get some actual chip data beyond the one operating datapoint Spondoolies has on the SP50 product page I don't really know. Could be higher, could be lower, but I don't want to build more than about 100W into it. A decent air cooler should be good for 100W. That might mean five chips but I don't have enough info yet so we're talking over various configurations of four chips. I've talked to Novak about building it to sell in the $200 range, so a 1BTC price point I'm preliminarily confident is doable.

I'm not too much on package design, but if someone wants to volunteer to help, I can share mechanical specs as they get ironed out.

If I build something bigger than the S1 scale, it'll probably be something along the lines of the rack machine that was being discussed in another thread, using seven or eight of the same boards. I'm not sure my little shop is up to the task of designing and manufacturing a whole product line of electronics quite yet, but a pod and the TypeZero board in various configurations should meet most needs between 50W and about 2500W. I don't feel like competing in the 1200W Jet Turbine or 17KW Shipanchor market sectors.

[VirosaGITS]

The pod board would not be designed for S1 compatibility. They'd be two separate PCBs.

If we get far enough into pod to make it viable, I've been needing to talk to a couple people that shipped me a few cases of Freezer 7 coolers last year for about fifty Technobit Minion boards that never got shipped. The coolers are still sitting in boxes on my hosting shelves, been there over 13 months by now. I could look into buying them off the guys and making that an optional sale alongside a pod PCB.

I think four chip is going to be easiest to wrangle, which leaves a 500GH top-end for the pod. I say top-end but until I get some actual chip data beyond the one operating datapoint Spondoolies has on the SP50 product page I don't really know. Could be higher, could be lower, but I don't want to build more than about 100W into it. A decent air cooler should be good for 100W. That might mean five chips but I don't have enough info yet so we're talking over various configurations of four chips. I've talked to Novak about building it to sell in the $200 range, so a 1BTC price point I'm preliminarily confident is doable.

I'm not too much on package design, but if someone wants to volunteer to help, I can share mechanical specs as they get ironed out.

If I build something bigger than the S1 scale, it'll probably be something along the lines of the rack machine that was being discussed in another thread, using seven or eight of the same boards. I'm not sure my little shop is up to the task of designing and manufacturing a whole product line of electronics quite yet, but a pod and the TypeZero board in various configurations should meet most needs between 50W and about 2500W. I don't feel like competing in the 1200W Jet Turbine or 17KW Shipanchor market sectors.

Sound good to me, i hope more people follow the thread and decide to at least mention their interests in the project.

I didn't see the stackable rack you mentioned but basically, if you make a small PCB and then lay a heatsink and then sandwich the heatsink between another PCB and another and another, you could probably just use a quiet but bigger form factor fan to push air between the PCB's, a bit like the S7 design, it could allow interesting scaling and efficient space usage.
It would also make directed airflow easier any maybe even use vinyl tubing to take the final exhaust outside or something.

And i hope you don't try to compete in the 1200W+ range, a smaller stackable/compact/space-wise unit of any kind that you can easily fit physically and electrically, that is quiet and relatively cost efficient sound like a great niche to work in.

[sidehack]

Discussion on the rackable unit was coming down to figuring out how best to stack seven or eight of the S1-sized board and heatsink combo in a 4U and allow for both external power cabling and internal interchangeable server PSUs in the 1Ux2U scale.

Sandwiching a PCB between two heatsinks is going to have a problem of now you're limited to a single-sided board with no components taller than your ASICs, or you have to mill out channels in your heatsinks or use individual stick-on chipsinks. In any case it's not really something I look forward to requiring. My goal for the TypeZero board is a max power dissipation around 320W, stock settings probably more like 250W or under, and able to run off an unmodified S1 chassis. Might need outside panels like the S3's housing or S5 provide but that's not really difficult.

I don't really want to build something like the S7. Power density may be great for folks whose costs go by the square or cubic foot, but it sucks for anyone trying to keep a rig quiet and stable and those are more likely my target customers.

[VirosaGITS]

Discussion on the rackable unit was coming down to figuring out how best to stack seven or eight of the S1-sized board and heatsink combo in a 4U and allow for both external power cabling and internal interchangeable server PSUs in the 1Ux2U scale.

Sandwiching a PCB between two heatsinks is going to have a problem of now you're limited to a single-sided board with no components taller than your ASICs, or you have to mill out channels in your heatsinks or use individual stick-on chipsinks. In any case it's not really something I look forward to requiring. My goal for the TypeZero board is a max power dissipation around 320W, stock settings probably more like 250W or under, and able to run off an unmodified S1 chassis. Might need outside panels like the S3's housing or S5 provide but that's not really difficult.

I don't really want to build something like the S7. Power density may be great for folks whose costs go by the square or cubic foot, but it sucks for anyone trying to keep a rig quiet and stable and those are more likely my target customers.

Ah, yeah i can see the problem with that idea.

And I don't personally mind the space efficiency that much, for my own devive i was thinking more of it creating an ease of airflow direction to dump it outside, since imo, home mining is all about cheap electricity and no cooling cost.

If the air flow straightforwardly like pretty much any of the Antminers, its easier to dump. Since the miner itself push air, you just need to have the exhaust already outside your place when it exit the miner.

I guess its probably best come to that "problem" when we have some specs out.