GekkoScience BM1384 Project Development Discussion

[QuintLeo]

I didn't see it either when I looked for it the day after the original announcement.

 The limited number of voltages listed has some interesting implications - seems like we are definitely hitting the hard limits on 28nm (not a shock if it's full custom), though I'm a bit supprised that the max voltage listed is so low.

 I'm even MORE supprised that the "default" is now a mid-voltage point, as opposed to Bitmain's normal "highest voltage is the default".
 THAT makes me wonder if the S7 is going to turn out to be a 17 chip per string device, not an 18 as some of us have speculated - but upon reflection, no way they'd have published that "230 watts at the wall per TH" claim if so.

[1714201949]

1714201949

[1714201949]

1714201949

[1714201949]

1714201949

[AJRGale]

I'm not sure you'll get the full 32.5 GH/s per chip with the undervolt to .6 volts per chip, but should be fairly close with a small undervolt like that at worst.
Depends on how much "overclock" headroom the chips turn out to have.

Trick is to keep the cost under 1/4 of whatever a S7 ends up selling for, otherwise it would make more sense to just get more S7s.

Are thel BM1385 chip specs posted yet anywhere?

Do you mean the datasheet that is available on Bitmain website since the announcement of the chip?

https://bitmaintech.com/files/download/BM1385_Datasheet_v2.0.pdf

looking at the footprint, yet another different layout Again! at least this one looks a little more logical! no more odd blocks and tracks to throw the VCC in.. buuut still have to physically see it it might still bee some odd block for the power..

[sidehack]

Wow, that datasheet is a heck of a lot better than they've put out before! I really like that the protocol is detailed.

Regarding the pinout and such, they did implement a lot of the stuff that's been mentioned in this thread. Given this chip's been in development for months I'd say they beat us to the punch, but that doesn't mean I don't like it. Also if I'm reading right, I think the _A pins are the signal pins with integrated level shifters? And an integrated 0.9V LDO for the PLL?

Dude. I really want to play with some of these chips. You know how much easier the designer's job gets with integrated level shifters? Holy cats.

[VirosaGITS]

If the work seem easier with these chips, maybe we could think of a crowdfunding plan for obtaining a batch of these chips? Kind of a group buy thing.

[alh]

If the work seem easier with these chips, maybe we could think of a crowdfunding plan for obtaining a batch of these chips? Kind of a group buy thing.

While that sounds great, I'd be stunned if Bitmain would give up any of the BM1385 parts any time soon, even for a King's ransom. Maybe sidehack can "sweet talk" them, but I don't expect they will yet.

[sidehack]

It's certainly worth looking into. I'd be hesitant to start it without first getting sample chips and having at least most of a working design. Hopefully they're as willing to get us sample chips as they were with the BM1384. I can understand not wanting to hand any out right now since they haven't released a miner built with them yet.

[VirosaGITS]

I see. Well that sound fair enough but if it come down to a bit extra fund i think me and several other wouldn't mind keeping it small and doing some kind of mini private funding. If that would let us get some custom PCB that work well, it could draw enough interest to do proper crowdfunding and scale it up a bit if desired. Some kind of limited gamble.

But i understand its a bit early for that, since i'm guessing Bitmain didn't even reply to or give more information on date and pricing of the next chip yet.

[philipma1957]

It's certainly worth looking into. I'd be hesitant to start it without first getting sample chips and having at least most of a working design. Hopefully they're as willing to get us sample chips as they were with the BM1384. I can understand not wanting to hand any out right now since they haven't released a miner built with them yet.

They may sell you samples if you agree to launch after the s-7

They would look pretty bad if your gear is way better then theirs is and it releases at the same time.

I would think they want your gear to release sixty days later then the s seven.

[sidehack]

I may very well have power and some peripheral controls ironed out before they agree to sell sample chips. I think between Compac sales and PlanetCrypto's cooperation I shouldn't run short of dev funding to get a prototype out once we have sample chips. Depending on how many chips we can get to start, that could mean several of a couple different boards for a few folks to play with. Be good advertising.

If I had sample chips right now, I probably still wouldn't be able to ship a thing but prototypes before December.

[VirosaGITS]

That doesn't sound like a particularly horrible timeline. 3 Months or so after having received some chips?

I personally wouldn't mind the wait for some PCB's, maybe prototypes, for testing and reviewing at a good price. Sound like fun and possibly rewarding.

[sidehack]

I wouldn't expect to have anything to ship within six week of the start of manufacture. I couldn't start manufacture until after probably several weeks of gathering funds. I wouldn't start gathering funds until I had a working prototype. If I had miners to ship for Christmas I'd be pretty happy with that timeline. Prototype boards before the end of September would be nice.

[VirosaGITS]

I see, reality seldom match dreams.

But that's still interesting. I wonder what are the odds of Bitmain coming up with a BM1386 or such in that time frame.

[sidehack]

I kinda doubt they do, at least not for general release. Right now they are their only real competition, so unless the other guys put up something substantially better than the BM1385 in the relatively near future they have no incentive to ship another new chip.

[notlist3d]

I kinda doubt they do, at least not for general release. Right now they are their only real competition, so unless the other guys put up something substantially better than the BM1385 in the relatively near future they have no incentive to ship another new chip.

They are very secretive as most are.  Until they release their own miner I doubt they will sell to anyone else.

They no doubt will want to be the first ones with a miner using it to grab those initial sales.  I hope after they do support projects like yours, but I don't know if they will until much later. 

[chiguireitor]

Wow, that datasheet is a heck of a lot better than they've put out before! I really like that the protocol is detailed.

This so much....

The previous datasheets looked like they didn't want anyone playing with their stuff, but this one, this one does reek of integration on everything, and then a little more...

Here's hoping the thing comes at least in QFP, would be OMG-SO-COOL.

[PlanetCrypto]

Wow, that datasheet is a heck of a lot better than they've put out before! I really like that the protocol is detailed.

Concur

Regarding the pinout and such, they did implement a lot of the stuff that's been mentioned in this thread.

Good, it's about time a manufacturer started paying attention to the intelligence available in the community at large.

Given this chip's been in development for months I'd say they beat us to the punch, but that doesn't mean I don't like it.

IMHO:
It's still a 28nm product, a 28nm process with some maturing improvements, but never the less it's old tech.
Of conspicuous notice is the last revision date (upper left hand corner of every page) of 8/20/2015. As of this writing that's a mere 9 days ago.
The datasheet has the kind of info S/W guys will like, but leaves the H/W guys guessing a bit.
Looks like they listened to sidehack et. al.'s comment/desire to bring the power in on one side and GRND on the pad (application of common sense).

Do I like it? Yup.
Can I get it? NOPE!.

My position remains: The bestest coolest tech in the world ain't worth the powder to blow it to hell if I can't buy it.
As such, releasing the datasheet is relegated to hyping their design for a complete miner (of which the profit margins are higher than chip sales) versus actually providing the community with a usable, reliably available, power consumption competitive raw component.

They released just enough info to get us talking about it, versus us talking about Innosilicon's 14nm chip, a community based dev effort, or something else.
This boils down to marketing BS nothing more.
It's commonly known that any press is better than no press, as such I'll have no further public comment about their "new" chip and/or S7.
Sell me 1K chips at a reasonable price and I'll change my tune.

HAMLET [Re-enter Players with recorders]
O, the recorders! let me see one. To withdraw with you: -- why do you go about to recover the wind of me, as if you would drive me into a toil?

GUILDENSTERN
O, my lord, if my duty be too bold, my love is too unmannerly.

HAMLET
I do not well understand that. Will you play upon this pipe?

GUILDENSTERN
My lord, I cannot.

HAMLET
I pray you.

GUILDENSTERN
Believe me, I cannot.

HAMLET
I do beseech you.

GUILDENSTERN
I know no touch of it, my lord.

HAMLET
Why, look you now, how unworthy a thing you make of me!
You would play upon me,
You would seem to know my stops,
You would pluck out the heart of my mystery,
You would sound me from my lowest note to the top of my compass;
and there is much music, excellent voice, in this little organ, yet cannot you make it speak.
'Sblood, do you think I am easier to be played on than a pipe?
Call me what instrument you will, though you can fret me, yet you cannot play upon me.

[philipma1957]

I see, reality seldom match dreams.

But that's still interesting. I wonder what are the odds of Bitmain coming up with a BM1386 or such in that time frame.

No the single biggest reason is the halving comes next July since they just spent money to build the 1385

They will want to get good money from it before they spend close to a million or two for the 1386.

After the halving happens and we all get a real idea of pricing they will decide if they want to spend money making a 1386.

Of course if price shoot to the moon like it did in 2013 money for the new 1386 could show soon rather then late.

[sidehack]

PlanetCrypto - what I meant by "beat us to the punch" is that the package and whatnot was probably already designed when we started talking about what we'd like to see, which means their designers had already come to the same conclusions which we voiced later.

If we assume Bitmain already has S7 on their shelves, we assume the chip was completed in the last several months. If we assume Bitmain paid attention to our considerations, it makes much more likely their not having any completed S7 because the chips weren't completed until much more recently.

I don't see why we shouldn't talk about BM1385 and someone else's chip, or our own chip. If you want to start your community dev discussion thread and have a center for conversing about it, this thread can be the center of discussion on the BM1385.

The data sheet doesn't actually disappoint me too much as a hardware guy. More than three reference points for a performance curve would be nice, but not necessary. Seeing the top-clock for the chip still gives a 10W output, and the 0.66V/7W is a midrange setting (which I was suspicious of) I'll have to rethink my layout again. A 20-chip board is probably still okay; at top clock it'll certainly run quite hot so I may cut that back down to 18 or even 16. 16-chip board at top clock would give us about 170W per board and 680W per machine, which is really pushing an S1 chassis pretty hard. I guess an S5 pushed to 425MHz has to clear 630W of heat.

An S1 outfitted with 16-chip boards would push 2.5TH at the top end (at about 700W if it didn't burst into flames) and clock down to 1.4TH at around 280W.

The Spec2 boards would probably run 24-chip, for a top-clock rating of 930GH at about 260W and bottom clock of 500GH at 100W. This would give an 8-board rack unit 7.3TH/2100W to 4TH/850W range.

If I wanted to push it to 30-chip we'd see 1.16TH at 325W and 625GH at 125W, for a machine-level range of 9.3TH/2550W to 5TH/1KW.

That's the estimated DC power consumption, so divide by PSU efficiency to get wall power.

I think it'd be safer for the Spec1 to go 16-chip instead of 20-chip. That reduces the potential long-term performance by 20% (in terms of hashrate, not efficiency) but reduces the initial cost (which could save $40 per 4-board kit just in chips) and retains a high top-clock hashrate with reduced potential to burn up your machine (versus someone trying to top-clock a 20-chip board and burning 850W on an S1 chassis). A 30-chip Spec2 is still very doable. The top-clock power draw is higher than I'd like to spec the per-board power draw for the machine (as seen in this thread: https://bitcointalk.org/index.php?topic=1158807.0) but not by much. Stock setting would probably be more like 275W with the opportunity to push them higher at your own risk made available.

[philipma1957]

PlanetCrypto - what I meant by "beat us to the punch" is that the package and whatnot was probably already designed when we started talking about what we'd like to see, which means their designers had already come to the same conclusions which we voiced later.

If we assume Bitmain already has S7 on their shelves, we assume the chip was completed in the last several months. If we assume Bitmain paid attention to our considerations, it makes much more likely their not having any completed S7 because the chips weren't completed until much more recently.

I don't see why we shouldn't talk about BM1385 and someone else's chip, or our own chip. If you want to start your community dev discussion thread and have a center for conversing about it, this thread can be the center of discussion on the BM1385.

The data sheet doesn't actually disappoint me too much as a hardware guy. More than three reference points for a performance curve would be nice, but not necessary. Seeing the top-clock for the chip still gives a 10W output, and the 0.66V/7W is a midrange setting (which I was suspicious of) I'll have to rethink my layout again. A 20-chip board is probably still okay; at top clock it'll certainly run quite hot so I may cut that back down to 18 or even 16. 16-chip board at top clock would give us about 170W per board and 680W per machine, which is really pushing an S1 chassis pretty hard. I guess an S5 pushed to 425MHz has to clear 630W of heat.

An S1 outfitted with 16-chip boards would push 2.5TH at the top end (at about 700W if it didn't burst into flames) and clock down to 1.4TH at around 280W.

The Spec2 boards would probably run 24-chip, for a top-clock rating of 930GH at about 260W and bottom clock of 500GH at 100W. This would give an 8-board rack unit 7.3TH/2100W to 4TH/850W range.

If I wanted to push it to 30-chip we'd see 1.16TH at 325W and 625GH at 125W, for a machine-level range of 9.3TH/2550W to 5TH/1KW.

That's the estimated DC power consumption, so divide by PSU efficiency to get wall power.

I think it'd be safer for the Spec1 to go 16-chip instead of 20-chip. That reduces the potential long-term performance by 20% (in terms of hashrate, not efficiency) but reduces the initial cost (which could save $40 per 4-board kit just in chips) and retains a high top-clock hashrate with reduced potential to burn up your machine (versus someone trying to top-clock a 20-chip board and burning 850W on an S1 chassis). A 30-chip Spec2 is still very doable. The top-clock power draw is higher than I'd like to spec the per-board power draw for the machine (as seen in this thread: https://bitcointalk.org/index.php?topic=1158807.0) but not by much. Stock setting would probably be more like 275W with the opportunity to push them higher at your own risk made available.

I  say less is more when it comes to the s-1 form factor.

   If you can build an s-1 upgrade kit that does 1.4th at 280w  it will sell and sell and sell.
If you can build ans s-3 upgrade kit thats does  that same spec 1.4th at 280w  it would also sell and sell and sell

Here is why that much power 280 watts in that size frame can be made to run quietly.  I run s-3's at my friends office.

I add the silverstone 141 fan as a pull keep the stock fan as a push  run the gear at stock speeds using 330 watts.

My results are fairly  quiet gear.

Many home miners are sound restricted.  And I know the s-1 or s-3  are not loud with just a little modding and tiny down clock.

1.4th and 280 watts   to me means :

 my current 4x s-3 setup  goes from 1.8 th at  1450 watts  to 5.6th at 1120 watts

you can sell that big time!

[Mudbankkeith]

There must still be loads of S1, S1/3 upgrades and S3, units sitting in lonely neglected corners, all waiting to be lovingly revived with nice shiny T/h capable boards.

Bitmain are unlikely to come back with anything suitable to replace them.

Bring on the BM1385 chip. It may be old tech 28nm cleaned up and repackaged, but it could still fill a gap for the home miners and see the community through to the block reward halving due next spring.