Bitmain Antminer S6 and S7 Speculation

[VirosaGITS]

S5, or S5+? The S5+, like the S4+, I have a hard time understanding. Now if instead of designing an S5+ with 7.7TH in a terribly hideous and cumbersome form-factor and very little improvement on technology already eight months old, or an S4+ using more-than-year-old chips in a mediocre and highly-overpriced offering worse than any miner sold this calendar year, they had retooled the S4 build around the BM1384, it might have been something. A 5x17 string instead of 3x17 of BM1382 would have built a 4.2TH miner in the same case, same heatsinks, same PSU. That would have been worth buying. A clumsy behemoth trying to extract 1KW from a box the size of an S1 and with about a 10% power savings over the stuff they were shipping last Christmas, with no good way to power it and no good way to even shelve the thing safely? Yeah, definitely a mess.

I got about a dozen S1 sitting on the shelf waiting, and several stripped-down dead S5 chassis as well.

I meant the S5 design seemed rush and very inefficient compared to what it should be able to produce. For instance see the huge efficiency gain from just undervolting the 1.91v board to 11.5v. Lose nearly no hashrate, save 12% in power. *Roll eyes*
You yourself managed to do much better with that chip.

For the S5+ i don't even want to talk about it anymore. Its more a Frankenstein thats beneficial to Bitmain than anything else. In comparison the S5 is a gem.

[QuintLeo]

Well... the BM1385 certainly beat our predictions in this thread... it's slated to be 0.23W per GH (or 230W per TH).  Now we just have to wait and see what the actual units will look like.  If they're still drawing the same 590W like the S5, that's about 2.5TH/s.

 18 chips per string, that's a given from the BM1385 published specs and Bitmain's long-standing tendancy to run their new designs at "max hash/max volts".

 That works out to 2.34 TH at about 540 watts "at the wall", on a 2 board 2 string per board "same as S5/S3/S1" form factor unit.
 I suspect the actual final announced specs will be a HAIR off from that estimate, more on the power side than the TH side.

 I would guess the unit will initially be priced at 3.5-4 BTC, but that's a LOT more speculative and I wouldn't be shocked at anything up to 5.


 Release date? Whenever Bitmain finishes replacing their "farm" S5s, or when a competator shows up with working hardware or at least announces working production of chips (other than SFards, whish is starting to make me think "Sfarts" of late) of those competators that sell to the public.

[VirosaGITS]

Well... the BM1385 certainly beat our predictions in this thread... it's slated to be 0.23W per GH (or 230W per TH).  Now we just have to wait and see what the actual units will look like.  If they're still drawing the same 590W like the S5, that's about 2.5TH/s.

 18 chips per string, that's a given from the BM1385 published specs and Bitmain's long-standing tendancy to run their new designs at "max hash/max volts".

 That works out to 2.34 TH at about 540 watts "at the wall", on a 2 board 2 string per board "same as S5/S3/S1" form factor unit.
 I suspect the actual final announced specs will be a HAIR off from that estimate, more on the power side than the TH side.

 I would guess the unit will initially be priced at 3.5-4 BTC, but that's a LOT more speculative and I wouldn't be shocked at anything up to 5.


 Release date? Whenever Bitmain finishes replacing their "farm" S5s, or when a competator shows up with working hardware or at least announces working production of chips (other than SFards, whish is starting to make me think "Sfarts" of late) of those competators that sell to the public.

Indeed, unless they need to kill off competition (again) they don't need to release this yet. This would just undermine their... mine.

[sidehack]

Don't forget the controller and fan power, which could easily be another 25W (becoming 28 more at the wall).

The S5 is inefficient for the chip, because it's running at the top of what's possible for it. The S[odd] miners always did that, and the S[even] miners came later with higher chip density and lower per-unit power. Inefficient or not, the S5 was the most efficient machine available at the time, and nobody's really beaten it in a cost-effective way in the last 8 months. Sucks to say, but that was a pretty optimal design when you consider $/GH and that efficiency was already "good enough".

I'm betting they announced the chip as a preemptive strike on Spondoolies or LK Group announcing new gear (speculative, of course). Quite possibly they're only toying with factory samples now, which is why they aren't announcing an S7 design or any performance curves beyond the one data point. Or they're only letting slip a little bit of info and they've actually had them for months and just aren't telling everyone about their already-deployed S7 stacks because they don't want to yet.

[VirosaGITS]

Don't forget the controller and fan power, which could easily be another 25W (becoming 28 more at the wall).

The S5 is inefficient for the chip, because it's running at the top of what's possible for it. The S[odd] miners always did that, and the S[even] miners came later with higher chip density and lower per-unit power. Inefficient or not, the S5 was the most efficient machine available at the time, and nobody's really beaten it in a cost-effective way in the last 8 months. Sucks to say, but that was a pretty optimal design when you consider $/GH and that efficiency was already "good enough".

I'm betting they announced the chip as a preemptive strike on Spondoolies or LK Group announcing new gear (speculative, of course). Quite possibly they're only toying with factory samples now, which is why they aren't announcing an S7 design or any performance curves beyond the one data point. Or they're only letting slip a little bit of info and they've actually had them for months and just aren't telling everyone about their already-deployed S7 stacks because they don't want to yet.

That's true, it is still a good unit and will still be for a while. However the S1 was easy to undervolt. And quiet.(Still ROI able to me at 1J/GH) +Wifi
The S3 downclock well and is quiet. +Wifi (but no proper external connector iirc)
The S5 was meant to be super underclockable as well (remember the 0.2J/GH portion of their first information out?), but then suddenly it seem like it had to be rushed and sold cheap. As if they just wanted to kill off the competition. Only the latest PCB seem to handle undervolting. But i guess i'm speculating quite a bit in what could have beens.

And Yep, releasing those right this moment is still a waste for them. But the right time may come to be soon. They are possibly just keeping it as ammo to kill off the competition as soon as they come out.

[QuintLeo]

Don't forget the controller and fan power, which could easily be another 25W (becoming 28 more at the wall).

 Note that the specs they published on the BM1385 were the "at the wall" spec numbers, NOT the chip numbers, for power.

 I'm STILL saying they already have S7's built and hashing away in their farm/Hashnest.

 I suspect the CHIP can go down to more like .12 at "optimal efficiency" undervolt.


 The only "waste" to them on actually putting the S7 up for sale right now would be if they still have any of those S5+ things or any S5's left in their hands unsold. I'm CERTAIN that's the only thing they are waiting on, though if someone else releases a miner that might preempt them.

[notlist3d]

Don't forget the controller and fan power, which could easily be another 25W (becoming 28 more at the wall).

 Note that the specs they published on the BM1385 were the "at the wall" spec numbers, NOT the chip numbers, for power.

 I'm STILL saying they already have S7's built and hashing away in their farm/Hashnest.

 I suspect the CHIP can go down to more like .12 at "optimal efficiency" undervolt.


 The only "waste" to them on actually putting the S7 up for sale right now would be if they still have any of those S5+ things or any S5's left in their hands unsold. I'm CERTAIN that's the only thing they are waiting on, though if someone else releases a miner that might preempt them.

It is very unlikely they have S7's already at data centers.  Don't really see a huge jump.  Unless they put in S5 slots and did not do a big batch.

I think it's likely they have prototypes or working models.  But I don't think they have a batch made yet for their dc.   I think once mass production they will likely list quickly on site.  We are not there yet in my mind.

[blindminer]

oh men    I bought 25 s5's last week! 

[sidehack]

 Note that the specs they published on the BM1385 were the "at the wall" spec numbers, NOT the chip numbers, for power.

I'm fairly certain that's false. The chip spec they posted was 0.216W/GH which translates to 216W/TH and the machine-level number they gave was 230W/TH assuming a 93% PSU; 216/0.93=232 which backs up that assertion.

My bet is 2x18 string, 32.5GH per chip makes 1.17TH per board at 253W per board so about 505W for the pair at 2.34TH. Add 25W for controller and really loud fan makes 530W. A typical 90% PSU takes that up to 589W at the wall, or a 93% gives you 570W wall. I certainly wouldn't complain if it came in under that. I also certainly wouldn't complain to see a chip-level performance curve chart like they posted for the BM1384.

[philipma1957]

 Note that the specs they published on the BM1385 were the "at the wall" spec numbers, NOT the chip numbers, for power.

I'm fairly certain that's false. The chip spec they posted was 0.216W/GH which translates to 216W/TH and the machine-level number they gave was 230W/TH assuming a 93% PSU; 216/0.93=232 which backs up that assertion.

My bet is 2x18 string, 32.5GH per chip makes 1.17TH per board at 253W per board so about 505W for the pair at 2.34TH. Add 25W for controller and really loud fan makes 530W. A typical 90% PSU takes that up to 589W at the wall, or a 93% gives you 570W wall. I certainly wouldn't complain if it came in under that. I also certainly wouldn't complain to see a chip-level performance curve chart like they posted for the BM1384.

so 570 to 600 at the wall for 2340 gh =  .2435 watts a gh to  .2564 watts a gh


even at .28 watts a gh (shit psu bronze or worse) it would still be a solid improvement

[VirosaGITS]

Basically we're likely to see a S5 design/fan/noise thing with around same watt consumption but twice the hashrate.

Unless they continue that monster trend of 3x units.

[QuintLeo]

 Note that the specs they published on the BM1385 were the "at the wall" spec numbers, NOT the chip numbers, for power.

I'm fairly certain that's false. The chip spec they posted was 0.216W/GH which translates to 216W/TH and the machine-level number they gave was 230W/TH assuming a 93% PSU; 216/0.93=232 which backs up that assertion.

 The spec I saw was the 230 watt / TH "at the wall" in that chart, the one that had the S5 listed at 510 watts/TH (which IME is a hair low, but some folks have reported hitting it in reviews).

 I'm not sure how much power a BB uses, the fan ALONE is about 20 watts though (presuming the use the same fan as on the S5, I don't see any reason they'd change that) and the linkage board/logic probably used a watt or two.

 I suspect the S7 eats a hair less power than the S5 did for it's hair-more-than-double hashrate - sidehack seems to be ending up at pretty much the same place as I did from slightly different start points. 9-)

It is very unlikely they have S7's already at data centers.  Don't really see a huge jump.  Unless they put in S5 slots and did not do a big batch.

 I don't see any reason why they would waste the time replacing Hashnest S5s with those S5+ things, only to have to turn around and replace them AGAIN with S7s. I do see a VERY FAINT possibility they have a "S7+/S8" design going into Hashnest.

 Those "used" S5s are getting replaced by SOMETHING, after all.

[mavericklm]

Unless they continue that monster trend of 3x units.

15 or 16th monster..... muahahahhahah
i'd buy!

[philipma1957]

Unless they continue that monster trend of 3x units.

15 or 16th monster..... muahahahhahah
i'd buy!

If they sell an 11th to 15th beast I will buy it.

If they go big

I rather they sell something that maxes at 2700 watts  not 3500 watts

[mavericklm]

i was thinking of underclocking the s5+ to ~2.8kw

[valkir]

I will prefer a 1500w-1700w max gear. So we could put that on a 15a 120v breaker. 

[dmwardjr]

Unless they continue that monster trend of 3x units.

15 or 16th monster..... muahahahhahah
i'd buy!

If they sell an 11th to 15th beast I will buy it.

If they go big

I rather they sell something that maxes at 2700 watts  not 3500 watts

I would have to agree with you there, Phil.

All of my circuits are 240 Volt @ 30 Amps = 7,200 watts.  It would be beneficial to put two big rigs on a 30 amp circuit without pushing it to its 7,200 watt limit for safety purposes.  If less than 2,700 watts, it would be nice to have the rig low enough to put 3 or 4 rigs on one 240V/30A circuit.

Does anyone have any speculation about the "cost" per GH on the S7?

[alh]

In terms of the cost for the S7, I'll just bet it's more than you'll want to pay. I think that the primary driver for the S7 price will be "Whatever the market will bear". Unless they have some serious competition in terms of gear, they will try and extract every dollar they can for each miner. I don't think they care about ROI times, except to the extent that their customers do calculations and then buy or refuse to buy. Since the price of BTC also factors into customer calculations on return, that will fold in, but the number one driver will be what folks are willing to pay, large or small.

Right now, they seem to have approximately zero credible competition. The SFARDS pricing provided them with plenty of room to the upside. Besides, it's easy to lower the price later if folks aren't buying. I expect they they think there is plenty of pent up demand for new hardware, and they are probably correct. That won't lead to a low price.

Just my rather cynical $.02 on the pricing guesses for the S7.

[dmwardjr]

In terms of the cost for the S7, I'll just bet it's more than you'll want to pay. I think that the primary driver for the S7 price will be "Whatever the market will bear". Unless they have some serious competition in terms of gear, they will try and extract every dollar they can for each miner. I don't think they care about ROI times, except to the extent that their customers do calculations and then buy or refuse to buy. Since the price of BTC also factors into customer calculations on return, that will fold in, but the number one driver will be what folks are willing to pay, large or small.

Right now, they seem to have approximately zero credible competition. The SFARDS pricing provided them with plenty of room to the upside. Besides, it's easy to lower the price later if folks aren't buying. I expect they they think there is plenty of pent up demand for new hardware, and they are probably correct. That won't lead to a low price.

Just my rather cynical $.02 on the pricing guesses for the S7.

Thanks for sharing your thoughts, alh.

I only wish I knew what that amount will be...

I would almost be willing to bet the first batch will be gone in less than 12 to 24 hours.  Many more would be miners with power costs as high as $0.20 cents per kWH will be priced into buying this rig with almost double the hash rate at half the power.  We already know large farms, small farms, hobbiest, etc... will want to get their hands on this rig.  I can see the difficulty getting as high as 100 to 110 Billion by end of April 2016.

I may not jump on the S7 band wagon until batch 2 or 3.

[alh]

I would also argue that without announced specifications (hashrate, power draw, etc.), and a fairly firm date for product delivery, price is kinda meaningless. Yes,  we got some good insight as a result of the BM1385, but the rest is still highly speculative. There are a variety of possible results using that chip, all of which have their advantages and drawbacks, along with associated costs.

As I've said before, price is the easiest thing to adjust on a product like this. All the engineering stuff is difficult to change once completed, but we've seen Bitmain tinker with the prices more than once in a week back in February.