Topic: rumors of 3.5Th antminer S7 "lite" - page 3. (Read 8777 times)

Not really, the First format on the S7 almost was, since 1200watts is close to the 12A 110V limit. Meanwhile 1000watts on the 1440watts 80% safety margin on 120V is kind of awkward. Got to find something else to do with the 400watts.

The S4 is the one that was really fittingd as a one miner, one circuit miner 120v wise, the PSU is even rated to exactly 1440watts.

I think this version could be some sort of sop to the 120V home market. The typical US 120V, 15 amp outlet can handle 12 amps max continuously. With a little tuning plus a decent power supply, this version can be a one electrical outlet/branch circuit miner.

I agree you.

OK so stepping back from the detail of Batch 6, how about this for a bit of speculation as to what is going on at Bitmain as regards the S7? I still strongly feel that this is not a "New or Lite" Miner it is a Design Correction.

So in a very short period of time we have seen 4 different versions / selections of the S7. We have had batches at 4.86TH, 4.66TH and some people have even had shipments at 4.45TH. Now we have a revised design with less chips at 4.05TH.

You don't plan things to go like this  and clearly they have been having real problems with achieving the original spec. There is evidence, with the stickers on the Boards, that they have been speed selecting and it may even be that they are doing the same with the BM1385 and at the System level?

However you just can't carry on doing this in a volume manufacturing environment so I think that "Management" said that's enough, change the design and get us out of this mess.

So my guess is that the chip reduced 4.05TH S7 is very new and may only exist on paper at this point in time. However "Marketing" wanted a Batch for Sale, so the Engineers get pressed into a specification and that is why the numbers just do not add up. The error on the chip count, the very strange quoting of Power Efficiency as 0.25 J/GH + 10% etc. I do not think we have seen the final spec yet and that will come when they make some measurements?

Where they have been lucky is that the comptition, or rather lack of competition, has given them a Free Ride, leaving them as the only Game in Town with this class of Miner, allowing them to get away with this mess while selling the Miners for a premium price.


Rich

I was expecting it to be quite different. Thought dual module and big change.  With new S7 change it is not near as big of change as I thought.

Lose hashpower and still have 10 PCIe ports needed.   I think it was a bad move.

 If you mean "falling diff", that isn't going to happen. Ditto "flat diff".

 Don't forget you don't GET the thing 'till sometime in December.

 My standard calc shows this thing has no more chance at RoI with anything but VERY cheap electric than the 4.66 or 4.86 batches did - they did NOT drop the price enough to account for the lower TH and the LONG DELAY IN SHIPPING.



 The only advantage to this unit is that the higher volts/chip should let it overclock and be more reliable on meeting it's specs - at the cost of efficiency (I do NOT believe that 0.25w/GH spec, but had to use it for my calc as it was what was available).

Yes very disappointed that they did not go along with our  2 Board design. I cannot afford do not want to spend this much money on a miner, but would like to move to S7 / BM1385 efficiency as even at the current BTC price ny S5's will not be profitable after the halving.

I do agree  with Phil


and think that relative to the earlier batches of S7, which are a bit on the edge for Voltage & Frequency, this one will offer more scope for Under & Over Volting & Frequency.

Rich

The day when we had 500 BTC and high 400's people got a bargain on better S7's then this and for cheaper.  Now it's a fixed 4 btc price so it does not happen again.

I think they should have went to 2 blade if going to downgrade it.  They kept the design with 10 PCIe for less hash... i was not thinking they would do that.

ah not what i was expecting sincerely, this is still too akin to the s7 as a power concern, but it seems that it cost much less, the s7 is still priced 7 btc with only 1/5 more of the hash

this can do roi in 4 months with free electricity and falt diff, with my bad electricity and a flat diff, i cna roi in 7 months without selling it before, mmh i'm a bit tempted

Still a big mystery to me as to what is going on with Batch 6? Yes the number of chips is clear at 135 but that has to mean that the chain length has been reduced from 18 to 15 chips. So shorter chain has to mean increased core voltage, which does make sense given the batches we have seen with a reduced Default frequency and the original very tight PSU spec.

However as I have detailed above the resultant 0.8V per chip just does not tie in with the BM1385 Data sheet, and as we all know increased Core Voltage = lower efficiency, and yet they are still quoting an unchanged 0.25J/GH.

So either the BM1385 Data Sheet is wrong, the efficiency has reduced relative to the earlier Batches, there is a significant change to the Hash Board design, or something else I have not thought of?  

Just noticed there is a subtle change to the efficiency spec.

Batch 1 - Power Efficiency: 0.25 J/GH (at the wall, with APW3, 93% efficiency, 25°C ambient temp)
Batch 6 - Power Efficiency: 0.25 J/GH + 10% (at the wall, with APW3, 93% efficiency, 25°C ambient temp)

However +10% would still only max it 0.275% and on my maths it's over 0.3J/GH



Rich

It may be a good unit for underclock & undervolt.  it says 11.6 to 13.0 volts   they may have altered the  voltage regulation like they did with the later s-5's

No more math with this, bitmain has correct the data:

 S7 Specifications:

1. Hash Rate: 4.05 TH/s

2. Power Consumption: 1042 W + 10% (at the wall, with APW3, 93% efficiency, 25C ambient temp)

3. Power Efficiency: 0.25 J/GH + 10% (at the wall, with APW3, 93% efficiency, 25°C ambient temp)

4. Rated Voltage: 11.60 ~ 13.00V

5. Chip quantity per unit: 135x BM1385

6. Dimensions: 301mm(L)*123mm(W)*155mm(H)

7. Cooling: 2x 12038 fan

8. Operating Temperature: 0 °C to 40 °C

9. Network Connection: Ethernet

10. Default Frequency: 600M

OK it's Morning in the UK so I am going to have another go at wrapping my head round what's going on with this Not So Lite S7? Excuse me while I think and Type, this may or may not lead anywhere?  

So here's the guts of the post from carlosmnk in the other thread that suggests 135 chips.


If you just look at the maths, the case for 135 chips is spot on, with the Hash per chip dividing down from 4.86TH with 162 Chips to 30GH and the multiplying neatly back up with 135 Chips to 4.050TH.

So what's the problem? Well the resultant Core Voltage on the Chips is the problem....

If we go back to the 4.86TH system, then 162 Chips are divided onto 3 Boards giving 54 Chips per board giving a Chain of 18 Nodes with 3 Chips / Node. Apply 12V to the 18 Chip string and we get 0.667V which is bang in the middle of the Core Voltage range shon in the BM1385 Data Sheet.



Now if we look at a 135 Chip System, then that has to be 3 Boards which gives 45 Chips per board and the only way that can be divided up is with a Chain of 15 Nodes with 3 Chips / Node. Apply 12V to this string and we get 0.8V which is  off the chart.

It's actually right up at the Max Recommended Core Voltage, but not over the Absolute Max so perhaps it's OK?

That is until you look at the new PSU spec which is "Rated Voltage: 11.60 ~ 13.00V"  Now if we apply 13V to the 15 Nodes we get 0.867V so yes still under the absolute Max but over the recommended.

]

The other "problem" is the W/GH efficiency. At the original 0.66V we have 0.219W/GH add say 14% to that for the PSU & other inefficiencies and you get to the rated 0.25J/GH
However take a guess at the W/GH for a 0.8V Core, say 0.3W/GH and even without adding the 14% you can see we are way off???

So what's going on ? Well assuming this is a 135 Chip system with 15 Chip tripples per string my only thought is that the Data Sheet must be wrong and in practice the Core voltages needed are more like those for the BM1384 used in the S5 which coincidentally has a 15 chip string?

So no firm conclusion, overall 144 Chips works out best if the BM1385 Data Sheet is correct, however 135 chips fit's the maths best, but in that case the Data Sheet has to be wrong. Makes you wonder if there is a third solution waiting to be found?

Let's hope that someone will take the case off and post a photograph?  


Rich
 
UPDATE. The Bitmain Website has been corrected and now shows 135 Chips. So that part of the mystery is solved. What is not solved is the Chip Core Voltage and reconciling that and the J/GH with the BM1385 Data Sheet?

I'm guessing it has to do with quality control.  How they kept having to split into to batches one faster one slower.   I'm guessing they tested this out and can hit target speed instead of 2 speeds.

Price is also locked at 4 BTC so you can't really get a deal by watching btc price and it.

It seem it contains 135 chips, instead 145 (it can be a mistake), so, 45 chip / hashboard...

We'll have to wait and see, but i am also not impressed, a design correction does not sound bad and a 4THs per unit also does not sound bad, but i would of wanted to see a consequential 20% price drop with the 20% hashrate drop.

Instead the unit is as expensive as ever, or more. I know BTC is doing very well right now and people are going hysterical and buying all they can get their hands on, but damn. Batch 4 and 5 were properly priced...

This is not at all what we had guessed. It's not a Lite Version at all. I would call it a Design Correction. We have seen from the very tight voltage spec on the original S7 followed by batches at reduced hash rates that they were having manufacturing problems. This is a more permanent fix that with a 12V supply increases the Core Voltage from 0.66V to 0.75V and should allow reliable running at 600MHz and also I would think some overclocking.

What is uncertain at this point is if they are able to select BM1385 Chips and continue with the 4.86TH Design or if we will see all production move to the 4.05TH Batch 6 style?

Rich

Im not to impressed to be honest.  I was hoping to see 2 blade version.  To go down in speed... and keep 3 hashing boards just is a pain.

I dont think its the "lite" I was hoping they did.  With this I would have rather them it the same on higher speed.

Yes I guess this is what we have been waiting for, although not quite what was expected.  

Looks like it's still 3 Boards but with a reduced chip count / Board. I think they have made a mistake and it will not be 145 Chips but 144.

Which means that each board reduces from 18 Tripple Nodes to 16 Tripple Nodes. Also the voltage range is now a more sensible 11.6V to 13.0V rather than the very tight must not be less than 12V.


Rich

Mr. Lees info was right on leak:

https://bitmaintech.com/productDetail.htm?pid=00020151105082702881F8f0B9I405FF

Price:       4 BTC
               ( 1481.32 USD )

1. Hash Rate: 4.05 TH/s

2. Power Consumption: 1042 W + 10% (at the wall, with APW3, 93% efficiency, 25C ambient temp)

3. Power Efficiency: 0.25 J/GH + 10% (at the wall, with APW3, 93% efficiency, 25°C ambient temp)