Topic: ANTMINER S7 is available at bitmaintech.com with 4.86TH/s, 0.25J/GH - page 114. (Read 527798 times)

Sounds like a plan.  The low temperature tonight is 32 degrees fahrenheit.  It's in the 30's now.  I have all of mine back to 700 MHz except for one.  It's at 675 with temps at 62, 59 & 61 on all 3 blades.  The other 3 rigs that has higher temps than others are in mid to upper 50's.  I'm wondering if the one's I have with higher temps are as another member mentioned about different fans installed.  I'm thinking about checking out the fans tomorrow to see if there is a difference.  If there is, I'm replacing the less effective fans with the 5,700 RPM Delta fans I have laying around.  They are extras I have on hand to replace fans in S4's if they were to fail.  I sold all of my S3's, S4's, S5's and SP20's.  Well, I still have one S5 remaining.

Agreed, its as if they are aggressively overclocked at factory - which is why I think there will be greater failure.  Higher temps, higher voltages = more burnouts.  Summer may be rough for S7's without underclocking.
Out of my 10 batch 8 Units, I initially underclocked 4 to 650.  The hashboard that failed was only running around 65 - maybe 67 at peak, so I had not underclocked it.  I've downclocked them all now to minimize failure.

Does the Controll board play into the clocking?

i get alot of erros at 600mhs on b6

if i clock to 606 the hash rate just drop to half, anything above that give alot of errors.

underclocvking dont seem to make much difference

That's the main thing I'm referring to... during the warmer wether.  I had to under clock 4 x S7's batch 8 last week with it 72 to 75 degrees outside because they were getting close to 80 Celcius.  It's not issue when it's in the low 50's or below outside.

Part of this had to do with their placement and the lack of grilles in the other door for air flow.  I'm putting 4 new grilles in the door next to the door that already has grilles this weekend.

In the photo below you will see the door on the right has grilles and the door on the left does not.  The rigs on all shelves that are closer to the door without grilles are warmer because heat manages to ricochet off the door without grilles and is sucked into the rigs closest to it.  Once grilles are installed in the other door, the exhaust fan will suck cool air in through those doors and reduce ricocheting into the rigs next to that door.

This mining closet is about to be totally reorganized during the power upgrade I'm getting sometime within the next 3 to 4 weeks.  The IBM 2880 PSU's will be laying flat with a wire shelf resting on top of them.  THEN, I will have the S7's sitting on top of the wire shelf that is on top of the 2880 PSU's.  This way, I can easily have 6 x S7's in front of each grill with a total of 12 x S7's on each of 5 shelves.  So, I will have some work to do when I install 14 more 30 amp outlets and reorganizing the shelves on the rack.  As well as installing 4 x new grilles in the other door.  There are 14 more S7's sitting in another closet waiting for power upgrade.

There are 5 shelves with 4 x S7's on each shelf at the moment.  You cannot see the bottom shelf in this photo.  You can see one S7 without PCI-e cords on a shelf sitting up against another working S7 to prevent warm air from ricocheting off the door into the working S7.  The temperature reduced by 4 degrees Celsius when I placed that rig next to it.

https://i.imgur.com/dW6vj9Y.jpg

Example of the wire shelf that will sit on top of 6 x IBM 2880's then 12 x S7's will sit on top of the wire shelf that is resting on top of the 2880 PSU's:  [By the way, I'm remodeling my home.  So, ignore the speckling on the wall and lack of silicon on the base molding with new paint.  That is new flooring I've installed.

https://i.imgur.com/99pdY0a.jpg

You see the 2880 and 2980 PSU's will have no problem with air flow.  The wire shelving resting on top of them will not impede air flow.

https://i.imgur.com/vtG5IV6.jpg

So go out and buy a small $40 IR temp sensor with a laser thingie. I've been having a ball with KNC Neptune/Titan boards, the hashing chips get so hot when people overclock they melt the solder underneath enough so when they are bumped you get shorts. Chip temp <> board temp...

No wonder there are scorch marks under the chips.

Is anyone doing this yet?

Hmm, but overclocking or downclocking the rig should change nothing on its longevity. It should be merely at what heat they are being run at. Its possible than when the rig show for say 70c, some component is being stressed at 150c+. So lowering the heat one way or another is required to avoid failure.

I really don't know what cause so much hardware failure but afaik if you keep them cool, the clock changes nothing.

Well, the others were factory clocked at 550, 600 and 625.  I say the batch 8 was already over clocked at the factory to 700 and should be at 600 like the batch 6 @ 4.05 TH/s with 135 chips.  I have since under clocked the 3 of mine that run hotter than the others to 650.  I'm thinking about lowering them to 600 [Especially, when summer time arrives] for the longevity for the rigs.

This does not give me much hope in the Batch 8 Order that will be delivered Monday!   I have been hearing quite a few problems with the Batch 8! 

Well lost my 3rd hashboard yesterday.  This was a batch 8 that had been hashing for 12 days.
Unfortunately Powercycle PSU does little.  Hashboard will come on for a few seconds and spew out copious HW errors.
Then it resets itself and hashboard shows as dead.  Opened another ticket with support, but not sure how well thats going to go with zero response still on the 2 batch 5 hashboards in RMA.

I noticed these run much warmer than the previouis batches.  Obviously due to the 700mhz.  I had underclocked a few units that were running hotter than others.  This unit was left at default 700mhz.  I've since set all units to 650 frequency to cool them down a bit (lower to mid 60c's)  In contrast to prior batches these are inferior.  I doubt they will have close to the lifespan the others will given the 700 frequencies.

They are both batch 8 23.11. 2015 shipped takes 1210W with BM APW3... and 23.12. shipped takes 1410W, something very bad happened in that month.
Early bird is little bit on the edge with 700 Mhz but works anyway at that, so I think core voltage is adjusted to later units with hardware or HW mods.

My testing for the very early batch 8 is posted here efficiency were 0,26W per Gh/s 600-700Mhz and if I remember correctly 400Mhz were also about 0,26.
Later batch 8 takes specsed 1293W +10%, with 0C air feed it took 75w less but I didn't get 4700Gh/s 700Mhz as miner was running cold and rpm was 19xx and didn't went slower.
Later is running to some specs but 0,25J/GH is bull.

I'm also curious about efficiency when under clocking with the 2880 & 2980 PSU's without having to use a different PSU to under volt.  This would also help me determine if it will be more cost effective to over clock, keep at stock settings or under clock.  It's important for me to make the right decision for PSU to be used with 200+ S7's in the future.

if it was set by the controller according to requested frequency, I believe that we shouldn't see people having trouble overclocking to 750 or 800 since the voltage would also rise (unless capped in the software).
Anyway, I'll try to get some time to play with the S7 on tuesday so we can get some answers

Thanks for taking the time to do this.  I don't think you realize how much this is helping me.  I'm running numbers for future revenue projections and making necessary decisions on when to move to get access to cheaper power, etc..  My goal is to acquire 200+ S7's by block halving.

As has been said, if the efficiency is significantly different then that requires a different core voltage. What I am not sure about is where the numbers shown are coming from?

The Batch 6 spec was 600MHz, 4.05TH/s, 1042W, 0.25J/GH
The Batch 8 spec was 700MHz, 4.73TH/s, 1293W, 0.25J/GH

If you do the maths yourself on the above TH/s & W you get

Batch 6 0.257J/GH
Batch 8 0.274J/GH

Will however be very interesting to see if the Buck Converter voltage is different between Batch 6 & 8. Also if it is Factory set or controlled by the Controller?


Rich

If the efficency is really different, it's very likely that the buck converter is set to another voltage.
I'm back home. Will answer PM, quote requests, arrange a few deliveries first.
Then, I'll look more in depth at what can be done on this buck converter.
Better efficiency should be easy by setting the output voltage slightly lower.
Overvolting might be tricky because the MOSFETs could already be close to their limits at 800MHz. I didn't check the data sheets.
(my batch 8 ran without any issue 10 days at 800, I set it back to 700 so it's not too loud now that we are back.

Just to make sure we are on the same page:

Batch 6 was factory set at 600 MHz for 4.06 TH/s with 135 Chips
Batch 8 was factory set at 700 MHz for 4.73 TH/s with 135 Chips

Hence the .25 watts versus .3 watts because of Batch 8 clocked at a higher frequency.

My question is, can I under clock Batch 8 down to 600 MHz and get the same .25 watts efficiency as the Batch 6 at 600 MHz?

I stocked on up the S5/S7 fans a while back. I sold about 10 of them already and have I think 10 or 12 left. They are brand new. Never used. I am in the US, so I would only ship in the US. If you want to buy any message me. I do charge a premium, but you can get these a lot cheaper if you look around a little, but more risk.

Default S5/S7 Fan
http://www.nidecpg.com/fanpdfs/va450dcf.pdf

The fan easily comes off with 4 screws. Remove the fan. Put the fan on top of the foam sheet and outline the fan. Cut away the excess foam. Install foam template onto miner then the fan. Install the fan with the 4 screws. Some times by just having that small gap between the foam fan and the miner frame makes a huge difference in sound. Hey, I agree, you shouldn't have to be doing any of this to a miner that cost that much money, but this is what a bitcoin miner does. This is an example of the foam. Also, difference thickness produces better/worse sound(s) from fan. This did not affect fan performance.

http://www.amazon.com/Foam-Sheets-2-Inch--2-Inch-50-Pack-Rainbow/dp/B00284NN82/ref=sr_1_3?ie=UTF8&qid=1451681628&sr=8-3&keywords=foam+crafts

I'll watch the thread for a bit. I've got the tools and probably some skills, but it's a balance: If Bitmain is doing RMAs then of course the best option is to go with them.

However if anyone has a fireballed S7 or one that is not rma-able due to "dirt" or something like that they want to send to me to research on, drop me a PM.

As I say on my Neptune/Titan threads, what's the worst that can happen? :-)

sorry, others already answered (135). So far B6 is doing well at 625 (~4.18Th average) with low power use (~0.26 J/Gh).
Re underclocking...I am sure that it does not pay to underclock right now from the profitability point of view. However, machine reliability is a whole another story. I had difficulty pushing B6 past 650, so I am amazed that they made essentially the same machine (B8) work (mostly) at 700.
Someone had made a spreadsheet in SP20 thread that calculated the optimal speed vs profitability, but in Sp20 there was a big difference in efficiency vs speed. S7 is quite profitable right now (if you are not taking hardware cost into consideration) based just on electricity usage. However, with current difficulty increases, it might run out of steam at 0.1/kwh by summer. That's where downclocking might give it a month or two extra and then they will be sold to super low electricity cost areas.