Topic: Pair of s-9s now running photos are up. - page 3. (Read 7738 times)

the number of hardware errors means nothing  the % does  0.0019  means 99.81 % are good.

but as  a suggestion clock at 643.75 for a day

S9 still running her legs on kano, solo.ckpool and zpool/sha256 via APAC MRR relay.

I manage the pools at MRR portal for ease of use.

I am ok with 13.7THs at stock clock.

Yes, the temps needs a bit more cooling but can someone explain to me the HW errors across all boards?

PSU used is the one that came with S7.

Many have advised me to test using a EVGA-1600w to stabilize the HW errors but I am thinking why?

Is Bitmain's PSU that bad to cause HW errors?

Okay - Here is 14hrs @668.75; Avg. 14.28

I had two batch twos on order to not offer me the option to send in 200 for fast track was dumb.

I am sure many batch two buyers would have sent in 100 more.

Yes, it does alter the sound. I was testing my s9 at work and using one of my DPS to power it. While moving it around I found that changing the airflow at the s9 fan intake changed the pitch/sound. Didn't have to restrict the flow, just change it. Then stuck some foam air filter in front of the fan and no more siren.

That extra money they could have made for a fast-track delivery is why I think something went wonky in their parts supply/production schedules with the line/subcontractor doing the b3's finishing first. Sticking strictly to parts allocation and per-batch tagging of sales led to the flub.

I wonder if that would alter the s-9 whistle sound?

My next two batch 2's should arrive soon although still not shipped.

Stupid of them to not offer batch 2's a fast track payment  I would have sent them the 200 extra to have them shipped on the 15th and not the 20th well thats 200 they lost.

I bet hundreds of batch 2 payers would have kicked in more money to quicken the batch 2 shipments.

I figure bitmaintech lost 20,000 by not offering that to the batch two purchases.

I hate to be the manager that did not think of that his boss will be annoyed to see all that money left on the table.

Do be aware that the DPS2k does not have its own fans... What I did for running 3 s5's off of one is put the PSU about an inch away from the intakes of the miners.:

http://i.imgur.com/cf5IRyB.jpg
Works beautifully.
Flippin A... veiwable pic not working so just weblink. sorry.

edit: Speaking of beautiful, while in my Imgur album came across what I did with the Dragon clone A1 rig that Joshua Zipkin of AMT finally sent us lucky few who got anything at all. and sure that pic link works...   Staying with direct link to stay more OT
http://imgur.com/ZRgUk8s.jpg

I decided to try o/c first  656.25 and 662.5 were not promising, but 668.75 has run for 1hr with only 3hw!

Avg. - 14.26

I'll post screen image after ~4 hours. This is with PCB @62/62/64 and die 94/91/97

PSU is DPS-2000BB w/ default voltage. Will be measuring efficiency and trying different input voltages soon.

with the BM-PSU I recycled from the S7 pushes the S9 to 13.6THs atm.

if your dps-2000bb can push 13.9THs, I think the 0.3THs additional hash may not be cost efficient in my case @ 650mhz

I am going to test a EVGA-G2-1600w next on the S9 and see if it has any difference or improvements.

Edit: I may keep this unit since it doesnt have the wooo sound, and the other units will go straight to the hosted place.

my two use these two of these

http://www.ebay.com/itm/IBM-Delta-Electronics-BladeCenter-DPS-2000BB-A-Power-Supply-39Y7359-24R2710-/222128625016?

 and I can get 13.9th  but since it is in a hotter place  I run a bit too hot and I  downclocked

http://www.ebay.com/itm/IBM-BladeCenter-1800W-Power-Supply-P-N-74P4400-FRU-P-N-74P4401-DPS-1600BB-/252423855278?

I ordered this it can do 147.5 amps at 12.2 volts = 1799.5 watts

it is cheap I have a board the works on the dps-2000bb  maybe it works on this one.
if not I will order a dps-2000bb

philip, in your other post at kano..... 1800w PSU? really? EVGA doesnt have that, so we need to explore the IBM/DELL/HP server ones?

I finally found the video!

https://youtu.be/nbQr9NDtIiA

The eVGA units are certainly more expensive, but well worth the 10yr warranty. I had one crap out after two months and they shipped me out a brand new unit, in box with full warranty, in 48hrs. As the video shows, they are demonstrably more efficient than the Bitmain units, and if someone is claiming they are more stable I would not hesitate to make the switch.

well the evga platinum 1600p2  = $$ 350-375 usd

the titanium 1600t2 = $$$ 399-425 usd

but I have a suspicion that the bitmaintech psu is not quite good enough to run these units.

anyone using other than the Bitmain's PSU (I recycled one of the S7 Bitmain PSU for the S9)?

I am reading another thread that using a EVGA PSU resulted better stability on the S9.

Granted that every PSU is a different quality (ie GOLD/PLATINUM, etc), power output should be decent enough if it can power up the S7. Are they any other difference?

my bad phil, i posted on wrong group - its should been in our other thread.

so 13950/14000 =  0.9964   or 0.0036  and your errors are 0.0004  at freq 650

please humor me as you have a really well preforming machine  could you set it at  freq 643.75 vs 650?

I have seen s-7 and s-5 run at 0.0000 error rates with tiny underclock.  If you gear were to do this you may get 13950 hash with 0.0000 and save a tiny bit of power.  other then that your machine seems to be a top notch batch 1 machine.

Just cooler weather here. The last 24hours were 25-27C and the results look the same, other than temps a few degrees higher. Blade 4 accounts for most of the HW errors; it would be nice to be able to clock the blades differently. Dunno if that works with their work generation/collection architecture.



(Batch 1 S9 / DPS-2000BB PSU / 8 Jun Firmware / 650MHz / 100% Fan / 25-27C Ambient / 24hr runtime)

deleted this post does not belong here.


I sold all my dao last week as I had a bad feeling about them.  Sad thing is this is going to hurt  ETH coin.

Maybe I will do some clarification here.

1) All previous chips until now are not able to read chip temperature because of missing temperature probe on chip die.
2) temp reading was done through temp probe at PCB, far away enough from chips to get much lower readings.
3) temp of PCB is extremely inacurate because it can't discover poor heatsink contact and chip overheat. PCB temp vs Chip temp accuracy is extremely affected by frequency (total heat generated) and fan speed. Low freq and low fan speed keeps PCB temp closer to chip temps.
4) generally, all chips follow the same temperature characteristic, so if datasheet says it is safe to operate the chip at 125C, it is true, because it is common to all other similar chips.
5) a lot of previous miners (S5 is the winner) suffered from high power density and poor heatsink contact, so there was big number of failed S5 due to overheating
6) knowing chip die temperature is the key to operate chip safely. Keeping the chip within safe margin keeps the chip running virtually indefinitely.
7) since antminer S5, all chips are able to withstand temperatures about 200C under load (under load is very importatnt). It is from my own experience when chips were running and due to overheat they desoldered themselves while soldering temperature is about 200C. Resoldering chips back was enough to fix the board. BTW: My Radeon 5970 GPU was working at 150C without destruction.


TLDR:

So, reading die temperature is the key to operate chips within safe region. Answers to your questions:

1) why so important: let us know key safety information
2) poor design: Yes, until now. With chip temp reading miners wouldn't fail.
3) are S7 temps lower/higher than S9: With the same PCB readings your chip temps will vary about +-10 degrees on both S7 or S9 miners. This is because PCB temperature is far away from chips and it is also cooled by fan. Depends on fan and frequency speeds.
4) OC suggestion: Keep your chips temp below 125C under any circumstances. Lower is better. From my experience chips can work even higher, but PCB componnents are aging and efficiency drops.

Hope it helped also others