Have the two S3+'s up running after a comedy of errors, mine.  They're running now something under 10 minutes, they've settled out at 461.12 & 464.57 @ 225.  Almost tempts me to kick the S3's up to 225 but I know the one problem device will be difficult.  One of the errors I made was to kill power while one was booting - I've found that to cause software errors later and that machine isn't giving an identity after the MAC address to nmap.  No biggie.  

The two miners are perfect externally, looking good.  Haven't opened them yet.  Will eventually to put wifi antennae.

Waiting since early this morning for my pair of S3+'s.  UPS, signature required.  It's 4PM.  Just got a call from a neighbor.  They have two packages addressed to me on their porch since around noon.

I don't get why they don't have both fans run the same speed as dictated by the exhaust temperature.  To have the intake fan run slower because it's cooler means, I'm guessing here, less efficient air flow.  If the slower fan is getting help from the draw of the faster fan it's just a bigger load on the faster fan.

The problem S3 which I changed out the thermal paste was down about 430 this morning.  I had moved both into the back room last night as after a warm day it got uncomfortable in my living room.  So, I restarted the problem S3 this morning and it has been getting warm in there, 91° 31%rh.  It has been running well since rebooting, the warmer room and low humidity.  6hr50min 440.91GH/s, 0 HW, but utility is 15, fewer shares tho the score seem equal to the other.  It's temps read 47 & 46 while the #1 unit is reading 44 & 42, running at 441.60 HW 160 in 15 hours but its utility is 16.

Having learned the art of building heatsinks for high voltage FETS in the late 1970's, what you want for perfection is two perfect surfaces in contact but since perfection isn't possible you want as little thermal paste as possible filling any air gaps between the surfaces.

When I took the inner heatsink off a hashing board, a good board by the way, the thermal paste seemed applied by machine in 4 rows along the top ASICs and 4 more rows along the bottom ASICs.  It was not well spread, just flattened.  There were air gaps between the rows.  Thermal paste conducts heat less well than the copper on the board or the aluminum.  So, essentially the heat went from the copper plane to the rows of thermal paste to the aluminum on the other side of the thermal paste.  I removed it all and applied an even thin layer of thermal paste to the surface and tightened from the center out.  When doing the outer heatsink I suggest you carefully remove it after removing the 4 spring'd screws but not to slide the heatsink while removing.  If you lift it up and off you won't build up thermal compound on the very fine ASIC pins as might happen if you slide the heatsink to the side.  I'm sure the stuff doesn't conduct but dust might adhere to it at the pins over time.  Big mistake is thinking more paste more conduction.

Also, as you remove the silver thermal paste from the inner heatsink, you may notice a couple or more small lumps of solid thermal compound stuck to the heatsink.  At least that was the case of the bad hashing board that had new paste applied in addition to the original.  No way that heatsink could have been flush, it stood away from the copper plane at least the depth of the hard lumps.

Is that really true?  The hashrate reflects an hourly average updated every 10 minutes.  So if you put a 44GH/s miner on toward the end of an hour, the average will be very low.  I don't really see what I'd call an accurate hashrate there until a move to the next block.

Yes, noon yesterday Slush was hashing at 6,527 and now 9,678 and share payout went from .004xxxxx to .003xxxxx.

S3+'s ordered the 19th, shipped to arrive by end of day Monday.

Figured out the different box size thing.  Florida S3 reseller got a return from a dissatisfied customer who marked the S3's box well with something like RETURN DEFECTIVE.  He couldn't just turn around and ship it to someone else because of the box markings but he had a box from an S1 or perhaps an S2 with its packing material.  He then sold the return as new in the used box.

I agree.  More likely guys with miner farms, get a dog of a unit, resell it quick, spread rumors under a different name that it was run at the factory for a long time.  

Still, they would have to be factory tested and a question might be how long they test the units for.

The one I got from Florida compared to the first one I got from the mid-west - both advertised new on Amazon.  The first was fine, the one from Florida not so much and that repackaged in a larger box.  Both S3's.  Perhaps he had a box from an S2 or earlier that was in best condition.  I'll have a better idea when my S3+'s arrive.   There are many creeps out there in the bitcoin world.

If the S3+'s come in the size box I got from the mid-west, I'll feel more sure giving a bad report to Amazon when they ask my opinion.

As in an earlier post, I thought the unit probably came up slow in testing and was sent to troubleshooting where new paste was applied and the screws tightened grossly tight on the bad hashing board thus bringing it up to "passing".  That would have been at the factory.  In industry we don't trash a new unit just because it isn't up to snuff.  As a matter of fact that troubleshooting of new failed units was one of my responsibilities at the last place I worked.  Of the two boards on that unit, the good hashing board had thermal paste that was less new, more dried, than on the questionable board.  That it may have been previously sold by the reseller and returned then shipped again is a possibility given what I see now as the condition of the plastic tape.  When my two S3+'s arrive I'll know if the box comes with clear tape or just stapled.  The USPO sometimes puts clear tape if a box isn't sealed with tape.  This has priority mail tape but the plastic tape under it had been opened and not by me.

A manufacturing facility needs to run new units as there's what's called an infant mortality rate.  Odd term to use in electronics but it's true.  Testing doesn't always run long enough to test but the failure rate drops sharply.

I just finished ripping #2 apart again and have just fired it up.  Both are now running in the cooler room side by side same height from the floor, same distance from the wall, both being hit by air from a 20" fan.  I've /etc/init.d/cgminer restart on both units.

The reason I ripped #2 down again is that the grossly overtightened screws had 2 that had tops partly stripped.  I had put those on the lower corners when putting the heatsink back on and couldn't tighten very well.  I later noticed that the controller board used the same screws.  I just swapped those two lower corner screws for two from the controller board.  At the same time I swapped the two fans taking care to make sure the air flow direction is correct.

#1 50min14sec, 443.59GH/s, Hw 6
#2 49min52sec 439.11GH/s, HW 0
------------
#1 2hr6min 445.00 GH/s (avg) HW 18
#2 2hr5min 439.97 GH/s (avg) HW 0

Yet there would be signs the tape had been removed and replaced.  There was quite a bit of additional USPS Priority Mail tape but there was no signs of repackaging that jumped out at me.  What I find curious is the Bitmain box dimensions difference.  Was about to measure the boxes and looked at the second first, this time more closely, and the bottom had been opened and resealed.  

#1 room temp 79°, 442.64GH/s(avg) HW 151, 15hrs 49min
#2 room temp 83°, 428.83GH/s(avg) HW 39, 245 (it just made a liar of me), 15hrs 53 min.

And no, when I buy a "new" Bitmain Antminer S3 on Amazon I do not expect the seller to have been hashing with it.  I expect to be sent a new miner untested by the reseller.

I bought two S3's from reseller's here in the US.  The first unit upon firing up for the first time started giving a smell of some electronic part cooking but that cleared after a few minutes.  I suspect that's normal for a brand new unit that has only been run long enough to check it's running okay.  The second S3, was in a somewhat larger box and looked the same.  No cooking electronics smell and I was looking for it.  That one never runs quite up to the standard set by the first.  It's much more prone to falling hashrate with temperature rise.  I took it apart and changed the thermal compound.  One of the two boards had screws that were grossly overtightened.  This second one may have failed its initial test and was subject to troubleshooting where a technician really mashed down the heatsink hold-down screws to improve the thermal paste contact which had been reapplied fresher and more liberally than on the first board.  I must restate how that second now runs.  Since removing and reapplying thermal paste, it runs with fewer hardware errors than the first.  I may remove and replace the thermal paste on the first but it runs well so I haven't yet.  I'd say you're lucky you smelled that new electronics smell.

So, the S3+ is the S3 with a firmware upgrade or is there a hardware mod as well or is that information not available?

I have noticed bursts of hash rate significantly over 500 even though my frequencies shouldn't get the rate that high as I'd get an X in the row of os when using the higher frequencies  one is on 237.5 and the other 231.25. just spotted one claiming 657.
[/quote]

Yes, I see the version number in the download URL.

Wonder what frequency the S3+'s will be shipped as.  Another 11 days or sooner....  And transit will take how long?

Haven't left for shopping yet, busy, but have some stats.

#1 53minutes (58m slow typist)
     436.51GH/s(avg)
     getworks 117
     accepted 911
     rejected 0
     HW 20
     Utility 16
     discarded 4,596
     stale 0
     localwork 703407
     WU 6,122
     DiffA 351659
     BestShare 1,327,130
     load 1.39 1.36 1.29

#2 53m (59min)
     438.76GH/s
     getworks 115
     accepted 1020
     rejected 0
     HW 33
     utility 18
     discarded 5,016
     stale 0
     LocalWork 705,714
     wu 6,143
     DiffA 258,344
     BestShare 205,663
     load 1.32 1.29 1.25

Pool score about the same for each
Pool hashrate a bit lower for #2 but both look good and running well.



    

    






#2

Reading the last few pages I opted to install the latest cgminer on what had been the problem S3, #2.  Before doing so I ran opkg list-installed | grep cgminer on both S3's so as to ascertain the present version.  Both had 3.12.0-1.  I downloaded the latest version on #2 and ran opkg list-installed again.  Right now I'm guessing that installing via wget and rebooting doesn't update the installed listing as after installation of the latest cgminer the installed list still showed 3.12.0-1.  Since that has the lower loads it could be the more efficient miner.

I will install the latest cgminer on #1 and check back a little later today.

The poolside hashrate show #1 to be lower now, GUI 438.38, # 1 having the higher load (2.46) and hardware errors.  #1, cat5 connected, seems to have rebooted 22 minutes ago for no reason, now has a low hardware error count (of course only running 22 minutes) of 5, DIFF 449, DiffA 165394, DiffR 0, DiffS 0.

#2 poolside hashrate higher by 11.7%, GUI 440.28, wifi connected, lower load (1.39), DIFF 367, DiffA 6,986,294, DiffR 5505, DiffS 0.

#2 has new thermal paste applied yesterday.  I'll be picking more up today and might rip #1 apart to change.  Again, I'll download the cgminer to #1 and get back later with new stats.

May I ask what determines load?  I have two Antminers,

#1 using cat5 connection and its wifi connection disabled has:
loads of 2.70, 2.56, 2.51,
hashrate 441.03,
DIFF 354,
Fan/Temp 3000/41 & 2220/41 room
Temp/Humidity 83°/-%;

and #2 at some distance running via wifi has:
loads of 1.62, 1.47. 1.42,
hashrate 440.80,
DIFF 367, 
Fan/Temp 2040/40 & 1860/39,
room Temp/Humidity 80°/39%.

So the first has loads 147% higher than the second?  Both S3's at stock frequency.

The new thermal paste seems to have helped.  What is helping more is reducing the wifi load.  With the storm I had put the living room Ant1 on wifi and disconnected the cat5 connection.  The backroom Ant2 can only run on wifi.  The combination of better thermal paste application, the cleaning of ASIC pins, and the reduced wifi load with Ant1 back on cat5 now allows Ant2 to be hashing at 439.11GH/s(avg) and slowly climbing.  Then again it might be that the temperature and humidity back there is now 83° and 42% rh.

Shut it down and stretched the springs on the hashing board that had overly tightened screws.   Up now 47 minutes, 426(5s) 428(avg) all ASICs o's but the miner is in a non-AC room and the humidity there is 43% rh tho temperature fairly low, 83°.  We had rain.  Perhaps a mass air flow sensor should be incorporated.  Wish the controller board had better air flow over its top.

I note that the hashing board on the left while facing the RJ45 port seems to have been a problem.  The screws from the board to the inside heatsink were grossly overtightened to the point 2 were almost impossible to remove.  When I got the heatsink off there was more thermal paste than on the the other hashing board.  The other hashing board has paste that was more dry.  This second with the over tightened screws had fresh paste.

I just worked on the second, over tightened screws board again.  Cleaning the ASIC of paste on its pins and closing it up, it's been up a little over 5 minutes but running slow, 432.03GH/s(5s) 418.33GH/s(avg) but all ASICs are showing o's.

I'm having a problem with an S3 I got from Florida.  It consistently runs slow.  I've just replaced the thermal paste, front and rear of both boards, and a single ASIC is malfunctioning.  I'd like to know how the o's are positioned relative to the actual ASICs.  For instance, looking at the miner from the RJ45 port end, is the top row of o's the right board?  And then is the first o the ASIC closest to the RJ45 port then progressively around so the last o in the line is the ASIC beneath the ASIC closest to the RJ45 port.  If this is the case then does the lower line of o's have the first o representing the top ASIC at the far end of the left board, farthest away from the RJ45 port?  If not, how?  Thanks.

soy