About the pics of the replacement boards they sent you... What's with 2 of them having a very deliberate looking pattern of no heatsinks on the PCB side? Related to if they are supposed to be center or outside boards?

Also note that that lil' ferrite I mentioned earlier is properly placed on all the replacements.

More OT, about ROI...
Current price for the batch 12 s9 @ 12.9THs gives $123 per THs. PSU not included.
The R4 @ $1k and 8.6 THs is $116 per THs... PSU included.
Hmm.

edit: Rechecked their gallery - shows a separate PSU.

My guess (hope) is that Chip Temp is the highest reported on the board. Given the simplicity of on-die chip sensing and the space afforded @ 16nm node it would be shear stupidity not to have each report its temperature... Hell, TSMC probably provided that IP block "free" of charge.

To me board level sensors are really only good for an indication of temps any large caps have to live with. For that, since BM claims to use ones rated for >135C, board temps under 70C are perfect and yes lower = better.

As for 'what is safe' for the chips.... Without BM providing a data sheet our only ref stating a MAX chip temp is the one for the BM''84 or '85 chips. Is 120c I believe

Since the s9 manual says to not exceed 115C, fine. Fits with the fact that at the 16nm node the effects of temp on the gate performance in the chips is not to be ignored and sets constraints on the window for allowable temp vs performance. Not to mention how the PCB and solder joints respond to it. However, if anticipated in the chip design, silicon can perform quite well even over 150C.

Is that or anything over 100 maybe 105C 'safe'.... Maybe.
If the temps are very stable, say less than a 5C swing in any 24hr period so the mechanical effects of thermal cycle are minimal. I say -- should be.

Actually not bad price for plug-and-play all the way from 110v up to 240v.
But 7 pairs of PCIe, making only 14 total? Rather odd number no? I would have gone with 20 for the option of 2 s9/s7's (barely) @ 240v or 1 s9/couple undervolted s7's @ 110v in.

edit: I really should start using more of the IBM DPS2000 supplies (with your breakouts & cables) I have. Right now only have 1 in use. Just that getting the Bitmain ones got to be too convenient since if I needed one I ordered the same time as getting a miner.
Still use the one setup like I did for my old s5's - just set the PSU an inch or so away from the the intakes so the miners do the airflow required. No fans needed.

Took another look at the pic on Bitmains site and yer right - it shows only 2 cables running from the controller to the hash boards. So, 2 hashboards slid into the partitions guiding air and giving nice support.

Well if the more common 19" racks are out, there is always the EIA 24" wide ones used in the AV industry...
edit: just looked, 610mm is 24.01568" so even EIA racks are out. But still, I see a lot of aftermarket case possibilities for this. Or branded casing from the retail chains. It *is* targeted at the home market after all so maybe just a couple per household. Ya know -- a mass-consumer Bling item

miner software authorship aside -- I've been waiting to see something like this from Bitmain or others. Long over due. Give it a snazzy looking case to look like AV gear and they will have a winner.

Reason: I'd bet beans to btc that they are eying the large retail store market. More specifically the ones in Asia as a whole at least to start.
The main hypermarket stores to a Western eye are -- insanely huge. Think Walmarts, large furniture/bedroom store, massive consumer electronics sections, groceries and eat-there, you name it. The one I visited in Korea (I think was a Lotte Mart in Daegue) even had their own moped and car dealership inside. All in one 8-story building.

In late 2014 before he cut all communications, Joshua Zipkin now aka Joshua Alexander was talking about him drumming up extreme interest from a Korean chain for his last-ditch efforts to keep AMT alive. Well at least be able to finally dump the scads of A1 chips he had...

It should be in the Skype AMT room chats link in my sig, somewhere around maybe late July or so.

Entirely different power densities involved so simply not possible.
Ja in a dreamers world diff would stay low so hash rates needed would stay low as well negating the need for today's zippy and power hungry ASICs. Blame Satoshi. He knew full well how technology would respond to diff increasing.

vs doing it how else?
Underclocking so top sinks not needed? 1 big top sink milled as needed for clearances?

Critique implies knowing a better or at least different way of doing something.

At least the pic from Bitmain answered the question about "missing PIN".
Let's substitute the word FIN, it fits better with um, component with fins on it.... aka heatsinks.

Since posting the pics, am surprised no one has mentioned the square gray component (looks like a ferrite of some sort) that is seen under the s9 name on the board. Ya know - the one that is definitely not placed squarely on the board and looks damn close to shorting itself out?

The declared value part...They claim if you want faster processing by Customs - keep the declared value low. I can understand that as it is the same for goods coming into the USA. In the past many folks here sorta pushed Bitmain to that 'suggestion'. Problem is: as Phillipma said you are also setting oneself up to take a big hit if a missing/damaged good claim has to be filed.

For controls -- really can't be beat. Programming is just choosing boxes as in/out and stringing them along.
Box A + box B, etc drive result (coil) Y.
Repeat with as many lines as needed.
Math functions or PID loops now just boxes with fill-in the blanks for variables.
All in fixed scan time, usually under 100us these days.

Ya know... All of this just screams to me get a small PLC from Automation direct.com and lose all the falderall. Something such as http://www.automationdirect.com/adc/Shopping/Catalog/Programmable_Controllers

It's um, part of the voodoo I do so well...

Cheap (ish). Utterly reliable. hugely expandable with standardized modules. Can integrate other sensors such as fire/smoke/temps, etc. Can be a web-server if remote is desired. Ja is programmed in ladder logics but so what? After all for the most part is is just a mass of switches and coils in software. very linear and obvious programming as in X1 in controls Cwhatever

Thinking cap on. Always dangerous...
If I do it for other than myself -- will provide switched 24vdc to contactor coils for power loads.
DISCLAIMER of liability: What and how YOU wire said contactors to is beyond my control (not liable for idiots and chicken-wire/duct tape types). Consult a licensed electrician.

Also will provide low power sink or source IO in banks of either 16 or 32 channels per module.

Touch screen panel for it? Maybe.

Hmm. how much leakage current can your DC_ON input take before it turns on?

Ust 2 wires per PSU and opto output assuming simple NPN:
the AUX 3 or 5v (also have seen 12v) from PSU to opto collector
Emitter from opto to you breakout.
Instant opto ON = Vsource ON  fed to the breakout but depending on the load a breakout presents, possibly not always opto OFF = exactly Vsource OFF.

Actually, I thinking of/referring to using a badge on the um, call it it 'Hobbyist" CPU board with the optos on it to drive yers or other breakouts. Just use 2 jumpers feeding the LED side to select active hi or low per bank of 4 in the opto chip. edit: or have a software switch for active high low...

Dunna have to solder to the PSU pins like you saw on the HP's you got with the CARE package from me.

edit 2: Ah. Brain fart. You mean your break out is a sink and want a source to drive it. Back to jumpers to select source/sink at the optos, this time perhaps feeding the typical +5vdc always-on aux voltage the psu provides to each one to maintain isolation? Still just makes at worst a 3-wire connection between the breakout and opto board.

no 220 = 240 for all intents and purposes
not the bitmaintech comes with 10 wires and will run the s-9.

so if you have 220 you are good.
[/quote]
[
I will add that in a commercial 3-phase setup all the above run perfectly on the typical >208VAC you get across the phases. Most of my ~18 or so kw farm at work runs on it.

Ja. Just have the GPIO drive some quad opto-isolators (preferably 16-pin DIP and socket so they can be replaced) and wire their outputs to the DC_ON terminals. That way the only common between the PSU's is their chassis ground so no cross-talk concerns.

Search the RasPi/Audrino/BeagleBone sites for a lot of home automation projects that do just this kind of thing. Why reinvent the wheel?

Most of the projects only take the power load so far (typically 15A) but no reason you can't use their outputs to drive heftier contactors. For that matter, all server PSU's have a low power (5vdc) toggle to turn the DC on and off...

So far all s9s good here including the 2 new batch 11's. Have all my fingers, toes and eyes crossed hoping they stay that way...
Any yes, is a bvyche typng liker that...

I have a Skype msg out to a buddy of mine to see if he has the time/interest to look at the firmware file(s) for the s9. Just waiting for him to show up.

Even by my standard the dude is by far a brilliant Dr. of Computer Science and does keep Blockchain discussions alive where he currently works (major Investments Firm). Leads a group whose job is to explore new ways of mining Big Data for investments modeling and making front ends for clients to use for perusing/manipulating the huge amounts of raw data being collected about, well, everything nowadays....

Anyway, between him and opieum2 from here (IT security) I figure that if something suspect is going on -- they will find it. Been a long time (AMT days) since opieum2 has been on here but I'll ask.

Ja. But, Phillipa said that he has also seen this on s7's and s5's which has code that can be checked.

Closest to I've seen to this (on a s7) is one board-shutting down for 15min to an hour and then coming back to life on its own. I have 2 s7's that do this every so often -- as in maybe once every few days to over a week or more between the board apparently taking a nap for a while then getting back to work.

The oddest thing about is that M'sMiner Monitor and Kano miner stats will show the 1/3rd drop in real-time speed while the average shows what it should be though after 15min or so Kano will start reporting the drop in average. Bitmains GUI reports the same drop in RT and average as normal. Also shows all boards/chips are alive. Do a soft or hard reboot and all is as it should be. Very odd...