Topic: ANTMINER S5: 1155GH(+OverClock Potential), In Stock $0.319/GH & 0.51W/GH - page 41. (Read 451032 times)

Agreed its pushing the connector even at stock for an S5 (if only 1 is connected per board).
On the other hand, the bolded is actually reverse to what you say. If the ambient is warmer / hotter you'll probably get more HW errors than run the risk of pushing the power draw on the connector(s), but have a cooler ambient and you'll get fewer HW errors and more power draw as the chips do not get reset.

Dinna know that it was part of OpenWRT. Figured it was open-source bit they tacked on.

As for bandwidth... was thinking more of getting hard data for if I move my 30TH farm at work to using a 4G HotSpot. Being in an industrial park the available internet solutions are limited mainly to Multiple T1 lines and they are pricey as well as shared with VOIP for the phones. That farm puts a sizable dent in the b/w we have at work...

Extrapolating from the S3 data I have I puts total monthly usage at a ballpark value of 8GB/month. How well does bandwidth usage scale with TH/s from the S3's across the S4 & s5 miners?

 SP20 limits power to 288 watts - which is identical to the spec for the connectors, probably deliberately.

 S5 running 300 on one connector per board is STUPID - that IS overstressing the connector. There's a REASON Bitmain insisted on putting 2 on each board (well, yes, overclocking, but even at stock clock they're pushing the spec on the connector pretty hard). I'd also bet that it's a rare thing to see.
 S5 at stock clock or below is within the spec of the connector (official 590 watts "at the wall" is more like 540ish or a little less DC, so 270ish per connector) but even running a stock clock S5 on one connector per board is pushing your luck a bit. Will probably work most of the time as long as the S5 is in a fairly cool room, don't bet on it in summer heat in most areas without A/C of some sort.

 I'll also point out comments and complaints about burnt connectors are fairly COMMON on the higher end KnC machines which run about 300 per connector.

The S1 and S3 use a slightly modified version of OpenWRT which had that feature built in, the other miners use a heavily customized variant of Angstrom Linux so it would have to be completely rewritten for that. Bandwidth however should be very low so it shouldn't be something you need to worry about.

The one request I have for Bitmain is put the network usage graphing back in like the S1 and S3's had! Since it is pretty much part and parcel with the Pi's and BeagleBone OS/s anyway please bring it back. Would be nice to see what the bw of these will be (mainly for planning G3/G4 WiFi Hotspot usage).

Tasty... Me likey

It just screams using a pair of the IBM 2kw supplies in load share which is easy-peasy using the breakouts from GekkoScience. Since I still just happen to still a couple  of each sitting idle and a ton of leads I'll be all set for one.

Only problem is will have to retire some S3's though as I'm at my free power limit of 15kw at work

If instead of a form factor of 3 S5 you had a "S5+' with the single form factor, then you would be left with a more expected upgrade to a S5, that still use different frame and heatsinks, enclosure, etc.

The difference would be small, a bit like the S3 and the S3+. It would have pretty much the same efficiency, just more under volted chips. Ultimately more heat crammed into the same space.

Ok have it your way.... But let's just look at this a different way. If S5+ hash boards were made available, as opposed to a complete 3 Bay S5 which is of no use to a Home Miner, then assuming firmware & interfacing to an S1 controller board was possible, which I suspect is straightforward as the interface to all of the BM series of chips is almost identical, then I think there would be a lot of interest. 

Don't want to quarrel too much about this point, but this monday i'll get into one of my mines, there's one S5 connector that came from the manufacturer with a clearly burnt PCB near the connector, still running fine, but you wouldn't expect that to come when running as the spec says.

---

Not all connectors are built equal, there's also the fact that not every manufacturer has certifications.


Indeed, this is one of the worries i have. Is this meant to have 30ºc+ ambient temp like the S5, or has to have a cooler setting? Also, push-pull was not required on the S5, why does this one needs it? The shroud also was disregarded as unnecesary some time ago, is this new design an acknowledgement of the mistake on the S5? Not trolling, just want some fab. acknowledgement plz, i can build a shroud quickly for my S5 units and enhance cooling, but i lack the skills to determine if it really helps or not (my gut says it is a waste of time, as i generally trust bitmain's engineering).

You were suggesting an S1 to S5+ upgrade. If you need to change the frame, the boards, the heatsink and add fans and a controller adapter. In the end the only thing you would be keeping is a single fan and maybe the s1 controller. So what exactly would you be upgrading?

I just bet its not possible even if they technically wanted to, since it would be just a disassembled package of a complete S5+, minus a fan?

The upgrade kit was designed for just shipping the new board, as its light, without having to reship the heatsink, the heavy parts, etc.

The Molex spec on the CONNECTOR commonly used for PCI-E 6-pin connectons states 8 amps per contact. 3 contacts for +12v = 24 amps = 288 watts per connector (the ground connectors carries the SAME current as a different part of the same circuit, so you can't count those into the total).

 190 watts per connector is actually fairly conservative based on the CONNECTOR specs.

 The PCI-E spec is actually INCREDIBLY conservative on usage of the connectors for both the 6-pin (75 watts) and the 8-pin (150 watts) varients.

 The PCI-E 8 pin connectors have the SAME limit as the 6 pin, as they STILL only use 3 pins for +12 - the extra 2 pins are BOTH ground, which IMO is just dumb.


 300 watts per connector like SOME miners needed (certain melt stuff fire-hazard KnC designs come immediately to mind) is stupid.



 The actual contacts used in those connectors are specified at 11 amps max, but that's an open-air rating and has to be derated when used in a multi-pin connector due to heat retention of the connector itself unless you WANT to melt the connector and end up probably shorting stuff out.



 I don't think "most" folks run an S5 on 2 PCI-E connectors - you pushing their limits hard stock clock, and OVERSTRESSING them if you overclock significantly with a SERIOUS risk of failure. You're be seeing a lot more "KnC Neptune / Titan" meltdown type complaints if this idea was common. Also, MOST power supply PCI-E cables aren't designed to handle that load per wire which would be a common issue even if NOT overclocking.


 The curved fin heat sinks aren't significantly IF ANY more efficient than a straight blade design with the same fin thickness in the same overall depth. A "tapered fin" design is MORE efficient than either, due to better heat conduction at the base evening out the heat transfer along the length of the fin - that's the design my one "oddball" recent used S5 has, and probably cools better despite the actual fin length being a hair shorter than the older "curved fin" design.


 If this thing is real, I'd STILL be worried about the heat dissipation - the airflow through the unit is NOT a good setup, though it appears that the heatsinks on the hashing chips themselves are bigger than most of these pictures suggest.

No, no its not. You're saying 190W PCI-E connectors are a fire risk when pretty much every minor ever has run at that level or higher. The ASICMiner Cube ran 225W when on 4, the SP20 300W at stock, even the S5 runs 300W when on 2.

I was not suggesting that the S1 heatsink would be used, but that the S5+ hash board would be supplied complete with heatsinks. As to the controller an S1 to S3 converter was relatively straightforward, so why not an S1 to S5+ converter? Yes you need a lot of air pressure to cool, but no more than the S5+ requires  and fans are relatively cheap.

I also realise that this is not going to happen, but I bet technically it could if they wanted to. 

Look at the picture, the chips arent being cooled the same was as before, the S1 heatsink would probably not be of use. You now need to put alot of pressure and air flow on the individual aluminium heatsink and there would not be any with the way the S1 is designed.

They used different connectors for the S5 boards from those on the S3 (and different from the S1 to S3 which introduced the controller board).
S5 PCB board to controller cables are 2 rows of nine pins, whereas the S3 PCB to controller connectors are 2 rows of 10 pins (and as you may have guessed, different sizes). The connectors on the monster look similar those on the S5 though.

9 boards -> 27 PCI-e + controller board which clearly has 1 PCI-e connector = 28 PCI-e connectors.

Running with just 2 PCI-e connectors is the irresponsible thing to do, no need to have a fire risk when mining.

I can see that an additional fan would be needed but why a new enclosure and why could the controller not be upgraded? Surely the interface to and driving any of the BM series of chips cannot not be that different?

You would most likely need a new controller, the boards, more fans and an enclosure and then just maybe you would be okay for dissipating 1200watts of power.

I think it would be great if the could offer the S5+ hash boards as an S1 upgrade with new firmware for the controller board?

Rich

It would be wonderful if Bitmain might consider offering separate controller boards if some of us want to buy one of these upcoming monsters to split into 3 different units for home mining.

Alternately, I wonder if I could repurpose old S3 controllers to drive these things.