All my rack stuff is on shelves or rails except the lightweight 24-port switch. I don't really trust the torques from front-mounted stuff.

The S4 was a darn good miner. If you get one that can undervolt (I think it was removed later?) they can be made more efficient. Not S7 territory, but approaching S5 at least, and not nearly as loud.

That said, the 135-chip S7 can undervolt to approaching T9/741 efficiency. Honestly I'd stay away from S9 and T9 if possible, they're way too power-dense, the chips run hot and they tend to break down. Spondoolies rack gear, in my experience, had the same problem. Avalon/Canaan builds their stuff more sturdily than Bitmain does, which means a higher upfront price but it'd probably pay itself back in longevity without warranty concerns.

Can you fit 12-14AWG wire in a crimp pin? If you're daisychaining, can you fit 2 12-14AWG wires in a crimp pin? Since you're building a GPU rig, does this question belong in a different section?

Not enough room for both. It's an either/or. This thing is already super busy.

Yeah, and I've got probably close to 200 S1-S5 chassis around the shop just waiting, and probably gonna collect more.

Now I've got some actual good help, and also a couple good guys I can delegate to (VH the driver whizbang, and Optimizer is doing buck design), things are getting done. If I can get closer to caught up on manufacturing I can have more R&D time.

I'm gonna try today to get one of these Terminus pods up to base functionality and cabled into a Pi that it's powering, so I can post a picture or two of the thing in operation. USB-A I think is a good idea.

The only thing that would do is light up anything pulling from the 5VDC line - so, USB chip, fan controller (but not the fans), node-level IO power, LEDs. Probably wouldn't draw more than a hundred milliamps and wouldn't do anything for the heavy-hitter ASICs.

If someone radioed in to say the pod wouldn't turn on from USB power, we'd be forced to openly ridicule that person, which is <sarcasm>very unfortunate</sarcasm>.

The general public being what they are, I have to make concessions to simplify things because most people don't like figuring things out for themselves, or even have the tools to do so. Apparently my upbringing was atypical.

Oh hey, question.

While I'm making changes - what would be better for 5V out (like Pi power), a 4-pin header or a USB A jack? The header is more flexible, but with the jack you could just plug in any USB cable that powers whatever controller you're using.

I posted an update in my Pod Miners hardware thread regarding the most recent revision of the Terminus PCB. Long story short, the support controller doesn't work how I'd like so it's going to take another slight revision to get all the features implemented. To explain how slight, know that I redid it all in the PCB layout program in about 15 minutes.

The end result being, depending on whether or not all the mining parts (which were almost wholly unchanged from the board revision sold in this auction) work as expected, I might have another dozen or so of these pods going out if I can get 'em all to work.

If I do, I'll probably open up another auction thread once I have the final head count, with a more traditional quantity@price bidding, and split the final take with Laura's donation.

The new revision won't be quite as jankety as these, since it integrates the tacked-on buck module by design. The fan mounting will also be better.

First off, it looks like you need the 2Pac support thread since you're running a 2Pac.

Second, why the -scrypt flag? Are you trying to mine litecoins?

Other than that, looks like everything is working as expected.

Oh hey guess what?

Chicken butt.

Also, an actual update.

I realized on Saturday, while sitting on my couch getting nothing accomplished, that the reason I haven't gotten much accomplished with design work lately is because I had been coming in on weekends for that and focusing on manufacturing Monday through Friday, but I've only had two or three weekends since about the first part of May that I've actually been in town and those days I was usually pretty darn beat and ended up staying home, or did come in for a while to dawdle on stuff.

So anyways we're a bit ahead of the game now that I have some decent help, so I took today to start working up Terminus pods. And the verdict is - it's gonna take a few more changes to the PCB to get 'em going proper.

So while I was laying out connections to the microcontroller I forgot a simple but non-obvious fact - one of the GPIO pins is pretty freakin' worthless. So I'm using an 8-pin micro, which means 6 pins of IO. I need 6 pins of IO - specifically, 2 I2C lines, 1 analog in and 3 outputs. So of the 6 IO pins, it turns out one can only be used as an input, not an output. So hey how about I use that as the input? Handy enough. Except oh wait I need ADC input and that's the only pin not tied into the ADC.

So I think I've figured up a half-decent workaround to the problem this causes, and I should be able to combine two of the outputs into a single output with a small additional circuit to handle the second function. But that means redesigning the PCB.

So now I'm left with about twenty Terminus boards whose internal controller won't fully function. It's supposed to handle fan speed and power shutdown off the temp sensor, reset the string when it detects a lockup, stuff like that. Power shutdown cannot work as it is. I started on simple problems first and haven't gotten to temp sensor interfacing yet but that could probably work for fan speed control, but without power shutdown it doesn't matter a whole lot anyways.

So, apparently I start every paragraph with the word "so". Additionally, I may populate one of these boards with the full miner (I started with just the power and base controls), lock the fans on full, wire up an RC reset, basically rolling back the advanced controller feature set to basic. We'd still have the better 5V onboard that can power a Pi, and the 6-pin power jack alongside the barrel, as improvements. And if it works I can have a dozen or so to sell on here as a "beta release" kind of thing if folks are interested, while I wait another two or three weeks on the revised PCBs to arrive.

These guys would be the final form factor, so I'll be doling a few out to resellers so they can get an idea of sizing and mounting requirements for enclosures or anything else.

I believe S9 hash boards are longer. S7LN also has a taller heatsink to fill the internal space and provide better cooling with a slower quieter fan.

Couldn't you just run the S9 with only two boards plugged in, powered by the S7LN PSU?

To comment on this - server PSUs are generally built to plug into a server backplane, and thus have zero PCIE connectors of their own. That's why some of us here build and sell breakout boards that convert the proprietary backplane connector into something usable, usually with anywhere from 6 to 30 PCIe jacks depending on the specific configuration. On top of that, cables can be had which convert one 6-pin to a 6-pin and 6+2 for powering a GPU. I've put together kits in the past for someone to run a PicoPSU and seven GPUs off a 1200W server PSU. I know people who run entire GPU datacenters off server PSUs.

Mostly I do it because I don't have anything else to do. All my friends skipped town, heck even my church closed its doors almost a year ago so I don't really have much else going on. Let's just say I haven't worked a 40-hour week since 2011, but it's okay because I actually like my job.

Hosting is mostly hands-off anyways, and I'm fortunate that customers are patient because sometimes it's a couple days before I can find time to troubleshoot or repair a miner with issues. That's one reason I won't take in GPU rigs, those things take way too much time to keep running stable. That and I don't give a rip about altcoins so BFD.

Depending how I'm feeling on Sunday, I may fetch a pizza and spend the day in the office catching up on design work. Last weekend I was back home helping my grandma run the food stand at the county fair (we were there until 1AM two nights, and the morning inbetween I got roped into pouring concrete at 7AM), and then this whole week has been literally a hundred degrees every day, so I might want to take some time off, but it's likely I'll be out there for at least a while.

The non-manufacture priorities right now are lighting up the BM1384 Terminus pod, which will require about a day's worth of coding after the basic hardware tests, and revamping a Bitfury test board to integrate some new signalling stuff VH and I conjured up. Office work probably means working on the Bitfury test board since that only means PCB layout; Terminus stuff is mostly workbench. Out where it's a hundred degrees. Which sucks. And is also pretty hard on the hosted machines, sorry about that.

Depends which "these" you're referring to - the "these" that I specified wouldn't be available for general purchasing after June 16th, or the "these" that I posted last week I had a few of for sale. For at least one of "these" the answer should be obvious.

I believe the minimum frequency step is 6.25MHz; the chips' PLL doesn't really go more granular than that. --suggest-diff is all positive integers.

The config issue is a bug in stock cgminer 4.10 as was mentioned here a couple months ago.

Well yeah, with a pool diff of 33K.

I can send you one to play with if you like. But as mentioned, even miners pulling 1300W off a 1500W PSU have no trouble, no real temperature elevation even at the contact points. S7, S9 and T9 miners pulling enough power to trip current on the DPS1200 (which is why I shifted them up to 1500W), same. The L3+ could have been trying to pull an extra 500 watts through a single cable pair without killing the PSU and 700 watts through a single cable pair could definitely cause a fire even on a good board. The PSU would have been outputting at least 12.5 volts.

Say whatever you want about KNC. Those guys knew how to get away with selling overpriced fire hazards a year behind schedule.

In general I'd recommend, like Optimizer, to use an over-spec PSU for improved efficiency. But occasionally that over-spec can cause a fire hazard if something else goes wrong and the PSU's overhead keeps it from tripping out before catastrophe occurs.

Seems unlikely, when there's problems on several different units of at least 3 different models of supply.

That's why I recommend using a PSU without as much trip-current overhead, and definitely not my boards on a PSU with more than 500W of overhead, especially not when three different boards on three different PSUs on three different L3+ caught fire.

Maybe heavier copper on Optimizer's boards would hold up to the localized surge currents better, but that just means the cables would fail next if the fault condition isn't fixed.

The best part about using stripped chips is they're all hand-mounted , so I'm manually lining up 0.5mm pitch QFN (no magnifier). Everything else is done on the robot, thank goodness, but I still have to spot-check every single part. I'm considering investing in a better one, 4 heads and a vision system, so it'd get everything done faster and with good accuracy.

My good screens are 17" 1920x1200, homemade; this laptop is 1920x1200 in 15.4". Nice and tidy.

I'm still backed up on 2Pac manufacture but having help is really nice and things are getting done quicker - though hours are limited this week because of the 100+F heatwave making conditions more miserable than usual. But hopefully before long I'll be able to take a day or two and get Terminus pods moving. And hopefully another day or two and get some Bitfury test boards out, sorry everything's way the heck behind schedule.

What ASICs were you running?

The las really well-made Bitmain miner was the S3. Right around that point they started realizing people just wanted the cheapest darn thing that they could keep running for at least 6-8 months. The S5 was sorely underbuilt, and then the S7 and S9 really can't be kept alive long with such extreme power density unless you have fairly cold intake air.

Lower heat density means a lot less heat stress on semiconductors and a lot less mechanical stress on fans. Better build quality means a higher cost, but as noted above, anyone looking for any kind of longevity of investment will appreciate reliability over low initial cost. The efficiency/time curve is starting to level off, and as the diff approaches infinity the percent change should taper off, meaning miners now have a substantially longer viable life than a year or two ago.

I don't recommend DPS1200 for L3+. Leastways I won't sell them to you if I know that's what you're using them on. The first five L3+ that came into my hosting, within a week three of them had set the PSU boards on fire. Looked like there was a flaw with the miner that caused potentially several hundred watts of additional power into one board, because every time the damage was centered around one pair of cables. The 1200 has about 500 watts of headroom over the L3+ stock power draw so it didn't trip overcurrent before the jacks started breaking down from the localized severe overcurrent draw. I could assume it's a flaw in the PCB, except that it happened to 3 of the exact same miner within seven days and not to any other of the same batch of boards, several of which I have powering S9/T9 off 1500W PSUs and they aren't even warm.
The owner of the miners said he saw similar trouble off pairs of Dell 750W, where one PSU would sometimes trip out (less headroom per supply). The L3+ are now driven off pairs of Seasonic 850W and I occasionally have to reset them.

My recommendation for the L3+, use no more than 1000W PSU and keep an eye on it.

My board works with the DPS800, which runs 800W on 120V and 1000W on 200+V, so that might be a viable option. I haven't tested it though.

But for safety I'm also not taking in more L3+ into hosting because I don't want my shop to burn down.