Honestly this is as much of a keep my skills/brain in shape sort of thing as it is a business. There's no way I could recoup the time spent trying to figure some of this stuff out and I've been fortunate enough to have a platform that actually has value over time. Trying to learn how to fix an Antminer is kind of pointless; they go obsolete within a few months of release, not worth the effort.

I can fix GPUs, but there's not enough of a market in that. Thought about the Innos, but they seem to be a slightly odd bunch and I'll deal with that if it happens.

C

Didn't see your reply dltrader, just a quote.

Meantime found a new and interesting failure: This controller board that just came in was doing the flashing white light, but instead of a short flash it was a longer one, which was odd. Hooked up a Neptune BBB to it and did some diagnostics. Sure enough the TPS chip was coming up but failing on PGOOD timeout, which was new.

So I tried removing the display board. Controller comes up, all is well. Oddly enough it looks like the display SPI driver circuit is bad; that's a new one to me, normally the displays just go dead without shutting down the board. Repair averted, board is fine, sending it back....

On to the next project.

Truth is the signal lines that short are on the outside edge of the CPU chip, right along the left and right sides. If I could jack the chip up a mm, I could get under it with some desoldering braid and pull off the solder around them. I would probably short the die's + and - supplies but who really gives a damn if the die is dead.

Trying this on some Neptune boards just to see what happens....

Marry Christmas everyone. Been busy working on a number of things, got some cards repaired, can fix just about everything except the shorts on pins 6/8.

The only way to fix that seems to be to replace the hashing chip. Not going to happen. But everything else has been mapped out and can be fixed; shorted controllers, dead FPGAs, even cubes with no dies.

I just wish KNC had not routed the power supply line to the chip control circuits through a via *under* the die. If there had been any other place one could isolate or cut it we would be in business. But no, drat.Moral: Route power and SPI signaling to your engines through a 0 ohm resistor. Would cost .0001c per board.

Meantime everything else is purring along. How long will Titans keep running I wonder.

Depends. Stocks to a large extent; I bought a fair bit in 2008 and tripled my winnings on the economic recovery. Now that the orange dingbat is in power I'm moving about 40% into cash reserves, then deciding what to do with that. The other 60% can stay in a mix of global technology (hell, the rest of the world can do all the tech stuff we will sink to mine coal or something) and basic dividend growth stocks (think toilet paper/stuff you need no matter what).

Bitcoin and gold are a nice counterbalance. I've been earning it for awhile now (and paying taxes in fiat on every satoshi). To date I don't think I have ever bought a bitcoin; not sure how actually. :-)

C

That is the big problem: For a retirement account to accumulate tax free it typically has to be specific funds at a brokerage. And the magic, GOD STRAIGHT truth applies in that case:

If you don't own the keys to your bitcoins
YOU DO NOT OWN BITCOINS!

Likewise there's nothing preventing a future govt from screwing you if they own the keys (this not happen? Hell, in the Phillipines you now have authorized death squads. Things can get weird). Likewise moving the money out of the current 401k/IRA's would hit me with a 30%-40% penalty.

Things can get really weird when weird people get into power and the people want weirdmageddon.

C

Ok if you're still having problems:

How many cubes do you have?
How many controllers do you have?
Are your controllers upgraded to the latest Neptune firmware?
Are all of the controllers showing a green status light?

How many cubes are generating heat (feel the exhaust air temperature)?

C

I keep thinking about transferring a chunk of my retirement funds into Bitcoin. Would be a good investment, especially with the recent currency manipulation being done by govts.

900?

Yes. I think the problem is the later model B+ chips use a new GPU/IPL loader that does not support Wheezy. This is a real bitch.

Question: Does anyone have a blown out bridgeboard that I can buy? I have just about everything working on the qberty boards but I need to move a component on an original BB to check a trace and I don't want to wreck my reference bridgeboard to do it. Will pay .1btc plus shipping.

Meantime more boards repaired, few more brought back to life. Don't run them too hot; don't burn out the plugs.

Clever. I was wondering what they used as a micro-controller to manage the data transfer. Oddly enough it's the CPU on the bottom card that manages the transfer. Everything else on the card is a dunsel.

Sure it's quickly built but let's face it: Sounds like they want to fix this problem ASAP. My guess is for the production units they order the chips flashed at the factory before mounting on the board. Saves a big step in production.

Dude, you're a God. Next time I'm in VA I'll msg you for a beer (root beer or otherwise). That would be the way to do it.

That would not be unusual: KNC did this with the Neptunes, and when they made the Titans they followed the exact same design. Because if the previous design works in the field why change it?

Let's move these discussions back to the main thread though; this is for fixing the A4's.

At this point you're reading way way way too much into this :-) But someone is already into the chip, and it is a possibility.

C

My guess is they are 1850 type power controllers. I remember those on the Jalapeno and BFL 60gh units, they're actually not too bad but pretty simple. Two channel support, they watch dv/dt across the upper FETs to determine current but they drive both FETs from the same gate source. One mistake BFL did was to use three high side and 3 low side FETs. The high side FETs are on only a very small period of time, but need to start up and shut down *QUICKLY* which requires a low capacatance gate trench. By putting three of them on the high side they increased the gate capacitance and they ran much hotter than they would have with only two. Low side tends to be on a lot more so capacitance isn't an issue and running three is a good idea.

Sorry, technobabble :-)


Good. Milling down an 8 bit 1990's controller MCU was enough of a bitch. :-) They may be busy, but see if you can reflash it with a 3.0 code base. What's the worst that can happen?

C

Oh you can pull the firmware? Then they didn't set the write protect bit, *good*.

I wonder if you could find someone to send you one blade from a 3.0 unit, then try pulling that firmware and flashing it to your 1.0/2.0 board. Or ask Inno to send you the firmware. I was thinking actually of building a "hat" that would go on the chip with a Pi behind it to just flash away but it sounds like Inno has the jig.

Edit: Looking at that jig I'm wondering why they have two sets of USB devices on there. Are they reprogramming one chip or two?

C

That would be a very good idea. As someone who has repaired a bunch of reworks all at once awhile back time is critical. It takes time to pull the unit out, remove the screws, take off the fans, not lose the screws, pull the boards, and place on the preheater (or jig in this case). Then putting it together. The time spent doing that is miniscule to the user, but when you multiply it by 500 you're talking serious time.

People should pull out their blades, and send just those in. Make sure Inno is ok with this, if someone botches the removal then the boards should be replaced under warranty or something if Inno approves. (It's an exception case, document those).

It sounds like Inno wants to just do one jig, and to be honest the only skill I could offer with the jig is my kinda-good reputation :-). Anyone should be able to do the reprogramming, LS will do a great job on it I'm sure. If it required reflow/rework/fun skills then I'd put my hat in but for now it seems to be under control.

Yup. At this point they seem to be thinking along the lines of sending some jigs to the US and having customers pass them around instead of swapping the MCUs. That could be one way to do it.

If I was doing reflow work, I'd charge .2 to .25 btc or so per board to swap the mcu's. Reason being:

Takes about 10 minutes to unpack, place, flux, and pre heat a board.
10 minutes to pull the MCU and cool the board down
10 minutes to place the new chip, heat up, and reflow the new component
30 minutes to hook it up and test it to make sure it works then re-box it.

So, hour a board. At .3btc per hour (my hourly billing rate for reflow work) that's a bit of a bargain but I can probably do two boards in parallel on the IR preheater.

For that I would also need a few 1.0 boards from Inno plus a bag of chips to get the process documented and down to a science.

What might make sense is to have the big mining shops get the jigs and reflash their dozens of boards, but the one-two offs send it over here to be swapped. Or they can send me a jig and I'll do it for .1btc per board to cover the time for overhead.

Meantime what's the part number on that MCU and anyone have a gen 1 and 3 board they want to risk?