The combined sidehack-novak usb stick review thread. AKA GekkoScience BM1384

[philipma1957]

Nothing leaves my bench without the LED flashing properly on shares. I'd suggest checking on the connection state of the small diode inbetween the SOT23 transistors to the left of the LED. That circuit basically acts as a timed pulse generator triggered by the RX to kick on the flash, and if the diode for some reason lost contact you'll get green only.

That diode is one of the hardest parts for the machine to place correctly because it's such a friggin' small package, smaller even than the 0603. It's the littles thing on the board.

The bulk of rework-requirement issues is, I'm fairly certain, due to the solderpaste we used. It got lost in shipping and arrived a few days late, in August, which meant the ice pack it was shipped with to keep the binder from evaporating was mostly a wasted effort and it's pretty dry. Doesn't spread evenly, has no tack, and by the time parts are placed it's basically cakey dust which means adhesion sucks.

The pick-and-place had a bit of a learning curve and took some calibration and such to get going. It was about four days before I had it figured out enough to place a whole panel of 30 sticks, and about two weeks after that before I had things straightened out enough that it could place a whole panel without requiring manual intervention for some things. I'll be optimizing placement speed on the next batch, and with better solder paste and a bit more calibration things should stay pretty nice.

Issues like the shifted capacitors are actually a result of a bit of rework I have to do on the big FETs, partly because the solder paste doesn't stick to the pins without help. I started watching specifically for that, and on the last four hundred or so sticks have been making sure to manually align the caps and add solder when necessary.

We did use a different board house for the full batch. The V0.4 were from a quick-turn prototype outfit, but the production V0.5 were ordered from the same US fab place we get 750W PSU boards from. I wasn't expecting the vias to be so huge, and that actually adds to some of the soldering problems, especially on the big FETs - some of the pads have vias adjacent or internal which wick away melted solder and can leave the pad fairly dry. I'll have to take this into account on future designs.

well I am 20 for 20 with the green so i am not complaining tomorrow I get new psu in for my stud hubs.  direct replacement and eff goes from 78% to 85%

[1715117596]

1715117596

[1715117596]

1715117596

[sidehack]

Gentlemen, meet Max.





Max here is a bit of an experiment. We beefed the main buck input caps from 66uF to 300uF, beefed the output from 770uF to 1240uF, the buck controller bypass from 2.2uF to 10uF and extended the voltage range to about 950mV top-end. Everything else is stock - FETs, inductor, primary heatsink and such.

Then we ran it at 500MHz. It didn't do quite as well as hoped, but honestly I think it's the hub's limitations. Output voltage was sagging, and considering the output currents involved the input cap upgrade wasn't really enough to overcome input impedance of the power leads on the hub (which I'd already beefed previously). Average draw was not measured but probably exceeded 4A; instantaneous burst draws could have been in excess of 20A. I don't think the hub was quite up to the challenge. Reported errors were around 6% and work units were in the 330 range.

It did, however, work just fine at 488MHz.



28 errors in 17 minutes, average hashrate 26.55GH out of 26.8GH expected. Pretty badass.

[hurricandave]

If your willing to experiment an play a little more with these. Think you could build that USB connector without the Positive terminal, but instead, hang a molex pigtail from the pad and power it straight off an ATX???

[Mudbankkeith]

Better still build a molex connector straight onto the PCB then plug in a standard floppy drive cable.
https://www.dropbox.com/s/dpjtcouau6ah0kj/2013-11-02%2012.41.13.jpg?dl=0

[Mikestang]

28 errors in 17 minutes, average hashrate 26.55GH out of 26.8GH expected. Pretty badass.

Super badass!  Max has pushed the limits of what's possible with a stick miner, who would have thought a few months ago that we'd be seeing almost 30GH from a single stick.

[sidehack]

First I want to see how far it'll go with the USB jack in place. I'll build a better USB like TheRealSteve uses for tests. Maybe if I can't push it far enough to melt the ASIC off that I'll do hardwired power.

The inductor will probably start failing before the ASIC does. But I've got a replacement I could scab on rated for at least twice the current.

[Mudbankkeith]

28 errors in 17 minutes, average hashrate 26.55GH out of 26.8GH expected. Pretty badass.

Super badass!  Max has pushed the limits of what's possible with a stick miner, who would have thought a few months ago that we'd be seeing almost 30GH from a single stick.

Thats cool man.    A year and a half ago the chilli was 36 to38 Gh(8chips and a power station)

[sidehack]

Yeah, two years ago I paid $400 for a Blade and pushed it to almost 15GH at about 150W. Now I can get that hashrate from a $25 USB stick and 7W power, or almost twice that hashrate from a somewhat-modified $25 stick and about 20W of power. Next time we get bored we'll rig something up to hit and exceed 500MHz but it'll take some rigging both in hardware and software.

By the way, the voltage can be pencil-modded to increase the upperbound by reducing the resistance of the blank resistor above left of the pot.

[RichBC]

Nice result with Max, what core Voltage were you running? As an aside if an S5 could be run at 488MHz that would be 2.41TH 

Rich

[sidehack]

If an S5 could be run at 488MHz it'd be 1.6TH but it'd also draw about 1KW and require about 13.5V PSU

I didn't scope the voltage, but it was nominally set to 890mV when we fired it up. Multimeter reading was more like 850, which I attribute to ripple probably due to input current limitations. I won't know for sure without scoping. When the voltage was turned up higher the performance actually got worse, which is a clue to me that it's an input current limitation because a higher voltage means a higher current draw. Increasing the inductance would decrease ripple current which would decrease the peak current bursts which might make it a bit better, and increasing input capacitance would help average the input current peaks, but decreasing input impedance will help quite a bit as well by allowing the hub to source more current without voltage dropping out.

[philipma1957]

Nothing leaves my bench without the LED flashing properly on shares. I'd suggest checking on the connection state of the small diode inbetween the SOT23 transistors to the left of the LED. That circuit basically acts as a timed pulse generator triggered by the RX to kick on the flash, and if the diode for some reason lost contact you'll get green only.

That diode is one of the hardest parts for the machine to place correctly because it's such a friggin' small package, smaller even than the 0603. It's the littles thing on the board.

The bulk of rework-requirement issues is, I'm fairly certain, due to the solderpaste we used. It got lost in shipping and arrived a few days late, in August, which meant the ice pack it was shipped with to keep the binder from evaporating was mostly a wasted effort and it's pretty dry. Doesn't spread evenly, has no tack, and by the time parts are placed it's basically cakey dust which means adhesion sucks.

The pick-and-place had a bit of a learning curve and took some calibration and such to get going. It was about four days before I had it figured out enough to place a whole panel of 30 sticks, and about two weeks after that before I had things straightened out enough that it could place a whole panel without requiring manual intervention for some things. I'll be optimizing placement speed on the next batch, and with better solder paste and a bit more calibration things should stay pretty nice.

Issues like the shifted capacitors are actually a result of a bit of rework I have to do on the big FETs, partly because the solder paste doesn't stick to the pins without help. I started watching specifically for that, and on the last four hundred or so sticks have been making sure to manually align the caps and add solder when necessary.

We did use a different board house for the full batch. The V0.4 were from a quick-turn prototype outfit, but the production V0.5 were ordered from the same US fab place we get 750W PSU boards from. I wasn't expecting the vias to be so huge, and that actually adds to some of the soldering problems, especially on the big FETs - some of the pads have vias adjacent or internal which wick away melted solder and can leave the pad fairly dry. I'll have to take this into account on future designs.

well I am 20 for 20 with the green so i am not complaining tomorrow I get new psu in for my stud hubs.  direct replacement and eff goes from 78% to 85%

the 2 hubs were doing 105 watts they dropped to 90 watts. any sticks left maybe i can get some more and fill the hubs up

[RichBC]

If an S5 could be run at 488MHz it'd be 1.6TH but it'd also draw about 1KW and require about 13.5V PSU

I didn't scope the voltage, but it was nominally set to 890mV when we fired it up. Multimeter reading was more like 850, which I attribute to ripple probably due to input current limitations. I won't know for sure without scoping. When the voltage was turned up higher the performance actually got worse, which is a clue to me that it's an input current limitation because a higher voltage means a higher current draw. Increasing the inductance would decrease ripple current which would decrease the peak current bursts which might make it a bit better, and increasing input capacitance would help average the input current peaks, but decreasing input impedance will help quite a bit as well by allowing the hub to source more current without voltage dropping out.

Ooops on the S5 @ 488MHZ. I had quickly typed the Frequency into my spreadsheet that had an S5 with 3 Hash Boards connected...

Yes a lot of power, but a lot easier to turn Server PSU's up to 13.5V than down to 9.5V 


Rich

[sidehack]

But it's a lot harder to take 500W out of an S1 heatsink without your boards bursting into flames. I can only estimate on the efficiency but it's about S3 level, so it'd be safer to just run some S3. Not nearly as fun though.

[sloopy]

Guys, can you tell me what this wire going around the pot is going to do for me, or is doing it already?

Just curious as I didn't see it on the others.

(*warning* fuzzy pic ahead)

[sidehack]

Ah, so you got that stick.

It was mentioned in a sales thread that there was a slight error in the PCB. There's a tiny hair trace which was hidden under the silkscreen layer so I didn't notice it when checking over the final revisions (seriously, it's ten thousandths of an inch square and underneath the silkscreen) that actually shorts the pot to ground. What this does is tie the regulator output to minimum (550mV) instead of giving you full-range control.

You might notice that every stick has a small nick in the board near the center terminal of the pot. This is me cutting that tiny trace. On about two percent of the sticks, my knife slipped and I nicked both the offending trace and the necessary one to which it was attached. That there is the first stick this happened to, and being early in the batch and the first stick with that issue I was in a bit of a rush to fix it and just routed directly around the cut trace with a wire. Some dozen or so other sticks out of the next few hundred also have nick issues, but I fixed them with a bit more finesse - which is to say, a smaller and much less intrusive wire.

Basically, if you bust that wire off your stick will stop working.

[sloopy]

Ah, so you got that stick.

It was mentioned in a sales thread that there was a slight error in the PCB. There's a tiny hair trace which was hidden under the silkscreen layer so I didn't notice it when checking over the final revisions (seriously, it's ten thousandths of an inch square and underneath the silkscreen) that actually shorts the pot to ground. What this does is tie the regulator output to minimum (550mV) instead of giving you full-range control.

You might notice that every stick has a small nick in the board near the center terminal of the pot. This is me cutting that tiny trace. On about two percent of the sticks, my knife slipped and I nicked both the offending trace and the necessary one to which it was attached. That there is the first stick this happened to, and being early in the batch and the first stick with that issue I was in a bit of a rush to fix it and just routed directly around the cut trace with a wire. Some dozen or so other sticks out of the next few hundred also have nick issues, but I fixed them with a bit more finesse - which is to say, a smaller and much less intrusive wire.

Basically, if you bust that wire off your stick will stop working.

What this does is tie the regulator output to minimum (550mV) instead of giving you full-range control.

 if you bust that wire off your stick will stop working.

I am definitely not going to take the wire off heh. Unless you told me it did something like allow me to overclock that stick higher, etc, but saying it is a no no is good enough for me.

I vaguely remember something along the lines of what you mentioned about "instead of giving you full range control". Could you help me understand a bit more there on why I wouldn't want full range control, if it is the voltage the pot controls?

I will go read back through the Sales thread, just being curious.

[sidehack]

You do want full-range control. You very much do want it. At the minimum voltage you'd be lucky if it lit up at 100MHz. They come from us at 610-630mV where they work at 150-200MHz. You can take it up to 800mV where it'll probably work at 425MHz.

[TheRealSteve]

Yeah, I figure the placement&soldering will have improved and continue to improve over time
I'll try poking at the components here and there when I get around to it to see if I can get the LED blinking away.

Review by 'Ken' at one of the reseller shops:

I bought two of these and just ordered another one. They work very well with most pools and can usually switch from pool to pool without having to unplug them. Gives me the ability to play around with various alt coins without a big investment. The LEDs are cool too.

[TheRealSteve]

I'd suggest checking on the connection state of the small diode inbetween the SOT23 transistors to the left of the LED

Finally got around to poking at things - that was indeed it, saves some hunting


Was going for an angle view ( https://i.imgur.com/abKI0X4.png ) but I think the top-down one shows the problem pretty well; the pin wasn't wetted by the solder, probably for the reason you've pointed out before.. paste drying out.

Should be a simple fix with a touch of flux and a heat gun

That diode is one of the hardest parts for the machine to place correctly because it's such a friggin' small package, smaller even than the 0603. It's the littles thing on the board.

Indeed - That text at the top is the tiny "GekkoScience" silkscreen. The smallest I personally ever work with is 0402, which is plenty tiny and uncomfortable enough, and yet far from the smallest passives found in modern mobile devices.

[sidehack]

If I were doing it again I'd buy a larger diode. Actually I might do that anyway since I need to re-buy for the second half of the batch. I haven't tested the new paste yet but honestly it can't be worse than the first batch's stuff.

I am glad that was the problem. By the time 400 sticks passed my bench, there really weren't many issues I couldn't identify on the first try. A lot of signal and power problems with the ASIC I could identify just by how much current it was pulling when I plugged in the stick.