Topic: The combined sidehack-novak usb stick review thread. AKA GekkoScience BM1384 - page 4. (Read 26372 times)

I had this happen on a ck build of cg 4.9.2 running 1 u2 on solo.ckpool.org

so it is not your build and it is not the sidehack stick.

since I was using a u2  and ck's stock build from his download of 4.9.2

but it is an antminer product to the fault may be with antminer chip.  note the u2 is the 1382 chip

and the sidehack stick is the 1384 chip. 

I am running the sidehacks on mmpool has not hit a block in 2 months

I am running the u2 on solo.ckpool has not hit a block in a few days.

if either pool hits a block and this happens I will post back.

Well, cgminer (my build - shouldn't matter, but just putting that out there) eventually hit a bump (apparently about an hour ago).  Pool sent a new block notification and cgminer exited with "Failed to create restart thread".  Number of USB resets was just under 50 - and none were on the screen before the error - so one potential culprint -ck mentions in a thread doesn't seem like it should apply (though I only count USB resets due to no valid hashes, and not restart_threads() counts, so who knows)

Regardless, not a problem with the stick itself - I've set it to hum along comfortably at 200MHz = 11Gh/s so I can do some other testing at some point
Edit: I should point my Block Erupter USBleep at it so next time I'll know it stopped because things stopped beeping..incessantly..and annoyingly.

Very nice. I couldn't even get my U3 to hash for that long!  Okay, sometimes it would but i'd rather have this bad boy running.

Still hashing away very stable after 1 day, 10 hours (123456 seconds because why not).  Tamb=24.5°C, Tdev=39.0°C.  I still think the USB resets ('X' figure in that screenshot) are software-avoidable, related to it expecting a U3, but I'm not touching that code any time soon

yeah  they are hard to mount j4bbrwocky   has been having troubles with it.

I think they could be a decent 1 chip LTC/scrypt miner since the scrypt side pulls less power.

I also think people may not like that the usb stick only mines scrypt and the btc/sha-256 was disabled.

It's also a 50W BGA chip built mostly for scrypt and I've stated numerous times that I don't want to work with either, and especially not both.

Sfards gear according to their website  runs best at all LTC or all BTC  not to sure it is efficient enough to bother with.

Or even SFards.  I would LOVE to see something from GekkoScience that does not cost the rumored 1200 dollars.   

The great news is GekkoScience has gained a following, and rightfully so.  So i'm hoping this will make companies more willing to let them use chips.  Bitmain letting them get some then being quiet is curious.  I hope bitmain provides them some more chips whether it be S5 or what ever is next.  We have seen Sidehack can do a heck of a lot already with current gen chips.

I'm hoping so. Bitmain said they'd let me know when they have more chips, and I'm really hoping they'll leak some info on new stuff but honestly I don't expect to know anything until after whatever official announcement they might have planned. But be sure we'll have cash in hand for sample chips the day we're able to buy some. It's been over a year since I had a project as fun as the stickminer, and the full-scale stuff could become some real business if it works as well as I hope.

Well maybe some new 1386 s-7 chips turn up  and you get to make a good 18 chip board this fall.

So would I. Designing the thing is no problem, but being able to mass-produce is a whole different matter (mostly of financing materials, since we'd have all the tools required).

Honestly he did a good job.  I really want them to succeed and I think they will.  They built a sold product my review is still running.

I really hope they keep making gear and have a series.  I would LOVE to see a large hashing unit some point in future when chips are able to be secured easier.

Good stuff right there TheRealSteve. I feel like with all the info the reviewers have provided, and constant updates from sidehack I already have my compac! 

Eeeeeexcellent  Looks like you were typing that up while I bumped my thread - added the link.

If there's anybody else who would like to set up a group buy for their region, feel absolutely free to use any of the pictures from my reviews, and/or this one with the heat sink closer to what it should be (with any luck).

So, this just happened.

https://bitcointalksearch.org/topic/run-2-closedsidehack-stickgekkoscience-compac-official-sales-thread-1126705


Yep. I put up an official sales thread with instructions and timelines and stuff.

Ah, not insulted at all - it's just a test setup and in part finding these edge cases is what it was about.  Means if somebody does come back to me with a gifted Compac saying it's making beeping sounds, doesn't seem to do anything, etc., I know what to tell them

Thank you for allowing my StickMiners thread to not die just yet - and thanks again to Bitmain for supplying the chips they had.

Yeah, depending on inline meters and such there could be a high-impedance (if only for AC) connection keeping burst current from getting through and dropping the voltage below a usable threshold momentarily. I didn't mean to insult your setup. I also tend to forget that regular people don't always have nice things, or know how to turn crappy things into nice things, so expecting a typical customer to have a USB hub actually capable of maintaining 5V under a 1.5A draw, while seemingly reasonable to tech people, doesn't take into account the reality that most people buy the cheapest crappiest black box capable of meeting immediate needs and don't know how to make it work any better. Which is, in a way, another reason to hold off on building Amitas.

Also, thanks for testing so extensively.


Additionally also, I just rechecked the datasheet. 4.4V is the minimum turn-on, and it's got a lot of hysteresis - the turn-off drop voltage is 3.9 typical.

Re: drop/singing - never had it go below 4.4.  Long before that things would drop out, bumping it back up, drop out again, etc.  It's likely a combination of factors (the various meters and boards in between, the generic power brick, etc.) or even this particular Compac
Will have to test at some point with a more direct setup, but at some point it's less of a stick miner and more of a PSU+HUB that also happens to mine   I'm quite happy with the performance you can still get off of a USB2.0 / 3.0 port, actually.  Hm, and I just noticed I didn't upload the right picture for that - fixed now (just showing 500mA and 900mA cut-off lines).

Good times.

I think sometime soon I'm going to set up my pair of sticks on ambient cooling here in the office (no fans) and push frequencies higher until one of them explodes, just to see what happens. If these things will ride 300MHz with no airflow in a 75F room would be pretty good to know.

Also, it'll take some software tweaking but I should have a way to drop that current burst on chip init, which especially at higher frequencies can be pretty bad. Unfortunately since there's no good way for cgminer to distinguish between the Compac and U3, we might have to make a Compac-specific cgminer with mods to the base U3 code that could make it not work with the U3 if we want the best performance from our Compac. As it is it works pretty good so I'm not too worried about it, but it's stuff we need to test anyway.

Maybe I've been using too good power supplies, or I overbuilt my rebuilt hub too well (it's got star-connected power leads per port, each with a local 470uF cap), because even with nine sticks initializing at 225MHz (which just the running current dropped my Vin to about 4.7, not sure what init current dropped it to) I had zero chatter on the buck. My pair of 300MHz sticks are running off what's apparently a pretty strong ATX, where the no-load 5V read in as 5.2
The input UVLO on the buck is internally set to 4.4V typical, so if your supply lines drop below that it power-cycles the buck. That could be the problem. If your supply lines drop below 4.4V it's fairly likely that at least one part of your setup sucks - either the hub's power or the hub itself. Or maybe the USB jack terminals are crusty and making a high-impedance connection.

Nudging things slightly back on topic with some testing.

Here's a GekkoScience Compac hashing away at 16.5Gh/s (300MHz clock):
[youtube.com]

I made a few adjustments to cgminer to show a bit more detail, give an estimate of Work efficiency, time elapsed since last reset/zero, etc.

That info was then used across several frequencies, measuring all the things, and doing further calculations (getting the effective hash rate and effect J/Gh, for example.)  The test setup has pretty much been unchanged throughout other than adding a power supply later on.  The fan is positioned 10cm above the Compac, so keep that in mind when looking at the temperature deltas; I wouldn't want to leave it passively cooled at the higher frequencies

Note: There's a knee in Vcore in the graph until I test those 'tween frequencies

I did run into a few occasions where the comms chip wasn't doing anything - power draw was pretty much near-zero* - and had to unplug/replug (switch on hub port makes this nice and easy).
* The green LED appears to be using ~3mA.  The blue/orange ones together add another ~6mA.  This affects efficiency quite minimally, but I'm still glad I taped that LED over a bit.

If all went well, and it would initialize and hash normally, potatoscope showed pretty much this (Iusb (dimensionless because what's a current probe? Can derive from graphs above though) at Vcore=~550mV):


If instead it failed to initialize, typically due to too low a Vcore, it would show this (Vcore=~550mV - yes, every once in a while it would be okay):

In that unclean state, more often than not, I had to power cycle the board to get it up and running again.

The other failure mode is in the upper regions of the clock frequency.  As more current gets drawn, cheap power supplies drop voltage quite a bit, and the power driver on the board would freak out a bit, producing a nice singing inductor coil:
[youtube.com]
Something to add to my little booklet's FAQ when I get around to updating it for the production versions (would love to know what changes are made there other than pot direction).  Solution is to use a better power supply or at least an adjustable one so you can bring Vusb back up.  In fact, on several tests, I had to bring Vusb up above 5V to prevent the singing.

As the clock frequency increases, the range in which Vcore will work and Vusb becomes finnicky narrows quite a bit; that's why the little graphs also become more narrow

Suffice to say that I haven't pushed it beyond 0.38J/[email protected]/s at this time, and haven't yet tweaked things subtly enough (a multi-turn pot would be nice) to get below the 0.339J/[email protected]/s I've gotten, those are pretty impressive stats for a StickMiner