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

[philipma1957]

Sometime soon I figure on pushing one up to 300MHz or better and see how it handles, see what the input current looks like. We did some rough comparisons today on heatsinks from some other miners and figure with moderate airflow the thing should be good up to about 6.5W chip heat, which would happen around 1.5A input current, which I should touch 300MHz before hitting that point. I think.

Unless one of y'all want to test that first. The only thing I know for sure is the buck's good for it. That's only about eight and a half amps output, pretty small potatoes for that circuit.

I may give a high power test tomorrow.  I have not passed .92 amps and freq 225.

the sticks got a bit warm.

I have some very good fans and maybe I will try really blowing  them out on friday.

I will use the usb bridge. that can go close to 1.8 amps at 4.8volts which is  over 7 watts

[philipma1957]

Just started higher overclock tests.

fullsize links

http://imageshack.com/a/img537/1732/b0a3Um.jpg
http://imageshack.com/a/img537/4849/w8BWvF.jpg
http://imageshack.com/a/img661/2889/CaCuoh.jpg
http://imageshack.com/a/img540/2382/b1LPCZ.jpg


only doing 1 stick. Since I have 2 if I do something wrong I will have one more left.
note my hub can do .9 amps per usb port. so I have a bridge  thus I can give a stick 1.8amps.  most hubs will not do this for you. note cooling fan

this is the original model I am doing freq 250.  it is doing 1.07 amps to get to 250 freq



the hub and the stick pull 8-9 watts on the kwatt



here is the bat file and the early cgminer numbers


note temp is 75c humidity is 62%

[philipma1957]

after some 1 stick testing at freq I moved to 2 stick testing,

full size images

http://imageshack.com/a/img673/3618/YXxKFG.jpg
http://imageshack.com/a/img905/876/feIAPf.jpg
http://imageshack.com/a/img633/9744/6Ef7U7.jpg
http://imageshack.com/a/img673/6385/HuAc8k.jpg
http://imageshack.com/a/img673/2489/tuh3vI.jpg
http://imageshack.com/a/img661/3984/LY7aLz.jpg

they use the  usb bridge and they use 1.11amps and 4.81 volts each the gold is the new one the black is a tiny bit taller and is the older one



the hub and 2 sticks use 14 watts



early cgminer numbers are  here


temp was 77f but shot is fuzzy

better temp photo


cgminer after an hour or so . 14 watts used to do 27 gh counting the 4 or 5 watts of the hub.

so 9 or 10 watts  to do 27gh at the sticks

[philipma1957]

If it enumerated as a COM port, you need to install the WinUSB driver for hotplug support and such (Zadig) which will allow cgminer to detect it natively. You'll also want to include "--au3-freq xxx" in the command line, where "xxx" is a valid operating frequency. According to the code I'm seeing (AntMiner U3 is under the Icarus driver), valid frequencies (and their associated per-chip GH/s) are:

100	5.5
125	6.88
150	8.25
175	9.63
193.75	10.66
196.88	10.83
200	11
206.25	11.34
212.5	11.69
218.75	12.03
225	12.38
237.5	13.06
243.75	13.41
250	13.75

The command line I use is "cgminer.exe -o stratum.mining.eligius.st:3334 -u 1BURGERAXHH6Yi6LRybRJK7ybEm5m5HwTr -p x --au3-volt 0 --au3-freq 150"

The --au3-volt is not really necessary, since the voltage is not software-adjustable on the Compac.

I have hit a dead end on testing.  I got freq 250 to load after that cgminer crashes.
I suspect I need to get into a file and add freq beyond 250, but my skill set with code is that of cut and paste.
I need to see a screen shot to copy what to do.

Sticks work well up to freq 250 I needed to get a better fan but in an 80f room temp in the garage I can hold the sticks with my hand and while warm they will not burn me. I found a fan that moves more air and I could  keep the sticks cooler by turning the fan up.
The black heatsink older stick likes 1.07 amps   4.81 volts
The gold heatsink newer stick like   1.12 amps  4.81 volts

the black one  has less errors and has just over 14gh
the gold one has more errors and just under 13.8gh

so 10.534 watts for 27.8 = 0.378 watts a gh.

not surprising the watts per gh drifted up a bit due to the 250 freq.
the variation in these two sticks is about 5%  but that is at top speed.
could be the newer gold one can go lower maybe a stable 100 freq my black heatsink stick was not stable at freq 100. it works as low as freq125

Still 2 sticks and close to 28gh.

[sidehack]

Yeah, I had forgotten the U3 code only has up to 250MHz in it. The actual code shouldn't be too hard to modify to add additional frequencies, it'd be dropped into the Icarus driver.

The black-heatsink stick was made with one of our new-from-Bitmain sample chips, the gold one actually has a pulled chip. The power dissipation also increases with heat, so the smaller heatsink on the gold stick probably adds to that slightly.

[philipma1957]

Yeah, I had forgotten the U3 code only has up to 250MHz in it. The actual code shouldn't be too hard to modify to add additional frequencies, it'd be dropped into the Icarus driver.

The black-heatsink stick was made with one of our new-from-Bitmain sample chips, the gold one actually has a pulled chip. The power dissipation also increases with heat, so the smaller heatsink on the gold stick probably adds to that slightly.

still 250 is pretty good.  someone that  knows how to mod the Icarus will need to give me a few screen shots on that.

I am pretty sure that with a quality high speed fan these will do more then freq 250  and 13.8 or 13.9 gh.

I will run them overnight at freq 250 then back down to 218.75

[TheRealSteve]

I think you'd need to adjust driver-icarus.c:

Code:

#define ANT_U3_MAXFREQ 250

Change to 450

Code:

} u3freqtable[] = {
	{ 100,		0x0783 },
	{ 125,		0x0983 }, 
[...]
	{ 243.75,	0x1306 },
	{ 250,		0x0982 }, 

Extend that with:

Code:

{ 275, 		0x0a82 },
{ 300,		0x0b82 },
{ 325,		0x0c82 },
{ 350,		0x0d82 },
{ 375,		0x0e82 },
{ 400,		0x08f2 },
{ 425,		0x0801 },
{ 450,		0x0881 },

( values from datasheet - there's a formula if you want to try other values*  If bfgminer works feed it hex values directly )

The function anu3_find_freqhex basically does a look-up of the frequency given on the command line in the table without exceeding it.  So supplying it with '166' will make it use the '150' freq.
The function set_anu_freq sends the frequency on over and then it's wait and see if things freak out or not

* No idea if making it use anu_find_freqhex instead wouldn't work.  Can't spot a reason for using a LUT for the U3 in the cgminer commits.

[sidehack]

The code I wrote to help with S1 undervolting (http://gekkoscience.com/misc/set_freq.cpp) should print out valid hex for the nearest possible setpoint to any arbitrary frequency, if you want more granularity than 25MHz steps just add whatever lines you want to it.

[philipma1957]

I think you'd need to adjust driver-icarus.c:

Code:

#define ANT_U3_MAXFREQ 250

Change to 450

Code:

} u3freqtable[] = {
	{ 100,		0x0783 },
	{ 125,		0x0983 }, 
[...]
	{ 243.75,	0x1306 },
	{ 250,		0x0982 }, 

Extend that with:

Code:

{ 275, 		0x0a82 },
{ 300,		0x0b82 },
{ 325,		0x0c82 },
{ 350,		0x0d82 },
{ 375,		0x0e82 },
{ 400,		0x08f2 },
{ 425,		0x0801 },
{ 450,		0x0881 },

( values from datasheet - there's a formula if you want to try other values*  If bfgminer works feed it hex values directly )

The function anu3_find_freqhex basically does a look-up of the frequency given on the command line in the table without exceeding it.  So supplying it with '166' will make it use the '150' freq.
The function set_anu_freq sends the frequency on over and then it's wait and see if things freak out or not

* No idea if making it use anu_find_freqhex instead wouldn't work.  Can't spot a reason for using a LUT for the U3 in the cgminer commits.

Okay so lets say I want to get my working version of cgminer 4.9.2 to add  for the sake of argument

freq 256.25
freq  262.50
freq   268.75
freq  275.00
freq  281.25
freq  287.50
freq  293.75
freq 300.00

which I THINK MY SETUP AND THE USB STICKS CAN DO.

I do not understand how to splice it into the  cgminer 4.9.2 I have.

give me a few minutes to show a screen shot or two. 

[philipma1957]

here is a screenshot of my 4.9.2

fullsize link


http://imageshack.com/a/img540/3146/eN85t7.png



My coding skills are poor I can do a  start bat file and not much else.

[sidehack]

Borrowed from an S1 thread, clock and hex:

256.25  1405
262.50  0a01
268.75  1505
275.00  0a81
281.25  1605
287.50  0b01
293.75  1705
300.00  0580

You're going to want to find driver-icarus.c. Change the ANT_U3_MAXFREQ to 300 and add these to the u3freqtable around line 400 then recompile. Doing that on Windows is going to be less than fun.

I've got my sticks on a Pi at the shop but I can ssh into it. I'll see if I can do it there.

[philipma1957]

Borrowed from an S1 thread, clock and hex:

256.25  1405
262.50  0a01
268.75  1505
275.00  0a81
281.25  1605
287.50  0b01
293.75  1705
300.00  0580

You're going to want to find driver-icarus.c. Change the ANT_U3_MAXFREQ to 300 and add these to the u3freqtable around line 400 then recompile. Doing that on Windows is going to be less than fun.

I've got my sticks on a Pi at the shop but I can ssh into it. I'll see if I can do it there.

may be over my head.  but I will keep trying a few ideas.

[sidehack]

So I added a bunch of frequencies to driver-icarus. You also have to modify a function in a few places in cgminer.c to take the accepted max up from 250 to 300.

I added that into the current git code (4.9.2) Recompiled and all that, cgminer starts and recognizes --au3-freq above 250 but doesn't see sticks. Of course I didn't test compile the stock code to see if it worked before I made my changes. Whoops.

[TheRealSteve]

Hey, at least you got that far   I haven't compiled cgminer on windows in a very long time; the instructions are outdated, several things have changed in various locations, and just when I thought it was getting somewhere... I'm hitting make complaining about the size of struct timespec being unknown
( cross-compiling on a linux install is much more fun )

[sidehack]

Actually, I'm on an XP laptop in my living room, SSH'd into a Debian server at the shop, re-SSH'd into a Pi with Minera on a shelf in the office, and building it on that.

Which cgminer 4.9.2 built straight from git isn't picking up my sticks either. But now the stock cgminer isn't picking them up either? Dunno what's up.

[TheRealSteve]

Actually, I'm on an XP laptop in my living room, SSH'd into a Debian server at the shop, re-SSH'd into a Pi with Minera on a shelf in the office, and building it on that.

Must. Go. Deeper.

In the mean time, I got mine to build.  This took well over 100 steps (admittedly, this includes clicking trivial buttons and such - was documenting it in screenshots just in case) and a whole bunch of googling and crossing fingers.. and the particular error I was facing semi-last (there were two more after that) is actually a MinGW issue that should be fixed soon.

For reasons I haven't explored yet, I also ended up needing 'libgcc_s_dw2-1.dll', but then it started right up and fired off on a Block Erupter USB without issue (so far - not a fan of the zadiggery and tend to run older cgminer).

@sidehack: what are the functions that need tweaking in addition to the define and LUT?

Edit: I suspect it's the set_float_100_to_250 cap on input parameters?

[sidehack]

Yeah, that's the one.

I won't have any more progress until tomorrow when I get into the shop. Not sure if something in the compile scripts hosed USB or if the hub's smoked out or what but I got no sticks even on the cgminer already built into minera, which actually was still working after the first unsuccessful rebuild from git (in a different location).

I'll talk to Novak tomorrow and see what he thinks, since linux and programming are both more his thing. It's funny, I'm the one with a CS degree to his Aero but he's the far better programmer.

[philipma1957]

TheRealSteve sent me a cgminer build. 

Will test it in the morning.

Thanks steve.

BTW these sticks float along at freq 250

30 plus hours  one has 2 or 3 hw's the 5 or 6.

Stats are

4.73v 1.12amps

 4.73v  1.07amps

I got a second meter. I will post a photo in a few mins.

 

[TheRealSteve]

Hopefully no magic smoke was released in an untimely coincidence.  I'll see about documenting the cgminer build (mis-)steps somewhere (not this thread, way too o/t).  If anybody wants to play with my binary, shoot me a PM - or just wait for sidehack/novak's build

[philipma1957]

Hopefully no magic smoke was released in an untimely coincidence.  I'll see about documenting the cgminer build (mis-)steps somewhere (not this thread, way too o/t).  If anybody wants to play with my binary, shoot me a PM - or just wait for sidehack/novak's build

Not trying the build yet but photos of my gear at freq 250.  I am impressed with these little sticks.

say 2.2 amps x 4.75 volts =  9.9 watts giving 27.46gh  this is 0.36 watts per gh at a high overclock.


http://imageshack.com/a/img907/1672/EqSGyN.jpg
http://imageshack.com/a/img908/3190/27EUgh.jpg
http://imageshack.com/a/img673/7530/VDJCEv.jpg
http://imageshack.com/a/img673/766/zfie6A.jpg


so in an 82 f garage  and a 140 mm pair of fans these sticks are merely warm

my pc running  cgminer and bitcoin and the hub running the 2 sticks use 46-47 watts


the 2 sticks and the hub use 16 watts



the cgminer has run more then a ½ day at this setting  8HW FOR ONE AND 2HW FOR THE OTHER!!
pretty good.