[GUIDE] BitFury Miner Support/Tuning

[tom99]

  Nice heatsink from ebay http://www.ebay.com/itm/100x140x12-7mm-Aluminum-Heatsink-for-Electronic-Computer-Electric-equipment-H157-/181110341808?ssPageName=ADME:L:OU:CA:3160

[1714201693]

1714201693

[1714201693]

1714201693

[1714201693]

1714201693

[1714201693]

1714201693

[klondike_bar]

Can you make a photo where to measure ... the current ... ?

beside each chip there is a bank of 8 small capacitors. most of these (seems like half of them in my situation) carry the voltage to the chip (0.73-0.85v = 1.5-2.7GH depending on tuning) and can be easily checked with a multimeter. I usually check at least 2 caps to verify that all chips receive similar voltage and ensure accuracy.

I find that on startup from cold pcb i have around 0.805v, which climbs to 0.825v after the system is running and getting toasty over an hour or so. during this time, error rates decrease slightly and hashrate climbs to optimal speeds of about 40-41GH for the board

ps: current (amperage) is different from voltage  

Does 40GH require you to use heat sinks or just fans blowing across the cards?

i have heatsinks in use, and plan to upgrade them to slightly larger ones (35x35x6mm) soon.

however, today i have had some issues in the past 3 hours when i lost hashrate at ghash.io and system reboot and start/stop miner ommands seem to only be temporary fixes where it returns to 40Gh, then slowly drops. error rates are within reason (slightly higher than normal), and chip 1 has 3 miso errors that dont seem to be replicating in any other chips. (typically it had 1 per 5min sample before)

Its hard to troubleshoot without >20min of data, but i might have to check and/or back down the voltage slightly as im unsure whether the issue is the card or the pool (the initial problem seems to correlate with my blockerupters acting similarly, but they recovered and run fine now)

edit: voltage is reading around 0.838V and this seems to be causing the chips to switch off or clock down towards default speeds (how would/does that work?) I think if it does not improve in the next hour i will take the system offline and try to tune down the pencil mod slightly. Ambient room temps are up 2-3 celcius in the last 24hrs now that heat in the building is on, and this may be the cause of lowered resistence=higher voltage

Many of my cards are at 20GH. It sounds like I could try to get them to 30 without adding heat sinks?

i would say that without adding heatsinks, but having proper airflow (a 120mm or 80mm fan pointed at the board is enough) you should be able to attain 30-35GH on a good board. If yours only runs 20GH by default, it sounds like some chips may be running slow so you may only see 27-30GH. Start at a voltage of 0.77v and see how it runs for 12hours. if it runs well, aim for 0.79v

[klondike_bar]

 Nice heatsink from ebay http://www.ebay.com/itm/100x140x12-7mm-Aluminum-Heatsink-for-Electronic-Computer-Electric-equipment-H157-/181110341808?ssPageName=ADME:L:OU:CA:3160

12.7mm height may be a little large for these boards, and may exceed the availale spacing between boards. 5-7mm would be better

[klondike_bar]

I just took my unit offline and checked the resistance of my pencil mod. immediate resistence was 1050k(ohm). after only 2 minutes it was at 1080k(ohm). I toned back the pencil mod and checked resistance at 1130k(ohm).

bootup voltage: 0.785V
3min voltage: 0.798V    - ~35GH
My ideal voltage is about 0.82-0.83V (~40.5GH)

over 0.835V leads to issues like those experienced today, where hashrate constantly struggles around half speed, with high error rates

[eestimees]

I dont know is it logical or rational, but could not get sleep and got an idea and made some scripts:

1) collect stats

# m h  dom mon dow   command
*/5 * * * * /home/pi/get_stat.sh

Code:

#!/bin/sh

cat /run/shm/.stat.log  | sed -E "s/[[:space:]]+/:/g" | cut -d ":" -f 1,3,4,5,6,7 | head -n 16 >> /home/pi/rw-stat.log

note I'm cutting 1st 16 lines there

then when it has run for some time, made this:

Code:

pi@bitfury ~ $ awk -F':' '{print $1,$4-($6*100/$5/100*$4),$2,$4,$5,$6; }' rw-stat.log | sort -r | uniq -f 3 | uniq -w 2 | sort -n | awk '{ if ( $2 ~ /[0-9]/ ) s+=$2;print $0;} END {printf "total: "s" Ghash/s\n"}'

result:

Code:

1 2.283 55 2.283 145 0
2 2.14944 55 2.177 158 2
3 2.135 55 2.135 169 0
4 2.177 55 2.177 167 0
5 1.97943 55 2.093 129 7
6 1.839 55 1.839 144 0
7 -nan 0 0.000 0 0
8 2.1187 55 2.135 131 1
9 2.315 55 2.315 172 0
10 2.23799 55 2.336 143 6
11 2.209 55 2.209 162 0
12 2.20839 56 2.283 153 5
13 2.072 55 2.072 156 0
14 2.19 55 2.220 148 2
15 2.26272 55 2.357 175 7
16 2.346 55 2.346 180 0
total: 32.5227 Ghash/s

explaning by columns:
1 - chip number
2 - Ghash/sec substracted % of error
3 - chip speed
4 - Ghash/sec from log
5 - Nonces/round
6 - false nonces

Idea was to let autotune to do the tuning for some days and then find the best setting where speed is highest while some errors are allowed. I was thinking that 2Ghash/s with 10% errors is 1.8Ghash/s and it is better that lower speed and 1.7Ghash/sec

maybe someone can get some ideas here or if my sleepy mind was useful (3.51AM at the moment) we can clean up that long and messy shell line.. and format output..

EDIT: added total has sum to the output end (this: 32.5227)

[klondike_bar]

^you should be getting far less than 10% errors if you updated to the newest chainminer (explained earlier in thread, a few pages back). Before i had 7-12% errors. now its about 2-4%

[eestimees]

^you should be getting far less than 10% errors if you updated to the newest chainminer (explained earlier in thread, a few pages back). Before i had 7-12% errors. now its about 2-4%

thanks, will try that

[Cablez]

^you should be getting far less than 10% errors if you updated to the newest chainminer (explained earlier in thread, a few pages back). Before i had 7-12% errors. now its about 2-4%

Can you do this with the V1 M-board as well?

[klondike_bar]

^you should be getting far less than 10% errors if you updated to the newest chainminer (explained earlier in thread, a few pages back). Before i had 7-12% errors. now its about 2-4%

Can you do this with the V1 M-board as well?

i dont beleive so, but dont take my word for it. I updated the chainminer on my v2 m-board from delayed august shipping and it drastically improved my error rates. a few chips produce 0 errors, most are in the range of 1-4

[Cablez]

Ok then if no one else has updated chainminer on a V1 then better to leave it alone.

[klondike_bar]

new heatsinks came in today



35x35x6mm with adhesive backings. I just swapped the 5 existing ones (15x15x8mm approx) for 4 of these. each one is just large enough to completely cover the backsides of 4 chips.

tweaked my pencil mod a bit to aim back at 40+GH. initial resistance was 1.13K(ohm) and initial voltage read was 0.805V

15 minute readings: voltage: 0.810    hashrate: 37.5-38GHash
my hope is that over the next 2-3 hours of operation, maximum temperature will be reached and the voltage will approach my ideal target of 0.825V. however, at this point it looks like it may not go past 0.815v eventually, so i may have to re-pencil it in a few hours once it is confirmed.

[tom99]

new heatsinks came in today



35x35x6mm with adhesive backings. I just swapped the 5 existing ones (15x15x8mm approx) for 4 of these. each one is just large enough to completely cover the backsides of 4 chips.

tweaked my pencil mod a bit to aim back at 40+GH. initial resistance was 1.13K(ohm) and initial voltage read was 0.805V

15 minute readings: voltage: 0.810    hashrate: 37.5-38GHash
my hope is that over the next 2-3 hours of operation, maximum temperature will be reached and the voltage will approach my ideal target of 0.825V. however, at this point it looks like it may not go past 0.815v eventually, so i may have to re-pencil it in a few hours once it is confirmed.

  you better off replace resistor work better.  You can buy resistor from this seller http://www.ebay.ca/itm/50x-3-3k-Ohm-1-16W-Chip-0603-SMT-SMD-Resistor-1-3k3-332-DE3803-/370879282748?pt=LH_DefaultDomain_0&hash=item565a20be3c

[Isokivi]

new heatsinks came in today



35x35x6mm with adhesive backings. I just swapped the 5 existing ones (15x15x8mm approx) for 4 of these. each one is just large enough to completely cover the backsides of 4 chips.

tweaked my pencil mod a bit to aim back at 40+GH. initial resistance was 1.13K(ohm) and initial voltage read was 0.805V

15 minute readings: voltage: 0.810    hashrate: 37.5-38GHash
my hope is that over the next 2-3 hours of operation, maximum temperature will be reached and the voltage will approach my ideal target of 0.825V. however, at this point it looks like it may not go past 0.815v eventually, so i may have to re-pencil it in a few hours once it is confirmed.

Im intreagued by those heat sinks, would you have a link handy to some specs, or a store selling them ?
...and please post a followup on how goot they work with, ambient temp and fan specs would be much appreciated also.

[klondike_bar]

new heatsinks came in today



35x35x6mm with adhesive backings. I just swapped the 5 existing ones (15x15x8mm approx) for 4 of these. each one is just large enough to completely cover the backsides of 4 chips.

tweaked my pencil mod a bit to aim back at 40+GH. initial resistance was 1.13K(ohm) and initial voltage read was 0.805V

15 minute readings: voltage: 0.810    hashrate: 37.5-38GHash
my hope is that over the next 2-3 hours of operation, maximum temperature will be reached and the voltage will approach my ideal target of 0.825V. however, at this point it looks like it may not go past 0.815v eventually, so i may have to re-pencil it in a few hours once it is confirmed.

Im intreagued by those heat sinks, would you have a link handy to some specs, or a store selling them ?
...and please post a followup on how goot they work with, ambient temp and fan specs would be much appreciated also.

i got them on ebay, around $1.50 a peice i think. they seem to work quite well and are definitely getting warm. they stuck on with ease (removing them may be a totally different story, the adhesive is on par with glue). I have a 120mm fan aimed directly at the back of the board from about 5" away, a 80mm pointed diagonally at the front face, and a 80mm on the other side of the front, so that its air channels between the RPi and h-board.

i think my fan use may be excessive for 1 board, but ive noticed 3 components need to be cooled:
1) the chips (duh) -  this is easily achieved by cooling the massive flat back of the PCB
2) the voltage regulator (?) (big cube on chip side of board) - i have some small heatsinks stuck to it that are perhaps a bit too small a footprint, but still work well. the 3rd fan i mentioned blows directly at this area, including component #3
3) the smallish chip/resistor (?) - on the chip side of the board beside component #2. it is very near the pencil-modded resistor and i think a major contributor to the heating and subsequent resistance drop as the unit heats up. I put a tiny heatsink on its backside as seen in the picture, where there are a lot of thermal vias. that heatsink is the warmest of all of them

I feel confident that later today i will try to push the pencil mod further. I would buy the resistor seen above but lack the solder tools and skill level to DIY it. (i can solder wires and big stuff, but surfacemount is a recipe for disaster)

[woodrake]

Heya,

I like being able to manage CipherMine's hardware via APIs. With our Avalons all we can do is read from them which is a bit frustrating (can't modify pool listings for instance, which is how our software switches between coins), but since the BitFuries are based on a Raspberry Pi and Debian (I think) I'm hopeful we can get the API working.

The presence of /home/pi/bfgminer suggests that they use bfgminer or some fork of it. /etc/rc.local calls /opt/bitfury/start-stratumproxy.sh and /opt/bitfury/start-miner.sh, the latter of which is presumably the thing that starts actual mining. start-miner.sh is pretty minimalist too:

Code:

#!/bin/sh

killall miner

cp /opt/bitfury/empty_stat.json /run/shm/stat.json

if [ -f /opt/bitfury/best.cnf ]; then
        cp /opt/bitfury/best.cnf /run/shm/.chip.cnf
fi
cd /run/shm

screen -d -m /opt/bitfury/chainminer/miner
exit 0

/opt/bitfury/best.cnf is not present on our units, nor /run/shm/.chip.cnf. That file is a potential candidate for some configuration, but if so I don't know what it should look like.

My first thought  /opt/bitfury/chainminer/miner (a binary file) is a custom compiled version of bfgminer, but having run it I've concluded that's not the case. I can achieve what I want just with some command-line options to bfgminer which is the frustrating thing. I tried changing the above to:

Code:

screen -d -m /opt/bitfury/chainminer/miner --api-listen --api-port 5001 --api-allow W:127.0.0.1

But that didn't appear to do anything:

Code:

root@bitfury:~# telnet 127.0.0.1 5001
Trying 127.0.0.1...
telnet: Unable to connect to remote host: Connection refused

Any suggestions?

Kate.


EDIT: Punin responded almost instantly - the man's dedicated to supporting his customers is awe-inspiring!

We are not in fact using BFGMiner by default although the latest version supports our product. Our own software is called chainminer and can be found in /opt/bitfury/chainminer

Our web based setup interface also works together with chainminer. If you want to run bfgminer, you must modify your /etc/rc.local or start-miner.sh to run bfgminer instead of chainminer. Please see github for latest version of BFGMiner that will work with your unit.

If you want to tweak things by hand please let the rig run for 5-10 mins then run:

cp /run/shm/.stat.log /opt/bitfury/best.cnf

and then edit the file.

I guess compiling bfgminer it is then!

[Anduck]

Chainminer is not BFGMiner. rPi and Debian + chainminer, yes.

You can tune your chips by copying /run/shm/.stat.log or /tmp/best.log to /opt/bitfury/best.cnf and then editing the file. There are instructions in the first post of this thread. A means autotune, a means autotune off etc...

Since chainminer isn't bfgminer, it doesn't have bfgminer-like api.

[Keefe]

i think my fan use may be excessive for 1 board, but ive noticed 3 components need to be cooled:
1) the chips (duh) -  this is easily achieved by cooling the massive flat back of the PCB
2) the voltage regulator (?) (big cube on chip side of board) - i have some small heatsinks stuck to it that are perhaps a bit too small a footprint, but still work well. the 3rd fan i mentioned blows directly at this area, including component #3
3) the smallish chip/resistor (?) - on the chip side of the board beside component #2. it is very near the pencil-modded resistor and i think a major contributor to the heating and subsequent resistance drop as the unit heats up. I put a tiny heatsink on its backside as seen in the picture, where there are a lot of thermal vias. that heatsink is the warmest of all of them

The big cube is the inductor (Pulse PA0513.441NLT). The small chip is the voltage regulator (TI TPS53355).

[zurg]

Mining should not need to access the SD card at all. I don't have the image, but is the swap disabled? There is no need of swap with revision 2 available memory.
Another reason could be that chainminer is storing data to often in it's logs on the file system. Not sure which logs are important and how big they get, but turning some of them off or storing all on a small ramdisk, may solve the dead SD problems.

If you can tell me what and how to check it I can. I am just a Linux n00b a bit.
Came to work.., rebooted Pi, it came up with no errors.
I do have a power switch that I can make Pi power cycle remotely by pinging it periodically.
Help@!, lol.

type free on the command prompt - if swap is used you will see it as the last line returned.
'dphys-swapfile swapoff' to turn it off and 'dphys-swapfile uninstall' to disable it

EDIT: to disable the service on startup use:
sudo update-rc.d -f dphys-swapfile remove

free reports:
102396 free and total
0 in use ( I tried randomly every few seconds etc.. but it never changes)

Now it's been nearly 3 days and my Pi took a dump again with the same errors.
My Power switch was still getting ping replies from Pi so I never rebooted. I was down for about nearly 4 hours.
Can anyone tell me how to get my Pi to auto reboot lets say every 24 hours?

Maybe I should redo my SD card again when I get to work tomorrow.

Thanks!

[klondike_bar]

new heatsinks came in today



35x35x6mm with adhesive backings. I just swapped the 5 existing ones (15x15x8mm approx) for 4 of these. each one is just large enough to completely cover the backsides of 4 chips.

tweaked my pencil mod a bit to aim back at 40+GH. initial resistance was 1.13K(ohm) and initial voltage read was 0.805V

15 minute readings: voltage: 0.810    hashrate: 37.5-38GHash
my hope is that over the next 2-3 hours of operation, maximum temperature will be reached and the voltage will approach my ideal target of 0.825V. however, at this point it looks like it may not go past 0.815v eventually, so i may have to re-pencil it in a few hours once it is confirmed.

24hr readings: voltage: 0.821-0.830 depending which capacitor is measured (closer to the volatge regulator = higher)   hashrate: 39-40GH
the ambient room temperature feels a few degrees warmer today, which may be causing this result. Either way, it should be noted that this indicates pencil modding results may increase over time and can result in bad performance if the final state is too high

based on these results, i am now trying to push chips 3-6 to 55 since they have lowest voltage compared to chips 9,10,15,16 (which run fine at 54 right now)
UPDATE: this method did not work. the chip error rates rose significntly, and are fine tuned back down to 54 now. I guess the voltage difference does not enable significant headroom for tuning. all 16 of my chips work best at 54 it seems, and any attempts to push for 55 typically takes it from a 1% error rate to 4-8% errors compared to marginal gain in hashspeed

[darkfriend77]

you can also ... regulate with clockspeed ... that's what i did ...

I had one board shutting down because graphite getting to low resistance on high temp ...


so I put all chips to clockspeed 53 ... now it's again stable ... and I can slowly approach again ...


It would be interesting to see which chip is where located ....

chips with big distance to the reg & resistor ... will probably have lesser influence ^^

Code:

+-+-+-+-+
|3|4|B|C|
+-+-+-+-+
|2|5|A|D|
+-+-+-+-+
|1|6|9|E|
+-+-+-+-+
|0|7|8|F|
+-+-+-+-+
  |||||
  +++++