[GUIDE] BitFury Miner Support/Tuning

[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|
+-+-+-+-+
  |||||
  +++++

[klondike_bar]

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|
+-+-+-+-+
  |||||
  +++++

im not sure. i clocked chips 3-6 from 54 to 55 and it resulted in errors. right now, all chips are tuned to 54 without an issue and my hashrate is around 39.5GH. I half expect this to rise a little bit, since the resistance seems to fluctuate with the ambient temperature. In the evening when my apartment is the warmest is when ive always seen it produce the highest hashrates but also push the limit where errors take over and start shutting down chips and shunting my hashrate.

42GH seems to be the upper bound before this occurs, at least with my prior heatsink configuration. Im happy to run a few days as it is now to follow the results and see how stability is. (in my mind, anything over 35 is sweet, so to be pushing the lower edge of 40GH is perfect)

[juhakall]

I received a new october card where the R02F resistor was 2.11k instead of 1.8k. Pencil modding it 2.05 boosted the valid hashrate from 32 GH/s to 34.5 GH/s, even though one chip dropped semi-dead after the modding.

[goxed]

Guys, looking for someone with a version 1 M board and successfully running 4 or more H cards

[Keefe]

Guys, looking for someone with a version 1 M board and successfully running 4 or more H cards

I have 4 cards, without problems most of the time. More are arriving later today.

[goxed]

Guys, looking for someone with a version 1 M board and successfully running 4 or more H cards

I have 4 cards, without problems most of the time. More are arriving later today.

wow, I have two stubborn version 1 M boards, that just wont hash properly with more than 3 H cards.

 What is your hashrate with 4 cards? Are you using the chainminer that came with the board? Any special settings? Thanks in advance.

[Keefe]

Guys, looking for someone with a version 1 M board and successfully running 4 or more H cards

I have 4 cards, without problems most of the time. More are arriving later today.

wow, I have two stubborn version 1 M boards, that just wont hash properly with more than 3 H cards.

 What is your hashrate with 4 cards? Are you using the chainminer that came with the board? Any special settings? Thanks in advance.

138.5 GH total, but 3 chips are dead (2 bypassed and 1 at the end of the chain). I overvolted them to an average of 0.78v. I'm using the stock software. Autotune is off, and some chips are set to 54 and some to 55, based on manual testing.

[goxed]

Guys, looking for someone with a version 1 M board and successfully running 4 or more H cards

I have 4 cards, without problems most of the time. More are arriving later today.

wow, I have two stubborn version 1 M boards, that just wont hash properly with more than 3 H cards.

 What is your hashrate with 4 cards? Are you using the chainminer that came with the board? Any special settings? Thanks in advance.

138.5 GH total, but 3 chips are dead (2 bypassed and 1 at the end of the chain). I overvolted them to an average of 0.78v. I'm using the stock software. Autotune is off, and some chips are set to 54 and some to 55, based on manual testing.

That's cool. I get 115GH/s with 3 H-boards. The moment I add the fourth it falls to 84GH/s. Using stock software, tried  both Autotune on and off. I guess I will have to dig deeper. Hashrate starts off good after power on and falls down within a minute or so.   Thanks for your help though