[GUIDE] BitFury Miner Support/Tuning

[klondike_bar]

A better test would be to power the card and measure voltage at the top of the Pulse inductor. If you get ~0.75v there (or whatever it's adjusted to), the power circuit (including that capacitor) is probably fine. I only say "probably" because I just fixed a board that was missing a cap on the back side. Before replacing that tiny back-side cap, the voltage at the Pulse top was correct, but the card wouldn't hash stably.

okay, i measured, it was 0.3V or 0V, what i can do next?

okay, lets start at the top:
1) you are forcing tuning to run at 55 when you use "aIfDSo 55". the fact you are seeing more than 10% errors means you should not have them higher than 53 or 54. change them to auto-tuning. (AIfDSo)
2) a few chips (such as 52) are set as "aifdso". this wont work in the bitfury system unless you were forcing custom timing clocks and other info to the chip (you aren't). change them to AIfDSo.
3) where are 1-30?
4) Your chips are producing way to many errors. half of their hash rate is actually errors. You need to cool your chips better. The chips currently at zero likely switched off because the autotuning was overwhelmed by the high error rate. you need to have good cooling fans, and likely add some heatsinks.
5) if you havent yet, update your chainminer following the method mentioned several times in this thread.

do the above and then check again. it should resolve 90% of your hashrate issues. as for the 0.3V you are measuring, try again after the changes. if its still <0.6V, check the values of the resistors R01F and R02F. in a board with capacitors these should be 1.0 and 1.25 k(ohm) respectively. In a board without capacitors they should be -?- and 1.7k(ohm) respectively (i dont know the R01F value for the board, but its likely not 1.0k like the first version)

if your resistors are not these values, thats the issue. if they are correct, then its onto the next step. probably the small TI chip found right beside them (it has about 12 or 14 pins) that is responsible for bringing 12v->0.7v

Wait...   are you saying that for chips that don't work/generate too many errors, setting them as aifdso doesn't work?? if not, what additional information needs to be provided to actually turn them off?? I can see a bunch of zero's in that line, which that seems to indicate they are technically switched off, so I don't understand why you're saying that aifdso alone doesn't fix the problem.... Thx for clarifying...

i meant for all the chips that are working, you have seriously high error rates with your manual tuning. autotuning will work a lot better.

[1714204035]

1714204035

[1714204035]

1714204035

[1714204035]

1714204035

[1714204035]

1714204035

[klondike_bar]

Okay, i made changes to settings, problem not solved...

And this is what i get...:

speed:2592 noncerate[GH/s]:96.651 (2.014/chip) hashrate[GH/s]:100.665 good:6751 errors:1585 spi-errors:1 mis$
4:      864     30.695  34.045  2144    339     0       1
8:      864     31.897  32.481  2228    849     0       0
C:      864     34.059  34.140  2379    397     1       0

Voltage still 0.3V

I measured resistors,
R01F:  2,120K ohm
R02F:  2,120K ohm

..so Ti chipis broken...?
 

1) switch to autotuning. clearly setting manual tuning to 54 across every chips is working terribly based on your 15-30% error rates.
2) R02F is quite high at 2.12K(ohm). try using the pencil mod to reduce it to 1.85K(ohm) and see what happens. If lucky, this is the issue and youll see the voltage rise to >0.6V. If it does, maybe the wrong resistor was placed and you should see if it can be changed for something in the 1.65-1.8K range

1. As you can see, m-board don't even detect BANK1 h-card(s). It is autotune mode now, but no help..
2. If I do pencil mod that, that will broke h-card warrantly...?

pencil modding is techincally messing with the warranty, but a tiny marking of pencil lead is not very hard to clean off and, unless you seriously manage to do it wrong, shouldnt cause anything worse then putting a bit too much voltage that in turn turns off the chips until you restart the miner (after dabbing away a fraction of the graphite with a fingertip)

try pulling out and re-inserting the board. i had a board dissapear the other day (after almost a week in place) and pulled it out, checked the connector and reinserted it. works fine again. I think if there is vibration from fans or weight from heatsinks it may be possible it simply isnt seated perfectly with all the connectors making contact.

[klondike_bar]

If anyone want to take a tracking of his Bitfury based on performance data of each card, here you find a macro for Excel which track all JSON-data each 5min. It delivers you some results as table and chart. You can analyze as you want, e.g. by card, bus, slot, SPIerr or MISOerr.
Just open in excel, enter IPadress and click start.

Ciao

Enrico

im getting debug errors through this whole chunk of code:

Workbooks(sFile).Worksheets("Config").Unprotect "bitcoin"
    Workbooks(sFile).Worksheets("Config").Cells(2, 2).Value = "logging stopped"
    Workbooks(sFile).Worksheets("Config").Cells(2, 2).Font.Bold = False
    Workbooks(sFile).Worksheets("Config").Cells(2, 2).Font.Color = RGB(0, 0, 0)
    Workbooks(sFile).Worksheets("Config").Protect "bitcoin"

and the similar chunk below for "logging started" ... "Bold = True" ... etc

am i missing something, or is there an underlying macro or feature i need to instigate?

[ricolein]

If anyone want to take a tracking of his Bitfury based on performance data of each card, here you find a macro for Excel which track all JSON-data each 5min. It delivers you some results as table and chart. You can analyze as you want, e.g. by card, bus, slot, SPIerr or MISOerr.
Just open in excel, enter IPadress and click start.

im getting debug errors through this whole chunk of code:

Workbooks(sFile).Worksheets("Config").Unprotect "bitcoin"
    Workbooks(sFile).Worksheets("Config").Cells(2, 2).Value = "logging stopped"
    Workbooks(sFile).Worksheets("Config").Cells(2, 2).Font.Bold = False
    Workbooks(sFile).Worksheets("Config").Cells(2, 2).Font.Color = RGB(0, 0, 0)
    Workbooks(sFile).Worksheets("Config").Protect "bitcoin"

and the similar chunk below for "logging started" ... "Bold = True" ... etc

am i missing something, or is there an underlying macro or feature i need to instigate?

Not your fault. Maybe you opened a different excel-file and work in it while the macro runs in background? Thats the only case i found for this problem.

I fixed it. Now you can work in different excel-files while data logging runs at background. But this is not a recommended using. If you get again an (other) error, please try at first this macro run alone without other files in background.
To get the new version please download here at http://www.bansemer.net/dl/BF-Datalogger.xlsm

Ciao

Enrico

[Jumpsu]

Okay, i made changes to settings, problem not solved...

And this is what i get...:

speed:2592 noncerate[GH/s]:96.651 (2.014/chip) hashrate[GH/s]:100.665 good:6751 errors:1585 spi-errors:1 mis$
4:      864     30.695  34.045  2144    339     0       1
8:      864     31.897  32.481  2228    849     0       0
C:      864     34.059  34.140  2379    397     1       0

Voltage still 0.3V

I measured resistors,
R01F:  2,120K ohm
R02F:  2,120K ohm

..so Ti chipis broken...?
 

1) switch to autotuning. clearly setting manual tuning to 54 across every chips is working terribly based on your 15-30% error rates.
2) R02F is quite high at 2.12K(ohm). try using the pencil mod to reduce it to 1.85K(ohm) and see what happens. If lucky, this is the issue and youll see the voltage rise to >0.6V. If it does, maybe the wrong resistor was placed and you should see if it can be changed for something in the 1.65-1.8K range

1. As you can see, m-board don't even detect BANK1 h-card(s). It is autotune mode now, but no help..
2. If I do pencil mod that, that will broke h-card warrantly...?

pencil modding is techincally messing with the warranty, but a tiny marking of pencil lead is not very hard to clean off and, unless you seriously manage to do it wrong, shouldnt cause anything worse then putting a bit too much voltage that in turn turns off the chips until you restart the miner (after dabbing away a fraction of the graphite with a fingertip)

try pulling out and re-inserting the board. i had a board dissapear the other day (after almost a week in place) and pulled it out, checked the connector and reinserted it. works fine again. I think if there is vibration from fans or weight from heatsinks it may be possible it simply isnt seated perfectly with all the connectors making contact.

I tried, no help... pheww... 0.0V

[klondike_bar]

I tried, no help... pheww... 0.0V

you went from 0.3v to 0.0v?  That seems wrong. try a different m-board slot and ensure proper insertion.

 If all else fails, ill buy your board for $200 shipped to toronto

[Jumpsu]

I tried, no help... pheww... 0.0V

you went from 0.3v to 0.0v?  That seems wrong. try a different m-board slot and ensure proper insertion.

 If all else fails, ill buy your board for $200 shipped to toronto

Lol, i am electronics technician, so i repair/change Ti-chip,I do not want to break the warranty...

[Swimmer63]

I just received two new h-boards, V1.2 as I have a V1 M-board.
Both cards are running at 33-34 Gh/s out of the box.
I measured the R02F resistor and both boards read 2.1KOhm.
My other boards read 1.3KOhm.
Can I still attempt to mod the two new boards?  If so, I don't want to move them much because they are already running high.  But do have very good cooling, 4 fans on 6 cards.
Thoughts?  Input from the wise men?

[Meizirkki]

Lol, i am electronics technician, so i repair/change Ti-chip,I do not want to break the warranty...

you will lose the warranty. But so long as you know what you're doing you wont need it. *high five* I'm an electronics technician too.

Can I still attempt to mod the two new boards?  If so, I don't want to move them much because they are already running high.  But do have very good cooling, 4 fans on 6 cards.
Thoughts?  Input from the wise men?

Iirc some where reporting problems when mixing old pencil modded cards with modded new ones. I'd say do it you can always clean the grphite off if problems arise.

[Isokivi]

I just received two new h-boards, V1.2 as I have a V1 M-board.
Both cards are running at 33-34 Gh/s out of the box.
I measured the R02F resistor and both boards read 2.1KOhm.
My other boards read 1.3KOhm.
Can I still attempt to mod the two new boards?  If so, I don't want to move them much because they are already running high.  But do have very good cooling, 4 fans on 6 cards.
Thoughts?  Input from the wise men?

Yes, you can pencil mod them and they respond as well as the previous boards, do not go as low tough. I've found that any board can take 1.72 and ones with very good cooling (3 fans per bank and heatsinks) can go as low as 1.63KOhm. I'd try to lower the resistance by 0.05KOhm at a step and see how the cards respond.

[Swimmer63]

I just received two new h-boards, V1.2 as I have a V1 M-board.
Both cards are running at 33-34 Gh/s out of the box.
I measured the R02F resistor and both boards read 2.1KOhm.
My other boards read 1.3KOhm.
Can I still attempt to mod the two new boards?  If so, I don't want to move them much because they are already running high.  But do have very good cooling, 4 fans on 6 cards.
Thoughts?  Input from the wise men?

Yes, you can pencil mod them and they respond as well as the previous boards, do not go as low tough. I've found that any board can take 1.72 and ones with very good cooling (3 fans per bank and heatsinks) can go as low as 1.63KOhm. I'd try to lower the resistance by 0.05KOhm at a step and see how the cards respond.

Makes sense.  Thanks guys.

[GandalfG]

For all those who measure the voltage applied to chip on H-Boards, as a minus (GND) use for this GND on the H-Board instead power suply or M-boards. These differences arising from the resistance of sockets and contacts on a card. They range tenths of volts. It turns out that, instead of 0.865V is 0.825 or even 0.760
Therefore, the chip receives voltage measured preferably on next chip capacitors or capacitors next to coils (PULSE) of DC-DC converter labeled C12 to C16.

https://www.dropbox.com/s/rxhufkcy1tgtwx2/20131110_220257.jpg

[Keefe]

For all those who measure the voltage applied to chip on H-Boards, as a minus (GND) use for this GND on the H-Board instead power suply or M-boards. These differences arising from the resistance of sockets and contacts on a card. They range tenths of volts. It turns out that, instead of 0.865V is 0.825 or even 0.760
Therefore, the chip receives voltage measured preferably on next chip capacitors or capacitors next to coils (PULSE) of DC-DC converter labeled C12 to C16.

https://www.dropbox.com/s/rxhufkcy1tgtwx2/20131110_220257.jpg

Check voltage on top of the inductor (grey box with PULSE printed on it and any of the GND pin from the PCIE power supply)

Just FYI... There is a small problem with measuring voltage like that, but it's still the easiest way in a crowded rig. There's a significant and varying voltage drop on the ground line between the power supply and a card. Just between the M-board's GND terminal and a slot's ground pins, there's a drop of between 0.023v (slot 0) and 0.067v (slot F) on my v1 rig. So while it may look like you're running a card at 0.900v when measured the easy way, it could actually be as low as 0.833v. And I didn't even account for any further drop from the slot to the chips' ground plane.

[inform]

Yeah nice

[greaterninja]

I have a version 2.3 M board.  I updated my Chainminer and now it only detects 2/5 cards.

there is no best.cnf file

and instead of 5x16 = 80 chips, my bitfury shows 88 chips when I know there are only 80 chips.


My question is what do I do now?

[Keefe]

I have a version 2.3 M board.  I updated my Chainminer and now it only detects 2/5 cards.

there is no best.cnf file

and instead of 5x16 = 80 chips, my bitfury shows 88 chips when I know there are only 80 chips.


My question is what do I do now?

Reboot it until the problem goes away. Try reseating boards also.

[kano]

Guess I should mention it in here.

I've written a cgminer driver (it's in the current cgminer 3.8.0/3.8.1) for the BlackArrow Bitfury V1 boards.
I've no idea if that is the same as a standard Bitfury board - I never had one and the twits would never sell me one either.
... maybe I should have put in the copyright header that Punin and Buzzdave aren't allowed to use the driver

Anyway, as I posted about it the other day:

For anyone who has a V1 BlackArrow Bitfury board - the latest cgminer 3.8.1 (3.8.0 also) has the RPi source for a driver now - consider it an early release since there are certain improvements I'll make to it that I've already sorted out what to do - in future versions - but it already runs with great performance.

The configure option is --enable-bab

It runs at 39.9GH/s valid shares with a bit under 10% HW errors above the 39.9GH/s on a standard board
i.e. if you were to also count bad nonces returned (which I prefer not to but I will mention in case anyone wishes to compare that number) it's returning 44GH/s of good+bad nonces.

I've run it with a single board on Raspbian to write and test it - it's based on the V1 initialisation of a single board.
I based the I/O code on the chainminer GPIO/SPI code as should be obvious for anyone who looks at my driver-bab.c and chainminer (as I also mention that at the top of driver-bab.c)

It's a bit CPU hungry so soon I'll be spending more effort on reducing that - 20% CPU with a single board - I've only tested with a single board.

It also doesn't tune the chip at all - so there is also assured performance improvement there - it runs at default settings at the moment.
That's the next effort I'll expend on the driver.

If anyone has a standard V1 BitFury board I'd also be curious to know if it works with them.

[AMD FTW]

Whoever gets me up and running 1 Btc here.

https://bitcointalk.org/index.php?topic=332035.msg3561229#msg3561229

[spegelius]

Guess I should mention it in here.

I've written a cgminer driver (it's in the current cgminer 3.8.0/3.8.1) for the BlackArrow Bitfury V1 boards.
I've no idea if that is the same as a standard Bitfury board - I never had one and the twits would never sell me one either.
... maybe I should have put in the copyright header that Punin and Buzzdave aren't allowed to use the driver

Anyway, as I posted about it the other day:

For anyone who has a V1 BlackArrow Bitfury board - the latest cgminer 3.8.1 (3.8.0 also) has the RPi source for a driver now - consider it an early release since there are certain improvements I'll make to it that I've already sorted out what to do - in future versions - but it already runs with great performance.

The configure option is --enable-bab

It runs at 39.9GH/s valid shares with a bit under 10% HW errors above the 39.9GH/s on a standard board
i.e. if you were to also count bad nonces returned (which I prefer not to but I will mention in case anyone wishes to compare that number) it's returning 44GH/s of good+bad nonces.

I've run it with a single board on Raspbian to write and test it - it's based on the V1 initialisation of a single board.
I based the I/O code on the chainminer GPIO/SPI code as should be obvious for anyone who looks at my driver-bab.c and chainminer (as I also mention that at the top of driver-bab.c)

It's a bit CPU hungry so soon I'll be spending more effort on reducing that - 20% CPU with a single board - I've only tested with a single board.

It also doesn't tune the chip at all - so there is also assured performance improvement there - it runs at default settings at the moment.
That's the next effort I'll expend on the driver.

If anyone has a standard V1 BitFury board I'd also be curious to know if it works with them.

Tried it, built with --enable-bab, but it didn't find any chips.

Edit: i have 5 v.12 boards, but i suppose that the version shouldn't matter. Had to run with sudo as cgminer wanted access to /dev/mem.

I'll be happy to try more builds/do debug runs, if there will be any development regarding these boards (although it seems that the motivation might be lacking ). Would love to replace the chainminer/stratumproxy combo with more sophisticated system.

[goxed]

Guys I have a question about the H-board.

Where can I download the H-board layout file? I am trying to get one manufactured.

Thanks