Problematic BFL Single

[JWU42]

Will try sending an e-mail too but they are still taking 7+ days to respond  

Last few days I have notice one single (running 808 FW) has been reporting much lower U (shares per minute) then the two other singles on the same firmware -- NOTE BFL2 - below.

Code:

cgminer version 2.5.0 - Started: [2012-07-08 09:47:51]
--------------------------------------------------------------------------------
 (5s):6820.5 (avg):7213.5 Mh/s | Q:64532  A:60760  R:130  HW:0  E:94%  U:98.9/m
 TQ: 12  ST: 12  SS: 0  DW: 839  NB: 68  LW: 0  GF: 6  RF: 17
 Connected to http://NOYB:8332 with LP as user XXXX
 Block: 000008c59a807b37fe8bfba3dc140a52...  Started: [20:00:12]
--------------------------------------------------------------------------------
 [P]ool management [G]PU management [S]ettings [D]isplay options [Q]uit
 GPU  0:  69.0C 2886RPM | 585.1/590.6Mh/s | A:5070 R:18 HW:0 U:   8.25/m I:11
 GPU  1:  68.0C 2892RPM | 631.5/626.5Mh/s | A:5481 R:14 HW:0 U:   8.92/m I:11
 GPU  2:  68.0C 2893RPM | 582.2/590.3Mh/s | A:4997 R:14 HW:0 U:   8.13/m I:11
 GPU  3:  69.0C 2690RPM | 626.8/625.9Mh/s | A:5406 R:11 HW:0 U:   8.80/m I:11
 BFL  0:  50.2C         | 804.2/795.8Mh/s | A:6714 R:12 HW:0 U:  10.93/m
 BFL  1:  52.0C         | 804.2/795.2Mh/s | A:6865 R:17 HW:0 U:  11.17/m
 BFL  2:  56.9C         | 804.1/795.6Mh/s | A:6023 R:10 HW:0 U:   9.80/m
 BFL  3:  51.6C         | 796.7/787.6Mh/s | A:6759 R:13 HW:0 U:  11.00/m
 BFL  4:  51.8C         | 764.1/755.7Mh/s | A:6223 R:11 HW:0 U:  10.13/m
 BFL  5:  52.5C         | 857.4/850.2Mh/s | A:7222 R:10 HW:0 U:  11.75/m
--------------------------------------------------------------------------------

This was after almost 12 hours - so long enough to rule out short-term "luck".

I removed the single from this linux rig and ran the auto-tune on it on a windows rig.  832 was the selected FW (higher than what I had been running) but it might be a tad cooler in my office.  When I started mining (tried BFGminer as well to ensure it wasn't a CGminer issue) the "nonce light" that is visible on the right side of the unit would occasionally turn off (not just blip at each new nonce).  This seems to be completely random and will last from .5 to 12 seconds.  This appears to be the equivalent of the device turning off - hence the lower U.  The weird part is why the Hash Rate still shows fairly high.

I have since flashed to the lowest firmware (768) and it is still happening (nonce light turning off).

RMA is probably not worth it as I will lose more income in not miningg than I would gain from a properly functioning unit?!?

PS - While typing his message the light has turned off 5+ times...  

PPS - BFL crew - any suggestions?!?

[1714857099]

1714857099

[1714857099]

1714857099

[1714857099]

1714857099

[1714857099]

1714857099

[1714857099]

1714857099

[rjk]

Does it do the same when a diagnostic is being run? The diagnostic uses known data instead of random data.

Wonder if your pool isn't providing work fast enough and cgminer isn't mentioning it (although it usually reports it religiously)

EDIT: I'm assuming that you have ruled out throttling as the cause; when that happens the same side light goes dim, but the front panel light also flashes for a while.

[JWU42]

RJK,

Yeah - I was thinking about this...

1.  No issues when using Easyminer
2.  Tried with 3 different pools - same issue repeats itself
3.  768 firmware seems to be best for now...

I think it might be temp related?!?  This is the hottest running single I have and using the lowest FW is keeping temps in mid 50's (where they seem to be happier).

And yes, to confirm, the unit is NOT throttling (front light blink).

[BFL-Engineer]

Will try sending an e-mail too but they are still taking 7+ days to respond 

Last few days I have notice one single (running 808 FW) has been reporting much lower U (shares per minute) then the two other singles on the same firmware -- NOTE BFL2 - below.

Code:

cgminer version 2.5.0 - Started: [2012-07-08 09:47:51]
--------------------------------------------------------------------------------
 (5s):6820.5 (avg):7213.5 Mh/s | Q:64532  A:60760  R:130  HW:0  E:94%  U:98.9/m
 TQ: 12  ST: 12  SS: 0  DW: 839  NB: 68  LW: 0  GF: 6  RF: 17
 Connected to http://NOYB:8332 with LP as user XXXX
 Block: 000008c59a807b37fe8bfba3dc140a52...  Started: [20:00:12]
--------------------------------------------------------------------------------
 [P]ool management [G]PU management [S]ettings [D]isplay options [Q]uit
 GPU  0:  69.0C 2886RPM | 585.1/590.6Mh/s | A:5070 R:18 HW:0 U:   8.25/m I:11
 GPU  1:  68.0C 2892RPM | 631.5/626.5Mh/s | A:5481 R:14 HW:0 U:   8.92/m I:11
 GPU  2:  68.0C 2893RPM | 582.2/590.3Mh/s | A:4997 R:14 HW:0 U:   8.13/m I:11
 GPU  3:  69.0C 2690RPM | 626.8/625.9Mh/s | A:5406 R:11 HW:0 U:   8.80/m I:11
 BFL  0:  50.2C         | 804.2/795.8Mh/s | A:6714 R:12 HW:0 U:  10.93/m
 BFL  1:  52.0C         | 804.2/795.2Mh/s | A:6865 R:17 HW:0 U:  11.17/m
 BFL  2:  56.9C         | 804.1/795.6Mh/s | A:6023 R:10 HW:0 U:   9.80/m
 BFL  3:  51.6C         | 796.7/787.6Mh/s | A:6759 R:13 HW:0 U:  11.00/m
 BFL  4:  51.8C         | 764.1/755.7Mh/s | A:6223 R:11 HW:0 U:  10.13/m
 BFL  5:  52.5C         | 857.4/850.2Mh/s | A:7222 R:10 HW:0 U:  11.75/m
--------------------------------------------------------------------------------

This was after almost 12 hours - so long enough to rule out short-term "luck".

I removed the single from this linux rig and ran the auto-tune on it on a windows rig.  832 was the selected FW (higher than what I had been running) but it might be a tad cooler in my office.  When I started mining (tried BFGminer as well to ensure it wasn't a CGminer issue) the "nonce light" that is visible on the right side of the unit would occasionally turn off (not just blip at each new nonce).  This seems to be completely random and will last from .5 to 12 seconds.  This appears to be the equivalent of the device turning off - hence the lower U.  The weird part is why the Hash Rate still shows fairly high.

I have since flashed to the lowest firmware (768) and it is still happening (nonce light turning off).

RMA is probably not worth it as I will lose more income in not miningg than I would gain from a properly functioning unit?!?

PS - While typing his message the light has turned off 5+ times... 

PPS - BFL crew - any suggestions?!?

Dear JWU42,

Three things might be happening:

1) If occasionnaly the nonce-led (the ones that are inside the box, not the one LED that is exposed via a lense) starts blinking,
   remains blinking for 15 seconds and then remains ON and the nonce LEDs start working again, then the unit is throttling.
   This is unlikely however, since you've mentioned that you had the same situation with a 768 MH/s firmware; Unless the unit was
   moved or is receiving hot-air from exhaust of other units/PSUs or for some other reason, the air around the intake is warm.
   
2) The pool is not providing job fast enough. If the front LED remains on, but the nonce leds inside the box remain off for like a
   second or two from time to time, this means the unit remains idle some times (not receiving jobs fast enough).
   This is usually a Pool/Network/Software issue. If you're running on Linux, then linux kernel or driver can have a role in it as well.

3) USB Hub issues. Please check the USB connection, and try connecting the unit directly to the computer to see if it changes.


Please let us know how it goes.


Regards,
BF Labs Inc.

[JWU42]

@BFL-Engineer - Thanks for the response!  To your points...

1.  The "nonce light" is not blinking - it will completely turn off for, as I said, up to 12 seconds.  Not blink - turn completely dark.  Even though Easy-Miner selected the 832 FW with auto-tune, I have tried using that, 808 and now 768.  With every FW the same issue.  Again, speaking of the "nonce light" and NOT the front facing "throttle light".

2.  Maybe there were some pool issues as it didn't happen with Easy-Miner.  I tried 3 different pools - still getting the issue.  Also, if a pool issue, why would it be limited to one unit?  I should see this issue with other units if it was a problem with the pool?!?

3.  Unit was originally on a linux rig (with USB hub) and moved to windows rig (different USB cable and direct to MB) - same issues.  Also tried BFGminer and CGminer - same issues.  I am trying to be methodical to rule out all items...

Unit is running "well" with 768 FW.  U of 9.66 - which is within the expected range based on the 768 FW.

I will run Easy-Miner again later, use that selected FW, likely see the light off issues again and will video it...

[BFL-Engineer]

@BFL-Engineer - Thanks for the response!  To your points...

1.  The "nonce light" is not blinking - it will completely turn off for, as I said, up to 12 seconds.  Not blink - turn completely dark.  Even though Easy-Miner selected the 832 FW with auto-tune, I have tried using that, 808 and now 768.  With every FW the same issue.  Again, speaking of the "nonce light" and NOT the front facing "throttle light".

2.  Maybe there were some pool issues as it didn't happen with Easy-Miner.  I tried 3 different pools - still getting the issue.  Also, if a pool issue, why would it be limited to one unit?  I should see this issue with other units if it was a problem with the pool?!?

3.  Unit was originally on a linux rig (with USB hub) and moved to windows rig (different USB cable and direct to MB) - same issues.  Also tried BFGminer and CGminer - same issues.  I am trying to be methodical to rule out all items...

Unit is running "well" with 768 FW.  U of 9.66 - which is within the expected range based on the 768 FW.

I will run Easy-Miner again later, use that selected FW, likely see the light off issues again and will video it...

Does the front LED ever turn off  or blink during you tests?
If so, can you describe when it happens and how?

Regards,
BF Labs Inc.

[P_Shep]

@BFL-Engineer - Thanks for the response!  To your points...

1.  The "nonce light" is not blinking - it will completely turn off for, as I said, up to 12 seconds.  Not blink - turn completely dark.  Even though Easy-Miner selected the 832 FW with auto-tune, I have tried using that, 808 and now 768.  With every FW the same issue.  Again, speaking of the "nonce light" and NOT the front facing "throttle light".

2.  Maybe there were some pool issues as it didn't happen with Easy-Miner.  I tried 3 different pools - still getting the issue.  Also, if a pool issue, why would it be limited to one unit?  I should see this issue with other units if it was a problem with the pool?!?

3.  Unit was originally on a linux rig (with USB hub) and moved to windows rig (different USB cable and direct to MB) - same issues.  Also tried BFGminer and CGminer - same issues.  I am trying to be methodical to rule out all items...

Unit is running "well" with 768 FW.  U of 9.66 - which is within the expected range based on the 768 FW.

I will run Easy-Miner again later, use that selected FW, likely see the light off issues again and will video it...

I'm seeing similar issues with one of mine. Haven't had chance to investigate fully yet though.
About 1/2 the time it'll take over 7s... 5s with the nonce light on (as usual) then it'll sit for the remaining 2s with the light off. No blinking on any of the lights. I'm wondering if something happened to the heatsink. Though it's been fine for the last 2 months.

JWU have your tried going back to 2.4.4 yet? I haven't.

[JWU42]

@BFL-Engineer,

Front light "Throttle Light" stays solid and does NOT blink at all when I am experiencing the "nonce light" turning off.


@P_Shep,

Happens with 2.4.4 and 2.5.0...

I am starting to see it correlate with "Pool X not proving work fast enough".  Not 100% sure but maybe there is something to what BFL-Engineer said...

[P_Shep]

K, I've been staring at the LEDs for a while.

I see there are 2 nonce LEDs (one for each chip, presumably) I see that (looking at my other units) they should be synchronised... Well they're not on this one.
Sometimes they work normally and come on at the same time, go out at the same time, off for a very short time than back on.
Sometimes only the outer one comes on, then the LED goes out for a short time then back on.
Sometimes the outer one comes on, then after a few seconds the inner one comes on for a second or so, then both go out for 4-5 seconds.

And as with JWU, the front light remains solid lit at all times

[P_Shep]

And on that note it's just started working at full speed again...

(but for how long?)

[BFL-Engineer]

@BFL-Engineer,

Front light "Throttle Light" stays solid and does NOT blink at all when I am experiencing the "nonce light" turning off.


@P_Shep,

Happens with 2.4.4 and 2.5.0...

I am starting to see it correlate with "Pool X not proving work fast enough".  Not 100% sure but maybe there is something to what BFL-Engineer said...

Can you run a heavy diagnostics on EasyMiner, and see if the same thing happens or not?


Regards,
BF Labs Inc.

[JWU42]

I think most of this is related to pool issues (GPUmax when passing through to ABCPool).  I still cannot understand why it was not effecting other Singles though 

I changed to the 800 FW and things have been running well (back on main linux rig) for the last 4 hours.

If it persists I will run the Heavy Diagnostics and advise.

Thanks!

[P_Shep]

Turns out taking the case off fixed it, as screwing it back on broke it. So it off again.
Guess there's a track broken somewhere.

[nbtcminer]

Turns out taking the case off fixed it, as screwing it back on broke it. So it off again.
Guess there's a track broken somewhere.

This sounds like the unit is over-heating causing it to throttle. Have you considered re-applying thermal paste to the heatsink / chip? Might want to check with BFL to see if this voids any warranties first tho.

[P_Shep]

No, it's not heat related at all. I don't need to take the case completely off for it to work again, just unscrew it. The board must be being twisted when it's screwed shut.

[nbtcminer]

No, it's not heat related at all. I don't need to take the case completely off for it to work again, just unscrew it. The board must be being twisted when it's screwed shut.

You mean closing the case actually warps the board??? How long are the screws?

[P_Shep]

No, it's not heat related at all. I don't need to take the case completely off for it to work again, just unscrew it. The board must be being twisted when it's screwed shut.

You mean closing the case actually warps the board??? How long are the screws?

It would seem to be the case. Slotting the cover on must just warp the board enough to break something. Screw length has nothing to do with it.