Avalon ASIC users thread

[Sitarow]

Just a 'somewhat' on topic aside ...

I have a BitBurner XX with 20 Avalon chips in it.
It runs at 400Mhz with about 1.4% HW

Code:

 BTB  0: 48/ 48C 1223mV | 10.86G/7.669Gh/s | A:149590 R:748 HW:2151 WU:108.6/m

24hrs runtime

I put it up to 450MHz and it was about 50% but appropriate cooling would have probably resolved that also.



So indeed the chips can do 450MHz

Using my numbers above:
(7.669 / 20) * 240 = 92GH/s at 400MHz

Nice work Kano!

[weirdgod]

I notice that sometimes the total MHS5s hashrate drops to weird low number (like 4 digits only)... is this normal?
One time I even noticed that the average MHSav was also similary low...
I did a restart of cgminer service, and it went back to 80k+ ...

Could we somehow implement this https://github.com/shazbits/cgminer-monitor instead of the cgminer-monitor that is run in cron?
any other solution to monitor the MHSav/5s and act appropriately?

thanks,
J

[mdbssm]

I notice that sometimes the total MHS5s hashrate drops to weird low number (like 4 digits only)... is this normal?

No.

If you are using an ethernet connection, you must delete the WWAN device from the 'network' interfaces, and under network/wifi/avalon_ap make sure it shows 'wireless network is disabled'. Until I did both of these step I had the same problem.

[Ytterbium]

Is it possible to connect the lan port to a mobile Internet device so that it can hash in public places?

You don't even need to do that, you can use it's wifi capabilities.  All you need is power cord.  In fact, you can even use a cellphone. (you'll need to configure it for wifi using ethernet, though.)

[kefky]

Unload the wifi kernel modules.

Thanks!I'll try.
I hope this will finally solve the problem.

[punin]

One of my batch 3 avalons is performing very poorly due to a high HW errors.  Unplugging all by one module, each module individually has between a 25 and 75% HW error rate at either 256 or 300Mhz.

This degraded performance starting after about 6 hours of mining at 300Mhz.

Is there anything I should try before trying to make a warranty claim?

I was calculating the hardware error rate wrong.  It should be HW/LocalWork, which brings the error rates i'm seeing to a much more reasonable 1%.

Regardless, 3 out of 4 of my batch 3 units end up with <10Gh/s after 3-12hrs of mining, some more frequently than others.

Could this behavior be due to the new temperature throttling feature?  I have the default temperature limits of 70C target and 90C cutoff, but I don't see the temps going over 70C.

I'm having this same issue.

EDIT: so is the solution is to disable wifi or to implement the load monitor with auto-restart? Is this issue resolved?

[exahash]

One of my batch 3 avalons is performing very poorly due to a high HW errors.  Unplugging all by one module, each module individually has between a 25 and 75% HW error rate at either 256 or 300Mhz.

This degraded performance starting after about 6 hours of mining at 300Mhz.

Is there anything I should try before trying to make a warranty claim?

I was calculating the hardware error rate wrong.  It should be HW/LocalWork, which brings the error rates i'm seeing to a much more reasonable 1%.

Regardless, 3 out of 4 of my batch 3 units end up with <10Gh/s after 3-12hrs of mining, some more frequently than others.

Could this behavior be due to the new temperature throttling feature?  I have the default temperature limits of 70C target and 90C cutoff, but I don't see the temps going over 70C.

I'm having this same issue.

EDIT: so is the solution is to disable wifi or to implement the load monitor with auto-restart? Is this issue resolved?

For me, disabling/removing wifi/wlan wasn't enough.

If you already removed wifi and it's still a problem, look at your load avg, if it's spiking, then you probably need the auto-restart I posted at https://bitcointalk.org/index.phptopic=140539.msg2898478#msg2898478

If the load avg is low (under 1.0) then you should check all the connectors inside your avalon.  I did have one of the wide ribbon cables wiggle loose (I probably bumped it installing the psu) and the result was about 20%+ lower peak hashes and odd hangups.

[-ck]

I've uploaded new firmware 20130813 with the latest version of cgminer 3.3.3.

http://ck.kolivas.org/apps/cgminer/avalon/20130813/

This release contained a concerted effort to decrease CPU usage for stratum mining, and makes a massive difference when mining on p2pool.

A reminder that my firmware does not contain code to detect batch3 devices so you'll have to manually enter the higher target and overheat temperatures into your configuration.

[hephaist0s]

I have a 3-module batch 2 Avalon, and I have just acquired a batch 3 module to add to it. I've connected it and, after changing the miner count to 32, it seems to be running fine. I have not yet started overclocking it.
Questions:
1) Has anyone else added a batch 3 module to a batch 2 machine? If so, have you encountered any issues?
2) Should I change any settings other than the miner count in order to best recognize / utilize the new 4th module?
3) Will the stock batch 2 750W power supply be enough to overclock all 4 modules?
4) The screw holes don't match up and the IDE cord I had wasn't long enough, so right now I have it running outside the case. Does this matter? If I do get it jammed in the case, it looks like it will be closer to the adjacent module than the other modules are to each other; is this worth worrying about, heat-wise? I have an air conditioner blowing right on the Avalon.
Thanks in advance for any advice / ideas!

[fhh]

I gave it a test, but the actual firmware produces more HW errors on my B1 Avalon so its throtteling down to 330-340 MHz (starting @350).
So I'm back on 20130703 and again waving around 357 MHz

[-ck]

Yes there's a small regression in the auto code, I'm working on it now.

[fhh]

Yes there's a small regression in the auto code, I'm working on it now.

great:)

are you considering to include showing %HW? :
https://bitcointalk.org/index.php?topic=254331.0

its not a big problem to do it myself, but I would like it to be included

[-ck]

Yes there's a small regression in the auto code, I'm working on it now.

great:)

are you considering to include showing %HW? :
https://bitcointalk.org/index.php?topic=254331.0

its not a big problem to do it myself, but I would like it to be included

Not really, no.

Here's updated firmware:

http://ck.kolivas.org/apps/cgminer/avalon/20130813-1/

[SolarSilver]

Yes there's a small regression in the auto code, I'm working on it now.

great:)

are you considering to include showing %HW? :
https://bitcointalk.org/index.php?topic=254331.0

its not a big problem to do it myself, but I would like it to be included

Not really, no.

Here's updated firmware:

http://ck.kolivas.org/apps/cgminer/avalon/20130813-1/

Hey Con,

not sure what you meant to modify between 20130813 and 20130813-1 (small regression in the auto code?) but could you integrate this in a future release?

Code:

--- cgminer.lua-20130813-1      Tue Aug 13 11:03:30 2013
+++ cgminer.lua-benturas        Tue Aug 13 11:01:01 2013
@@ -39,6 +39,8 @@
    for line in summary do
       local elapsed, mhsav, foundblocks, getworks, accepted, rejected, hw, utility, discarded, stale, getfailures, localwork, remotefailures, networkblocks, totalmh, wu, diffaccepted, diffrejected, diffstale, bestshare = line:match("Elapsed=(%d+),MHS av=([%d%.]+),Found Blocks=(%d+),Getworks=(%d+),Accepted=(%d+),Rejected=(%d+),Hardware Errors=(%d+),Utility=([%d%.]+),Discarded=(%d+),Stale=(%d+),Get Failures=(%d+),Local Work=(%d+),Remote Failures=(%d+),Network Blocks=(%d+),Total MH=([%d%.]+),Work Utility=([%d%.]+),Difficulty Accepted=([%d]+)%.%d+,Difficulty Rejected=([%d]+)%.%d+,Difficulty Stale=([%d]+)%.%d+,Best Share=(%d+)")
       if elapsed then
+         local mhw = string.format("%d(%1.2f%%)",hw,(100*hw/(diffaccepted+diffrejected+hw)));
+         local mrj = string.format("%d(%1.2f%%)",rejected,(100*rejected/(accepted+rejected)));
         local str
         local days
         local h
@@ -72,8 +74,8 @@
            ['foundblocks'] = foundblocks,
            ['getworks'] = getworks,
            ['accepted'] = accepted,
-           ['rejected'] = rejected,
-           ['hw'] = hw,
+           ['rejected'] = mrj,
+           ['hw'] = mhw,
            ['utility'] = utility,
            ['discarded'] = discarded,
            ['stale'] = stale,

Does your reply "Not really, no." mean you are NOT putting it in ever, or that you have not thought about doing it yet and are considering it? Just confused ;-)

[-ck]

I don't have any intention of including the hw error % as a separate statistic. I also don't understand why you'd start deleting statistics from debug output like that, and I have to tell you your calculations are completely wrong.

[SolarSilver]

I don't have any intention of including the hw error % as a separate statistic. I also don't understand why you'd start deleting statistics from debug output like that, and I have to tell you your calculations are completely wrong.

You don't think adding the hw error% will cut down on all those questions from people what have difficulty interpreting the confusing numbers?

Not sure what you mean by deleting statistics from debug output and for the record, I am only referring to the thread by Ben Turas https://bitcointalk.org/index.php?topic=254331.0

My math is even worse than his ;-)

It's probably

Code:

(100*hw/(diffaccepted+hw))

?

[-ck]

I don't have any intention of including the hw error % as a separate statistic. I also don't understand why you'd start deleting statistics from debug output like that, and I have to tell you your calculations are completely wrong.

You don't think adding the hw error% will cut down on all those questions from people what have difficulty interpreting the confusing numbers?

Not sure what you mean by deleting statistics from debug output and for the record, I am only referring to the thread by Ben Turas https://bitcointalk.org/index.php?topic=254331.0

My math is even worse than his ;-)

It's probably

Code:

(100*hw/(diffaccepted+hw))

?

That is the correct equation.

...however, what are you doing with that figure?

[SolarSilver]

...however, what are you doing with that figure?

OK, at the risk of this being a trick question, I'm going to humiliate myself by saying "we display the value thus adding readability"?

[-ck]

...however, what are you doing with that figure?

OK, at the risk of this being a trick question, I'm going to humiliate myself by saying "we display the value thus adding readability"?

I'm not trying to humiliate you. I'm asking if you have a valid use for the figure. Whether it changes your mining in any way?

[thorvald]

...however, what are you doing with that figure?

OK, at the risk of this being a trick question, I'm going to humiliate myself by saying "we display the value thus adding readability"?

I'm not trying to humiliate you. I'm asking if you have a valid use for the figure. Whether it changes your mining in any way?

it will be easier for the users to change the clock and see a real figure (to keep the hw error under 2% and really see it  and not to do the calculation manually)
overclock the lazy way