OFFICIAL CGMINER mining software thread for linux/win/osx/mips/arm/r-pi 4.11.1

[-ck]

New version - 2.7.1, 21st August 2012

Bugfixes galore \o/ The emphasis was just on outstanding bugfixes on this release to try and get a nice stable release out after the frantic pace of development recently.

Human readable changelog:
The occasional strange behaviour where lots of work would end up on backup pools in failover mode should be fixed.
The occasional scenario where one pool dies and the others behave like they're slow to provide work should be fixed.
Very high hashrate (U > 100) machines should now be able to work unhindered even on pools that don't support rolltime.
The -r and -R options no longer do anything and are deprecated in preference for faster communications and simpler code.
Hopefully the dynamic mode for GPUs is fixed on windows.
cgminer dying on switching users on windows should be fixed (it may actually restart itself to recover unless --no-restart is specified).
Hopefully the 7 day windows crash bug has been fixed
Share submissions that are re-sent will have (resend) appended to the log
Hopefully the gpu-map feature is working for more adl devices than opencl.
BFL autodetect fixes on windows.
New API features
New miner.php features.


Full changelog:
- Update windows build instructions courtesy of sharky.
- Increase max curls to number of mining threads + queue * 2, accounting for up
and downstream comms.
- Queue enough requests to get started.
- There is no point trying to clone_work in get_work() any more since we clone
on every get_work_thread where possible.
- There is no point subtracting 1 from maxq in get_work_thread.
- Only set lagging flag once there are no staged work items.
- select_pool does not switch back to the primary once lagging is disabled.
- miner.php allow page title to be defined in myminer.php
- Free work before retrying in get_work_thread.
- Increment total work counter under mutex lock.
- Increment the queued count after the curl is popped in case there's a delay
waiting on curls and we think we've queued work when in fact we're waiting
- API new command 'coin' with mining information
- Do the dynamic timing in opencl code over a single pass through scanhash to
make sure we're only getting opencl times contributing to the measured inte
- Increase curl reaping time to 5 minutes since comms between  curl requests can
be 2 mins apart with lots of rolltime.
- No need for extra variable in hash_push.
- Remove short options -r and -R to allow them to be reused and remove readme
entries for deprecated options.
- Avoid attempting to recursively lock the console mutex by disabling warnings
in gpu_fanpercent when fanspeed monitoring fails on windows. Debugged by luke-jr
- Deprecate the opt_fail_pause parameter, leaving a null placeholder for
existing configurations.
- Don't pause after failed getwork, set lagging flag and reassess.
- Add message to share if it's a resubmit.
- We should not be pausing in trying to resubmit shares.
- Get rid of the extending fail pause on failed connects since we discard work
after a period.
- get_work always returns true so turn it into a void function.
- get_work never returns false so get rid of fail pause loop.
- Get rid of pause and retry from get_upstream_work so we only do it from one
place.
- Deprecate the opt_retries feature as no one wants cgminer to automatically
abort. Leave a null placeholder for configurations that still have it.
- Reinstate fix ADL gpu-map not working when there are more ADL devices than
openCL patch by Nite69. Add virtual adl mapping for when none is specified o
- miner.php show summary Diff1 Shares total
- miner.php fix Work Utility totals
- miner.php format new Work Utility and Diff1 Shares
- API V1.17 show Work Utility and Diff1 Shares

..and now I may lie down for a week to recover...

[1715126300]

1715126300

[1715126300]

1715126300

[1715126300]

1715126300

[1715126300]

1715126300

[crazyates]

..and now I may lie down for a week to recover...

Even God rested on the seventh day. 

[bitcoindaddy]

I noticed on btcguild I only get around 80 percent efficiency but on ozcoin.net I'm getting 270 percent. Can I assume btcguild doesn't support rolltime and that I'm getting more coin with ozcoin.net? (PPS vs. PPS)

[bitcoindaddy]

Also, 2.7.1 segfaulted on me. I couldn't find a core file, is there normally one produced?

[crazyates]

I noticed on btcguild I only get around 80 percent efficiency but on ozcoin.net I'm getting 270 percent. Can I assume btcguild doesn't support rolltime and that I'm getting more coin with ozcoin.net? (PPS vs. PPS)

The nonce size in bitcoin is too small - it is 32bits (sorta like MS-DOS saying 640K will always be big enough)
Each piece of work you get from a pool allows you to attempt 2^32 (~4 billion) hashes - i.e. one for each nonce value.

For a single MiniRig, this means that you need to get over 350 pieces of work from the pool per minute.

The hack solution to this, that someone come up with in the past, was to instead edit the time field (roll it forward)
Thus if you add 1 second to the time field, you can then do another 4 billion hashes on the same piece of work you got from the pool (without having to ask for and wait for more work from the pool) since you are now hashing something different.

Thus most pools now support this.
If they don't - get a new pool.

It means you get less work from the pool to do the same amount of hashing.
The E: % value will tell you how much less.
e.g. if you have an E: of 200% that means you are rolling each piece of work you get, on average, once
Thus you are getting half as much work from the pool to do the same amount of hashing.

If a pool doesn't support roll-n-time then perfect "no rolling" E: is 100%

It doesn't reduce the number of shares you have to send back to the pool however.
The pool would have to support high difficulty shares to reduce that also.
But it does reduce the amount of work you request from the pool.

You can add up to 7200 seconds to the time field and bitcoind will still accept it as a valid block solution if you find a block
... thus why the timestamps in the bitcoin blockchain are not to be taken too seriously.

A larger nonce range in bitcoin would solve this correctly, rather than a time hack, however that would require a hard fork, and the bitcoin devs are scared of hard forks coz they do soft forks so badly.

https://bitcointalk.org/index.php?topic=89278.0

Edit:
The other solution (that no doubt, one of which will be used) that are being bandied about are to move the pools work to the miner.
Thus pretty much all the pool will be doing is counting shares.
The miner (or some other program run by the miner) will be doing the work of dealing with longpolls, transaction selection and everything related to that.

If this were to take place, then I would suggest anyone who implements such a solution, also put a fee in the coinbase to cover the work you are doing for the pool .........

From Kano's description, an E: of < 100% sounds like a pool that does not support rollntime. As long as the pool can provide you with enough work, and do it fast enough, you should be getting the same results. If the pool can't provide work fast enough, then you could have down time.

[gyverlb]

New version - 2.7.1, 21st August 2012

Something odd :
I have 6 mining rigs. 4 have multiple cards, 2 have a single card with a total of 21 cards. Until 2.7.1, I've had next to zero hardware errors reported by cgminer, only one card in the 21 had occasional (maybe 1/day) errors, it was a 5870 in a 5-card rig.
With 2.7.1, I've got hardware errors reported on all cards in the rigs with multiple cards (in fact 18 out of 19 cards have errors). The cards in a single-card rig don't have errors reported (even a 5970 with its 2 GPUs, so if it may be linked to the number of cards, it's not with the number of GPUs).

Could something have prevented errors to be reported with previous versions ?

Some additional, maybe related information :
I had hardware errors on a 7970 when I used 2 threads on it, switching everything to dynamic with a 250ms target made these errors disappear (with no noticeable difference in hashrate).

All the cards are overclocked/downvolted/... but are set in a configuration which is(/was ?) known to be stable (ie: each time a card is SICK/DEAD or crashes cgminer it is downclocked), it's only when a new card is added that a rig can be temporarily unstable the time I find its sweet spot.

Note that I detect card making cgminer crash by monitoring their temperature: on my configuration a card was most of the time not detected SICK (I suspect it as something to do with my using the API to monitor the rigs: it seemed to crash cgminer before the 60s SICK delay) but crashed cgminer, before each crash I could witness the temperature of the offending card dropped markedly.

[kano]

Also, 2.7.1 segfaulted on me. I couldn't find a core file, is there normally one produced?

On linux to produce a core file you usually have to type something like:
ulimit -c 2097152
before you start cgminer, to get a core file on a crash (that number I think means 2GB limit on the core size)
I always set a limit since once I got a strange crash that was about 60GB and I was running it on a network drive (at home) and my home network was rather slow for a few minutes, while it created the core file

Edit: then to get a possibly useful trace of the core file core.nnn to pastebin:
gdb cgminer core.nnn
bt

[ubbe_mainer]

Hi,

Big problems with my miner. I have been mining 2-3 months, but I had to make fresh install of my backup computer. All steps and logs are here:

https://bitcointalk.org/index.php?topic=101758.0

Please, someone? RCP calls and job curls are missing... I have spotted the part in cgminer.c code, but so far I have not added own debugging messages while I hope someone knows what is wrong.

[The00Dustin]

So I tried to use scrypt on cgminer 2.7.1 just to see what kind of hashrate I would get with my 5830.  This is the command I ran:

GPU_MAX_ALLOC_PERCENT=100 GPU_USE_SYNC_OBJECTS=1 DISPLAY=:0 cgminer --shaders 1120 --intensity 12 --scrypt  (tried with --benchmark and with a pool in case --benchmark wouldn't work).  Instead of getting an idea of my hashrate, I got a failure.  Perhaps my SDK is too old, but I'm not sure.  Anyone have any clue what this is about?:

Code:

 [2012-08-21 12:22:05] Started cgminer 2.7.1

/tmp/OCLKybKEH.cl(583): error: the opencl builtin function can only takes
          [unsigned] char/short/int/long[n] as first argument
                B[i+4] = EndianSwap(tmp[i]);
                         ^

/tmp/OCLKybKEH.cl(583): error: mixed vector-scalar operation not allowed
          unless up-convertable(scalar-type=>vector-element-type)
                B[i+4] = EndianSwap(tmp[i]);
                         ^

/tmp/OCLKybKEH.cl(590): warning: unrecognized #pragma
  #pragma unroll
          ^

/tmp/OCLKybKEH.cl(594): warning: unrecognized #pragma
  #pragma unroll
          ^

/tmp/OCLKybKEH.cl(607): warning: unrecognized #pragma
  #pragma unroll
          ^

/tmp/OCLKybKEH.cl(611): warning: unrecognized #pragma
  #pragma unroll
          ^

/tmp/OCLKybKEH.cl(624): warning: unrecognized #pragma
  #pragma unroll
          ^

/tmp/OCLKybKEH.cl(642): warning: unrecognized #pragma
  #pragma unroll
          ^

/tmp/OCLKybKEH.cl(663): warning: unrecognized #pragma
  #pragma unroll
          ^

/tmp/OCLKybKEH.cl(677): warning: unrecognized #pragma
  #pragma unroll
          ^

/tmp/OCLKybKEH.cl(636): warning: variable "ySIZE" was declared but never
          referenced
        const uint ySIZE = (1024/LOOKUP_GAP+(1024%LOOKUP_GAP>0));
                   ^

/tmp/OCLKybKEH.cl(707): warning: unrecognized #pragma
  #pragma unroll
          ^

12 errors detected in the compilation of "/tmp/OCLKybKEH.cl".

 [2012-08-21 12:22:05] Failed to init GPU thread 0, disabling device 0
 [2012-08-21 12:22:05] Restarting the GPU from the menu will not fix this.
 [2012-08-21 12:22:05] Try restarting cgminer.

[Krak]

Hopefully the dynamic mode for GPUs is fixed on windows.

What about on Linux? I've been noticing the -10 bug pop up since I upgraded to 2.7 last night. I've had to start just manually switching my intensity from 4 to 6.

[d3m0n1q_733rz]

Other vector sizes have been tried. They suck.

They've been tried with scrypt you say?  Because, with scrypt, a lot of the overhead is shifted to the loading and unloading of data.  And I recall having 8 vectors working for me better than 4 with the VLIW5 based GPU as long as the code could be written to support it.
Instead of just one uint8, we have two uint4 variables that have to load back and forth.  Now, it might just be me, but that seems rather inefficient doing that many reads and writes.
If someone would give it a try, I think we could determine it once and for all.

[Morblias]

When I went to 2.7.1, at some point one of my cgminers switched pools, and then would not switch back:



Bonuspool is definitely up and the cgminer on the right is connected to it, but on the left, it keeps saying it is online, won't switch to it, stays on pool 1, then says pool 0 is not responding (even though it isn't offline).

[farfie]

When I went to 2.7.1, at some point one of my cgminers switched pools, and then would not switch back:

https://i.imgur.com/6PKRn.jpg

Bonuspool is definitely up and the cgminer on the right is connected to it, but on the left, it keeps saying it is online, won't switch to it, stays on pool 1, then says pool 0 is not responding (even though it isn't offline).

I can confirm this. At first I thought it was because I manually added a pool while cgminer was running and that messed it up, but now he's posting a similar issue. Did you add a pool like me or did it switch and fail to switch back on its own Morblias?

[Morblias]

I can confirm this. At first I thought it was because I manually added a pool while cgminer was running and that messed it up, but now he's posting a similar issue. Did you add a pool like me or did it switch and fail to switch back on its own Morblias?

It switched and failed to switch back.

[-ck]

Other vector sizes have been tried. They suck.

They've been tried with scrypt you say?  Because, with scrypt, a lot of the overhead is shifted to the loading and unloading of data.  And I recall having 8 vectors working for me better than 4 with the VLIW5 based GPU as long as the code could be written to support it.
Instead of just one uint8, we have two uint4 variables that have to load back and forth.  Now, it might just be me, but that seems rather inefficient doing that many reads and writes.
If someone would give it a try, I think we could determine it once and for all.

Blow yourself away, because I am most certainly not going to give one more second's thought to scrypt.

[-ck]

I can confirm this. At first I thought it was because I manually added a pool while cgminer was running and that messed it up, but now he's posting a similar issue. Did you add a pool like me or did it switch and fail to switch back on its own Morblias?

It switched and failed to switch back.

Was bonuspool the common element?

[-ck]

Hopefully the dynamic mode for GPUs is fixed on windows.

What about on Linux? I've been noticing the -10 bug pop up since I upgraded to 2.7 last night. I've had to start just manually switching my intensity from 4 to 6.

Hmm then that's almost certainly related to one long getwork skewing the results. In which case it is not fixed in either linux or windows on 2.7.1. Darn, well at least I know what to fix next.

[-ck]

So I tried to use scrypt on cgminer 2.7.1 just to see what kind of hashrate I would get with my 5830.  This is the command I ran:

Needs sdk 2.6+

[-ck]

New version - 2.7.1, 21st August 2012

Something odd :
I have 6 mining rigs. 4 have multiple cards, 2 have a single card with a total of 21 cards. Until 2.7.1, I've had next to zero hardware errors reported by cgminer, only one card in the 21 had occasional (maybe 1/day) errors, it was a 5870 in a 5-card rig.
With 2.7.1, I've got hardware errors reported on all cards in the rigs with multiple cards (in fact 18 out of 19 cards have errors). The cards in a single-card rig don't have errors reported (even a 5970 with its 2 GPUs, so if it may be linked to the number of cards, it's not with the number of GPUs).

Could something have prevented errors to be reported with previous versions ?

Some additional, maybe related information :
I had hardware errors on a 7970 when I used 2 threads on it, switching everything to dynamic with a 250ms target made these errors disappear (with no noticeable difference in hashrate).

All the cards are overclocked/downvolted/... but are set in a configuration which is(/was ?) known to be stable (ie: each time a card is SICK/DEAD or crashes cgminer it is downclocked), it's only when a new card is added that a rig can be temporarily unstable the time I find its sweet spot.

Note that I detect card making cgminer crash by monitoring their temperature: on my configuration a card was most of the time not detected SICK (I suspect it as something to do with my using the API to monitor the rigs: it seemed to crash cgminer before the 60s SICK delay) but crashed cgminer, before each crash I could witness the temperature of the offending card dropped markedly.

Nothing I can think of that "prevented errors" directly apart from the fact that 2.7.0+ will work the GPUs harder than ever, with very little chance of rest between work unless you have a significant outage.

[SAC]

just to see what kind of hashrate I would get with my 5830.

~256kh/s running at 870,1170, worksize 256, thread-concurrency 6720, vectors 4, gpu-threads 2, intensity 18 or should be if it is anything like the ones I have. Oh and it takes ~170w per card for LTC as opposed to 140w for BTC running at 1.1 volts core.